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
12 * afs_MarkServerUpOrDown
20 * afsi_SetServerIPRank
32 #include <afsconfig.h>
33 #include "afs/param.h"
37 #include "afs/sysincludes.h" /* Standard vendor system headers */
40 #if !defined(AFS_LINUX20_ENV)
43 #include <netinet/in.h>
46 #include "h/hashing.h"
48 #if !defined(AFS_HPUX110_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_DARWIN_ENV)
49 #include <netinet/in_var.h>
50 #endif /* AFS_HPUX110_ENV */
52 #include <net/if_var.h>
54 #endif /* !defined(UKERNEL) */
56 #include "afsincludes.h" /* Afs-based standard headers */
57 #include "afs/afs_stats.h" /* afs statistics */
58 #include "rx/rx_multi.h"
60 #if defined(AFS_SUN5_ENV)
62 #include <inet/common.h>
63 #include <netinet/ip6.h>
64 #define ipif_local_addr ipif_lcl_addr
66 # define V4_PART_OF_V6(v6) v6.s6_addr32[3]
71 /* Exported variables */
72 afs_rwlock_t afs_xserver; /* allocation lock for servers */
73 struct server *afs_servers[NSERVERS]; /* Hashed by server`s uuid & 1st ip */
74 afs_rwlock_t afs_xsrvAddr; /* allocation lock for srvAddrs */
75 struct srvAddr *afs_srvAddrs[NSERVERS]; /* Hashed by server's ip */
78 /* debugging aids - number of alloc'd server and srvAddr structs. */
79 int afs_reuseServers = 0;
80 int afs_reuseSrvAddrs = 0;
81 int afs_totalServers = 0;
82 int afs_totalSrvAddrs = 0;
86 static struct afs_stats_SrvUpDownInfo *
87 GetUpDownStats(struct server *srv)
89 struct afs_stats_SrvUpDownInfo *upDownP;
90 u_short fsport = AFS_FSPORT;
93 fsport = srv->cell->fsport;
95 if (srv->addr->sa_portal == fsport)
96 upDownP = afs_stats_cmperf.fs_UpDown;
98 upDownP = afs_stats_cmperf.vl_UpDown;
100 if (srv->cell && afs_IsPrimaryCell(srv->cell))
101 return &upDownP[AFS_STATS_UPDOWN_IDX_SAME_CELL];
103 return &upDownP[AFS_STATS_UPDOWN_IDX_DIFF_CELL];
107 /*------------------------------------------------------------------------
108 * afs_MarkServerUpOrDown
111 * Mark the given server up or down, and track its uptime stats.
114 * a_serverP : Ptr to server record to fiddle with.
115 * a_isDown : Is the server is to be marked down?
121 * The CM server structures must be write-locked.
125 *------------------------------------------------------------------------*/
128 afs_MarkServerUpOrDown(struct srvAddr *sa, int a_isDown)
130 struct server *a_serverP = sa->server;
132 osi_timeval_t currTime, *currTimeP; /*Current time */
133 afs_int32 downTime; /*Computed downtime, in seconds */
134 struct afs_stats_SrvUpDownInfo *upDownP; /*Ptr to up/down info record */
137 * If the server record is marked the same as the new status we've
138 * been fed, then there isn't much to be done.
140 if ((a_isDown && (sa->sa_flags & SRVADDR_ISDOWN))
141 || (!a_isDown && !(sa->sa_flags & SRVADDR_ISDOWN)))
145 sa->sa_flags |= SRVADDR_ISDOWN;
146 for (sap = a_serverP->addr; sap; sap = sap->next_sa) {
147 if (!(sap->sa_flags & SRVADDR_ISDOWN)) {
148 /* Not all ips are up so don't bother with the
149 * server's up/down stats */
154 * All ips are down we treat the whole server down
156 a_serverP->flags |= SRVR_ISDOWN;
158 sa->sa_flags &= ~SRVADDR_ISDOWN;
159 /* If any ips are up, the server is also marked up */
160 a_serverP->flags &= ~SRVR_ISDOWN;
161 for (sap = a_serverP->addr; sap; sap = sap->next_sa) {
162 if (sap->sa_flags & SRVADDR_ISDOWN) {
163 /* Not all ips are up so don't bother with the
164 * server's up/down stats */
171 * Compute the current time and which overall stats record is to be
172 * updated; we'll need them one way or another.
174 currTimeP = &currTime;
175 osi_GetuTime(currTimeP);
177 upDownP = GetUpDownStats(a_serverP);
181 * Server going up -> down; remember the beginning of this
184 a_serverP->lastDowntimeStart = currTime.tv_sec;
186 (upDownP->numDownRecords)++;
187 (upDownP->numUpRecords)--;
188 } /*Server being marked down */
191 * Server going down -> up; remember everything about this
192 * newly-completed downtime incident.
194 downTime = currTime.tv_sec - a_serverP->lastDowntimeStart;
195 (a_serverP->numDowntimeIncidents)++;
196 a_serverP->sumOfDowntimes += downTime;
198 (upDownP->numUpRecords)++;
199 (upDownP->numDownRecords)--;
200 (upDownP->numDowntimeIncidents)++;
201 if (a_serverP->numDowntimeIncidents == 1)
202 (upDownP->numRecordsNeverDown)--;
203 upDownP->sumOfDowntimes += downTime;
204 if ((upDownP->shortestDowntime == 0)
205 || (downTime < upDownP->shortestDowntime))
206 upDownP->shortestDowntime = downTime;
207 if ((upDownP->longestDowntime == 0)
208 || (downTime > upDownP->longestDowntime))
209 upDownP->longestDowntime = downTime;
212 if (downTime <= AFS_STATS_MAX_DOWNTIME_DURATION_BUCKET0)
213 (upDownP->downDurations[0])++;
214 else if (downTime <= AFS_STATS_MAX_DOWNTIME_DURATION_BUCKET1)
215 (upDownP->downDurations[1])++;
216 else if (downTime <= AFS_STATS_MAX_DOWNTIME_DURATION_BUCKET2)
217 (upDownP->downDurations[2])++;
218 else if (downTime <= AFS_STATS_MAX_DOWNTIME_DURATION_BUCKET3)
219 (upDownP->downDurations[3])++;
220 else if (downTime <= AFS_STATS_MAX_DOWNTIME_DURATION_BUCKET4)
221 (upDownP->downDurations[4])++;
222 else if (downTime <= AFS_STATS_MAX_DOWNTIME_DURATION_BUCKET5)
223 (upDownP->downDurations[5])++;
225 (upDownP->downDurations[6])++;
227 } /*Server being marked up */
229 } /*MarkServerUpOrDown */
233 afs_ServerDown(struct srvAddr *sa, int code, struct rx_connection *rxconn)
235 struct server *aserver = sa->server;
237 AFS_STATCNT(ServerDown);
238 if (aserver->flags & SRVR_ISDOWN || sa->sa_flags & SRVADDR_ISDOWN)
240 afs_MarkServerUpOrDown(sa, SRVR_ISDOWN);
241 if (sa->sa_portal == aserver->cell->vlport)
242 print_internet_address("afs: Lost contact with volume location server ",
243 sa, "", 1, code, rxconn);
245 print_internet_address("afs: Lost contact with file server ", sa, "",
251 /* return true if we have any callback promises from this server */
253 afs_HaveCallBacksFrom(struct server *aserver)
259 AFS_STATCNT(HaveCallBacksFrom);
260 now = osi_Time(); /* for checking for expired callbacks */
261 for (i = 0; i < VCSIZE; i++) { /* for all guys in the hash table */
262 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
264 * Check to see if this entry has an unexpired callback promise
265 * from the required host
267 if (aserver == tvc->callback && tvc->cbExpires >= now
268 && ((tvc->f.states & CRO) == 0))
274 } /*HaveCallBacksFrom */
278 CheckVLServer(struct srvAddr *sa, struct vrequest *areq)
280 struct server *aserver = sa->server;
283 struct rx_connection *rxconn;
285 AFS_STATCNT(CheckVLServer);
286 /* Ping dead servers to see if they're back */
287 if (!((aserver->flags & SRVR_ISDOWN) || (sa->sa_flags & SRVADDR_ISDOWN))
288 || (aserver->flags & SRVR_ISGONE))
291 return; /* can't do much */
293 tc = afs_ConnByHost(aserver, aserver->cell->vlport,
294 aserver->cell->cellNum, areq, 1, SHARED_LOCK, 0,
298 rx_SetConnDeadTime(rxconn, 3);
301 code = VL_ProbeServer(rxconn);
303 rx_SetConnDeadTime(rxconn, afs_rx_deadtime);
305 * If probe worked, or probe call not yet defined (for compatibility
306 * with old vlsevers), then we treat this server as running again
308 if (code == 0 || (code <= -450 && code >= -470)) {
309 if (tc->parent->srvr == sa) {
310 afs_MarkServerUpOrDown(sa, 0);
311 print_internet_address("afs: volume location server ", sa,
312 " is back up", 2, code, rxconn);
315 afs_PutConn(tc, rxconn, SHARED_LOCK);
320 #ifndef AFS_MINCHANGE /* So that some can increase it in param.h */
321 #define AFS_MINCHANGE 2 /* min change we'll bother with */
323 #ifndef AFS_MAXCHANGEBACK
324 #define AFS_MAXCHANGEBACK 10 /* max seconds we'll set a clock back at once */
328 /*------------------------------------------------------------------------
329 * EXPORTED afs_CountServers
332 * Originally meant to count the number of servers and determining
333 * up/down info, this routine will now simply sum up all of the
334 * server record ages. All other up/down information is kept on the
344 * This routine locks afs_xserver for write for the duration.
347 * Set CM perf stats field sumOfRecordAges for all server record
349 *------------------------------------------------------------------------*/
352 afs_CountServers(void)
354 int currIdx; /*Curr idx into srv table */
355 struct server *currSrvP; /*Ptr to curr server record */
356 afs_int32 currChainLen; /*Length of curr hash chain */
357 osi_timeval_t currTime; /*Current time */
358 osi_timeval_t *currTimeP; /*Ptr to above */
359 afs_int32 srvRecordAge; /*Age of server record, in secs */
360 struct afs_stats_SrvUpDownInfo *upDownP; /*Ptr to current up/down
361 * info being manipulated */
364 * Write-lock the server table so we don't get any interference.
366 ObtainReadLock(&afs_xserver);
369 * Iterate over each hash index in the server table, walking down each
370 * chain and tallying what we haven't computed from the records there on
371 * the fly. First, though, initialize the tallies that will change.
373 afs_stats_cmperf.srvMaxChainLength = 0;
375 afs_stats_cmperf.fs_UpDown[0].sumOfRecordAges = 0;
376 afs_stats_cmperf.fs_UpDown[0].ageOfYoungestRecord = 0;
377 afs_stats_cmperf.fs_UpDown[0].ageOfOldestRecord = 0;
378 memset(afs_stats_cmperf.fs_UpDown[0].downIncidents, 0,
379 AFS_STATS_NUM_DOWNTIME_INCIDENTS_BUCKETS * sizeof(afs_int32));
381 afs_stats_cmperf.fs_UpDown[1].sumOfRecordAges = 0;
382 afs_stats_cmperf.fs_UpDown[1].ageOfYoungestRecord = 0;
383 afs_stats_cmperf.fs_UpDown[1].ageOfOldestRecord = 0;
384 memset(afs_stats_cmperf.fs_UpDown[1].downIncidents, 0,
385 AFS_STATS_NUM_DOWNTIME_INCIDENTS_BUCKETS * sizeof(afs_int32));
387 afs_stats_cmperf.vl_UpDown[0].sumOfRecordAges = 0;
388 afs_stats_cmperf.vl_UpDown[0].ageOfYoungestRecord = 0;
389 afs_stats_cmperf.vl_UpDown[0].ageOfOldestRecord = 0;
390 memset(afs_stats_cmperf.vl_UpDown[0].downIncidents, 0,
391 AFS_STATS_NUM_DOWNTIME_INCIDENTS_BUCKETS * sizeof(afs_int32));
393 afs_stats_cmperf.vl_UpDown[1].sumOfRecordAges = 0;
394 afs_stats_cmperf.vl_UpDown[1].ageOfYoungestRecord = 0;
395 afs_stats_cmperf.vl_UpDown[1].ageOfOldestRecord = 0;
396 memset(afs_stats_cmperf.vl_UpDown[1].downIncidents, 0,
397 AFS_STATS_NUM_DOWNTIME_INCIDENTS_BUCKETS * sizeof(afs_int32));
400 * Compute the current time, used to figure out server record ages.
402 currTimeP = &currTime;
403 osi_GetuTime(currTimeP);
406 * Sweep the server hash table, tallying all we need to know.
408 for (currIdx = 0; currIdx < NSERVERS; currIdx++) {
410 for (currSrvP = afs_servers[currIdx]; currSrvP;
411 currSrvP = currSrvP->next) {
413 * Bump the current chain length.
418 * Any further tallying for this record will only be done if it has
421 if ((currSrvP->flags & AFS_SERVER_FLAG_ACTIVATED)
422 && currSrvP->addr && currSrvP->cell) {
425 * Compute the current server record's age, then remember it
426 * in the appropriate places.
428 srvRecordAge = currTime.tv_sec - currSrvP->activationTime;
429 upDownP = GetUpDownStats(currSrvP);
430 upDownP->sumOfRecordAges += srvRecordAge;
431 if ((upDownP->ageOfYoungestRecord == 0)
432 || (srvRecordAge < upDownP->ageOfYoungestRecord))
433 upDownP->ageOfYoungestRecord = srvRecordAge;
434 if ((upDownP->ageOfOldestRecord == 0)
435 || (srvRecordAge > upDownP->ageOfOldestRecord))
436 upDownP->ageOfOldestRecord = srvRecordAge;
438 if (currSrvP->numDowntimeIncidents <=
439 AFS_STATS_MAX_DOWNTIME_INCIDENTS_BUCKET0)
440 (upDownP->downIncidents[0])++;
441 else if (currSrvP->numDowntimeIncidents <=
442 AFS_STATS_MAX_DOWNTIME_INCIDENTS_BUCKET1)
443 (upDownP->downIncidents[1])++;
444 else if (currSrvP->numDowntimeIncidents <=
445 AFS_STATS_MAX_DOWNTIME_INCIDENTS_BUCKET2)
446 (upDownP->downIncidents[2])++;
447 else if (currSrvP->numDowntimeIncidents <=
448 AFS_STATS_MAX_DOWNTIME_INCIDENTS_BUCKET3)
449 (upDownP->downIncidents[3])++;
450 else if (currSrvP->numDowntimeIncidents <=
451 AFS_STATS_MAX_DOWNTIME_INCIDENTS_BUCKET4)
452 (upDownP->downIncidents[4])++;
454 (upDownP->downIncidents[5])++;
457 } /*Current server has been active */
458 } /*Walk this chain */
461 * Before advancing to the next chain, remember facts about this one.
463 if (currChainLen > afs_stats_cmperf.srvMaxChainLength) {
465 * We beat out the former champion (which was initially set to 0
466 * here). Mark down the new winner, and also remember if it's an
469 afs_stats_cmperf.srvMaxChainLength = currChainLen;
470 if (currChainLen > afs_stats_cmperf.srvMaxChainLengthHWM)
471 afs_stats_cmperf.srvMaxChainLengthHWM = currChainLen;
472 } /*Update chain length maximum */
473 } /*For each hash chain */
476 * We're done. Unlock the server table before returning to our caller.
478 ReleaseReadLock(&afs_xserver);
480 } /*afs_CountServers */
484 ForceAllNewConnections(void)
487 struct srvAddr **addrs;
491 ObtainReadLock(&afs_xserver); /* Necessary? */
492 ObtainReadLock(&afs_xsrvAddr);
495 for (i = 0; i < NSERVERS; i++) {
496 for (sa = afs_srvAddrs[i]; sa; sa = sa->next_bkt) {
501 addrs = afs_osi_Alloc(srvAddrCount * sizeof(*addrs));
502 osi_Assert(addrs != NULL);
504 for (i = 0; i < NSERVERS; i++) {
505 for (sa = afs_srvAddrs[i]; sa; sa = sa->next_bkt) {
506 if (j >= srvAddrCount)
512 ReleaseReadLock(&afs_xsrvAddr);
513 ReleaseReadLock(&afs_xserver);
514 for (i = 0; i < j; i++) {
516 ForceNewConnections(sa);
519 afs_osi_Free(addrs, srvAddrCount * sizeof(*addrs));
523 CkSrv_MarkUpDown(struct afs_conn **conns, struct rx_connection **rxconns,
524 int nconns, afs_int32 *results)
530 for(i = 0; i < nconns; i++){
532 sa = tc->parent->srvr;
534 if (( results[i] >= 0 ) && (sa->sa_flags & SRVADDR_ISDOWN) &&
535 (tc->parent->srvr == sa)) {
537 print_internet_address("afs: file server ", sa, " is back up", 2,
538 results[i], rxconns[i]);
540 ObtainWriteLock(&afs_xserver, 244);
541 ObtainWriteLock(&afs_xsrvAddr, 245);
542 afs_MarkServerUpOrDown(sa, 0);
543 ReleaseWriteLock(&afs_xsrvAddr);
544 ReleaseWriteLock(&afs_xserver);
546 if (afs_waitForeverCount) {
547 afs_osi_Wakeup(&afs_waitForever);
550 if (results[i] < 0) {
552 afs_ServerDown(sa, results[i], rxconns[i]);
553 ForceNewConnections(sa); /* multi homed clients */
560 CkSrv_GetCaps(int nconns, struct rx_connection **rxconns,
561 struct afs_conn **conns)
568 caps = afs_osi_Alloc(nconns * sizeof (Capabilities));
569 osi_Assert(caps != NULL);
570 memset(caps, 0, nconns * sizeof(Capabilities));
572 results = afs_osi_Alloc(nconns * sizeof (afs_int32));
573 osi_Assert(results != NULL);
576 multi_Rx(rxconns,nconns)
578 multi_RXAFS_GetCapabilities(&caps[multi_i]);
579 results[multi_i] = multi_error;
583 for ( i = 0 ; i < nconns ; i++ ) {
584 ts = conns[i]->parent->srvr->server;
587 ts->capabilities = 0;
588 ts->flags |= SCAPS_KNOWN;
589 if ( results[i] == RXGEN_OPCODE ) {
590 /* Mark server as up - it responded */
594 if ( results[i] >= 0 )
595 /* we currently handle 32-bits of capabilities */
596 if (caps[i].Capabilities_len > 0) {
597 ts->capabilities = caps[i].Capabilities_val[0];
598 xdr_free((xdrproc_t)xdr_Capabilities, &caps[i]);
599 caps[i].Capabilities_val = NULL;
600 caps[i].Capabilities_len = 0;
603 CkSrv_MarkUpDown(conns, rxconns, nconns, results);
605 afs_osi_Free(caps, nconns * sizeof(Capabilities));
606 afs_osi_Free(results, nconns * sizeof(afs_int32));
609 /* check down servers (if adown), or running servers (if !adown) */
611 afs_CheckServers(int adown, struct cell *acellp)
613 afs_LoopServers(adown?AFS_LS_DOWN:AFS_LS_UP, acellp, 1, CkSrv_GetCaps, NULL);
616 /* adown: AFS_LS_UP - check only up
617 * AFS_LS_DOWN - check only down.
618 * AFS_LS_ALL - check all */
620 afs_LoopServers(int adown, struct cell *acellp, int vlalso,
621 void (*func1) (int nservers, struct rx_connection **rxconns,
622 struct afs_conn **conns),
623 void (*func2) (int nservers, struct rx_connection **rxconns,
624 struct afs_conn **conns))
626 struct vrequest *treq = NULL;
629 struct afs_conn *tc = NULL;
634 struct srvAddr **addrs;
635 struct afs_conn **conns;
637 struct rx_connection **rxconns;
638 afs_int32 *conntimer;
640 AFS_STATCNT(afs_CheckServers);
643 * No sense in doing the server checks if we are running in disconnected
646 if (AFS_IS_DISCONNECTED)
649 if ((code = afs_CreateReq(&treq, afs_osi_credp)))
651 ObtainReadLock(&afs_xserver); /* Necessary? */
652 ObtainReadLock(&afs_xsrvAddr);
655 for (i = 0; i < NSERVERS; i++) {
656 for (sa = afs_srvAddrs[i]; sa; sa = sa->next_bkt) {
661 addrs = afs_osi_Alloc(srvAddrCount * sizeof(*addrs));
662 osi_Assert(addrs != NULL);
664 for (i = 0; i < NSERVERS; i++) {
665 for (sa = afs_srvAddrs[i]; sa; sa = sa->next_bkt) {
666 if (j >= srvAddrCount)
672 ReleaseReadLock(&afs_xsrvAddr);
673 ReleaseReadLock(&afs_xserver);
675 conns = afs_osi_Alloc(j * sizeof(struct afs_conn *));
676 osi_Assert(conns != NULL);
677 rxconns = afs_osi_Alloc(j * sizeof(struct rx_connection *));
678 osi_Assert(rxconns != NULL);
679 conntimer = afs_osi_Alloc(j * sizeof (afs_int32));
680 osi_Assert(conntimer != NULL);
683 for (i = 0; i < j; i++) {
684 struct rx_connection *rxconn;
690 /* See if a cell to check was specified. If it is spec'd and not
691 * this server's cell, just skip the server.
693 if (acellp && acellp != ts->cell)
696 if (((adown==AFS_LS_DOWN) && !(sa->sa_flags & SRVADDR_ISDOWN))
697 || ((adown==AFS_LS_UP) && (sa->sa_flags & SRVADDR_ISDOWN)))
700 /* check vlserver with special code */
701 if (sa->sa_portal == AFS_VLPORT) {
703 CheckVLServer(sa, treq);
707 if (!ts->cell) /* not really an active server, anyway, it must */
708 continue; /* have just been added by setsprefs */
710 /* get a connection, even if host is down; bumps conn ref count */
711 tu = afs_GetUser(treq->uid, ts->cell->cellNum, SHARED_LOCK);
712 tc = afs_ConnBySA(sa, ts->cell->fsport, ts->cell->cellNum, tu,
713 1 /*force */ , 1 /*create */ , SHARED_LOCK, 0,
715 afs_PutUser(tu, SHARED_LOCK);
719 if ((sa->sa_flags & SRVADDR_ISDOWN) || afs_HaveCallBacksFrom(sa->server)) {
721 rxconns[nconns]=rxconn;
722 if (sa->sa_flags & SRVADDR_ISDOWN) {
723 rx_SetConnDeadTime(rxconn, 3);
729 } else /* not holding, kill ref */
730 afs_PutConn(tc, rxconn, SHARED_LOCK);
731 } /* Outer loop over addrs */
733 afs_osi_Free(addrs, srvAddrCount * sizeof(*addrs));
736 (*func1)(nconns, rxconns, conns);
739 (*func2)(nconns, rxconns, conns);
742 for (i = 0; i < nconns; i++) {
743 if (conntimer[i] == 1)
744 rx_SetConnDeadTime(rxconns[i], afs_rx_deadtime);
745 afs_PutConn(conns[i], rxconns[i], SHARED_LOCK); /* done with it now */
748 afs_osi_Free(conns, j * sizeof(struct afs_conn *));
749 afs_osi_Free(rxconns, j * sizeof(struct rx_connection *));
750 afs_osi_Free(conntimer, j * sizeof(afs_int32));
751 afs_DestroyReq(treq);
753 } /*afs_CheckServers*/
756 /* find a server structure given the host address */
758 afs_FindServer(afs_int32 aserver, afs_uint16 aport, afsUUID * uuidp,
765 AFS_STATCNT(afs_FindServer);
767 i = afs_uuid_hash(uuidp) % NSERVERS;
768 for (ts = afs_servers[i]; ts; ts = ts->next) {
769 if ((ts->flags & SRVR_MULTIHOMED)
771 (memcmp((char *)uuidp, (char *)&ts->sr_uuid, sizeof(*uuidp))
772 == 0) && (!ts->addr || (ts->addr->sa_portal == aport)))
777 for (sa = afs_srvAddrs[i]; sa; sa = sa->next_bkt) {
778 if ((sa->sa_ip == aserver) && (sa->sa_portal == aport)) {
785 } /*afs_FindServer */
788 /* some code for creating new server structs and setting preferences follows
789 * in the next few lines...
792 #define MAXDEFRANK 60000
793 #define DEFRANK 40000
795 /* Random number generator and constants from KnuthV2 2d ed, p170 */
801 a is 0.73m should be 0.01m .. 0.99m
802 c is more or less immaterial. 1 or a is suggested.
804 NB: LOW ORDER BITS are not very random. To get small random numbers,
805 treat result as <1, with implied binary point, and multiply by
807 NB: Has to be unsigned, since shifts on signed quantities may preserve
810 /* added rxi_getaddr() to try to get as much initial randomness as
811 possible, since at least one customer reboots ALL their clients
812 simultaneously -- so osi_Time is bound to be the same on some of the
813 clients. This is probably OK, but I don't want to see too much of it.
816 #define ranstage(x) (x)= (afs_uint32) (3141592621U*((afs_uint32)x)+1)
821 static afs_int32 state = 0;
824 AFS_STATCNT(afs_random);
829 * 0xfffffff0 was changed to (~0 << 4) since it works no matter how many
830 * bits are in a tv_usec
832 state = (t.tv_usec & (~0 << 4)) + (rxi_getaddr() & 0xff);
833 state += (t.tv_sec & 0xff);
834 for (i = 0; i < 30; i++) {
844 /* returns int 0..14 using the high bits of a pseudo-random number instead of
845 the low bits, as the low bits are "less random" than the high ones...
846 slight roundoff error exists, an excercise for the reader.
847 need to multiply by something with lots of ones in it, so multiply by
848 8 or 16 is right out.
851 afs_randomMod15(void)
855 temp = afs_random() >> 4;
856 temp = (temp * 15) >> 28;
862 afs_randomMod127(void)
866 temp = afs_random() >> 7;
867 temp = (temp * 127) >> 25;
872 /* afs_SortOneServer()
873 * Sort all of the srvAddrs, of a server struct, by rank from low to high.
876 afs_SortOneServer(struct server *asp)
878 struct srvAddr **rootsa, *lowsa, *tsa, *lowprev;
881 for (rootsa = &(asp->addr); *rootsa; rootsa = &(lowsa->next_sa)) {
883 lowsa = *rootsa; /* lowest sa is the first one */
884 lowrank = lowsa->sa_iprank;
886 for (tsa = *rootsa; tsa->next_sa; tsa = tsa->next_sa) {
887 rank = tsa->next_sa->sa_iprank;
888 if (rank < lowrank) {
890 lowsa = tsa->next_sa;
891 lowrank = lowsa->sa_iprank;
894 if (lowprev) { /* found one lower, so rearrange them */
895 lowprev->next_sa = lowsa->next_sa;
896 lowsa->next_sa = *rootsa;
903 * Sort the pointer to servers by the server's rank (its lowest rank).
904 * It is assumed that the server already has its IP addrs sorted (the
905 * first being its lowest rank: afs_GetServer() calls afs_SortOneServer()).
908 afs_SortServers(struct server *aservers[], int count)
913 AFS_STATCNT(afs_SortServers);
915 for (i = 0; i < count; i++) {
918 for (low = i, j = i + 1; j <= count; j++) {
919 if ((!aservers[j]) || (!aservers[j]->addr))
921 if ((!aservers[low]) || (!aservers[low]->addr))
923 if (aservers[j]->addr->sa_iprank < aservers[low]->addr->sa_iprank) {
929 aservers[i] = aservers[low];
933 } /*afs_SortServers */
935 /* afs_SetServerPrefs is rather system-dependent. It pokes around in kernel
936 data structures to determine what the local IP addresses and subnet masks
937 are in order to choose which server(s) are on the local subnet.
939 As I see it, there are several cases:
940 1. The server address is one of this host's local addresses. In this case
941 this server is to be preferred over all others.
942 2. The server is on the same subnet as one of the this host's local
943 addresses. (ie, an odd-sized subnet, not class A,B,orC)
944 3. The server is on the same net as this host (class A,B or C)
945 4. The server is on a different logical subnet or net than this host, but
946 this host is a 'metric 0 gateway' to it. Ie, two address-spaces share
948 5. This host has a direct (point-to-point, ie, PPP or SLIP) link to the
950 6. This host and the server are disjoint.
952 That is a rough order of preference. If a point-to-point link has a high
953 metric, I'm assuming that it is a very slow link, and putting it at the
954 bottom of the list (at least until RX works better over slow links). If
955 its metric is 1, I'm assuming that it's relatively fast (T1) and putting
957 It's not easy to check for case #4, so I'm ignoring it for the time being.
959 BSD "if" code keeps track of some rough network statistics (cf 'netstat -i')
960 That could be used to prefer certain servers fairly easily. Maybe some
963 NOTE: this code is very system-dependent, and very dependent on the TCP/IP
964 protocols (well, addresses that are stored in uint32s, at any rate).
967 #define IA_DST(ia)((struct sockaddr_in *)(&((struct in_ifaddr *)ia)->ia_dstaddr))
968 #define IA_BROAD(ia)((struct sockaddr_in *)(&((struct in_ifaddr *)ia)->ia_broadaddr))
970 /* SA2ULONG takes a sockaddr_in, not a sockaddr (same thing, just cast it!) */
971 #define SA2ULONG(sa) ((sa)->sin_addr.s_addr)
976 #define PPWEIGHT 4096
980 #ifdef AFS_USERSPACE_IP_ADDR
982 #define afs_min(A,B) ((A)<(B)) ? (A) : (B)
985 * The IP addresses and ranks are determined by afsd (in user space) and
986 * passed into the kernel at startup time through the AFSOP_ADVISEADDR
987 * system call. These are stored in the data structure
988 * called 'afs_cb_interface'.
990 * struct srvAddr *sa; remote server
991 * afs_int32 addr; one of my local addr in net order
992 * afs_uint32 subnetmask; subnet mask of local addr in net order
996 afsi_SetServerIPRank(struct srvAddr *sa, afs_int32 addr,
997 afs_uint32 subnetmask)
999 afs_uint32 myAddr, myNet, mySubnet, netMask;
1000 afs_uint32 serverAddr;
1002 myAddr = ntohl(addr); /* one of my IP addr in host order */
1003 serverAddr = ntohl(sa->sa_ip); /* server's IP addr in host order */
1004 subnetmask = ntohl(subnetmask); /* subnet mask in host order */
1006 if (IN_CLASSA(myAddr))
1007 netMask = IN_CLASSA_NET;
1008 else if (IN_CLASSB(myAddr))
1009 netMask = IN_CLASSB_NET;
1010 else if (IN_CLASSC(myAddr))
1011 netMask = IN_CLASSC_NET;
1015 myNet = myAddr & netMask;
1016 mySubnet = myAddr & subnetmask;
1018 if ((serverAddr & netMask) == myNet) {
1019 if ((serverAddr & subnetmask) == mySubnet) {
1020 if (serverAddr == myAddr) { /* same machine */
1021 sa->sa_iprank = afs_min(sa->sa_iprank, TOPR);
1022 } else { /* same subnet */
1023 sa->sa_iprank = afs_min(sa->sa_iprank, HI);
1025 } else { /* same net */
1026 sa->sa_iprank = afs_min(sa->sa_iprank, MED);
1031 #else /* AFS_USERSPACE_IP_ADDR */
1032 #if (! defined(AFS_SUN5_ENV)) && (! defined(AFS_DARWIN_ENV)) && (! defined(AFS_OBSD47_ENV)) && (! defined(AFS_FBSD_ENV)) && defined(USEIFADDR)
1034 afsi_SetServerIPRank(struct srvAddr *sa, struct in_ifaddr *ifa)
1036 struct sockaddr_in *sin;
1039 if ((ntohl(sa->sa_ip) & ifa->ia_netmask) == ifa->ia_net) {
1040 if ((ntohl(sa->sa_ip) & ifa->ia_subnetmask) == ifa->ia_subnet) {
1042 if (SA2ULONG(sin) == ntohl(sa->sa_ip)) { /* ie, ME!!! */
1043 sa->sa_iprank = TOPR;
1045 t = HI + ifa->ia_ifp->if_metric; /* case #2 */
1046 if (sa->sa_iprank > t)
1050 t = MED + ifa->ia_ifp->if_metric; /* case #3 */
1051 if (sa->sa_iprank > t)
1055 #if defined(IFF_POINTOPOINT) && !defined(UKERNEL)
1056 /* check for case #4 -- point-to-point link */
1057 if ((ifa->ia_ifp->if_flags & IFF_POINTOPOINT)
1058 && (SA2ULONG(IA_DST(ifa)) == ntohl(sa->sa_ip))) {
1059 if (ifa->ia_ifp->if_metric >= (MAXDEFRANK - MED) / PPWEIGHT)
1062 t = MED + (PPWEIGHT << ifa->ia_ifp->if_metric);
1063 if (sa->sa_iprank > t)
1066 #endif /* IFF_POINTOPOINT */
1068 #endif /*(!defined(AFS_SUN5_ENV)) && defined(USEIFADDR) */
1069 #if (defined(AFS_DARWIN_ENV) || defined(AFS_OBSD47_ENV) || defined(AFS_FBSD_ENV)) && defined(USEIFADDR)
1071 #define afs_min(A,B) ((A)<(B)) ? (A) : (B)
1074 afsi_SetServerIPRank(struct srvAddr *sa, rx_ifaddr_t ifa)
1076 struct sockaddr sout;
1077 struct sockaddr_in *sin;
1078 #if defined(AFS_DARWIN80_ENV) && !defined(UKERNEL)
1084 afs_uint32 subnetmask, myAddr, myNet, myDstaddr, mySubnet, netMask;
1085 afs_uint32 serverAddr;
1087 if (rx_ifaddr_address_family(ifa) != AF_INET)
1089 t = rx_ifaddr_address(ifa, &sout, sizeof(sout));
1091 sin = (struct sockaddr_in *)&sout;
1092 myAddr = ntohl(sin->sin_addr.s_addr); /* one of my IP addr in host order */
1096 serverAddr = ntohl(sa->sa_ip); /* server's IP addr in host order */
1097 t = rx_ifaddr_netmask(ifa, &sout, sizeof(sout));
1099 sin = (struct sockaddr_in *)&sout;
1100 subnetmask = ntohl(sin->sin_addr.s_addr); /* subnet mask in host order */
1104 t = rx_ifaddr_dstaddress(ifa, &sout, sizeof(sout));
1106 sin = (struct sockaddr_in *)&sout;
1107 myDstaddr = ntohl(sin->sin_addr.s_addr);
1112 if (IN_CLASSA(myAddr))
1113 netMask = IN_CLASSA_NET;
1114 else if (IN_CLASSB(myAddr))
1115 netMask = IN_CLASSB_NET;
1116 else if (IN_CLASSC(myAddr))
1117 netMask = IN_CLASSC_NET;
1121 myNet = myAddr & netMask;
1122 mySubnet = myAddr & subnetmask;
1124 if ((serverAddr & netMask) == myNet) {
1125 if ((serverAddr & subnetmask) == mySubnet) {
1126 if (serverAddr == myAddr) { /* same machine */
1127 sa->sa_iprank = afs_min(sa->sa_iprank, TOPR);
1128 } else { /* same subnet */
1129 sa->sa_iprank = afs_min(sa->sa_iprank, HI + rx_ifnet_metric(rx_ifaddr_ifnet(ifa)));
1131 } else { /* same net */
1132 sa->sa_iprank = afs_min(sa->sa_iprank, MED + rx_ifnet_metric(rx_ifaddr_ifnet(ifa)));
1135 #ifdef IFF_POINTTOPOINT
1136 /* check for case #4 -- point-to-point link */
1137 if ((rx_ifnet_flags(rx_ifaddr_ifnet(ifa)) & IFF_POINTOPOINT)
1138 && (myDstaddr == serverAddr)) {
1139 if (rx_ifnet_metric(rx_ifaddr_ifnet(ifa)) >= (MAXDEFRANK - MED) / PPWEIGHT)
1142 t = MED + (PPWEIGHT << rx_ifnet_metric(rx_ifaddr_ifnet(ifa)));
1143 if (sa->sa_iprank > t)
1146 #endif /* IFF_POINTTOPOINT */
1148 #endif /*(!defined(AFS_SUN5_ENV)) && defined(USEIFADDR) */
1149 #endif /* else AFS_USERSPACE_IP_ADDR */
1151 #ifdef AFS_SGI62_ENV
1153 afsi_enum_set_rank(struct hashbucket *h, caddr_t mkey, caddr_t arg1,
1156 afsi_SetServerIPRank((struct srvAddr *)arg1, (struct in_ifaddr *)h);
1157 return 0; /* Never match, so we enumerate everyone */
1159 #endif /* AFS_SGI62_ENV */
1161 afs_SetServerPrefs(struct srvAddr *const sa)
1163 #if defined(AFS_USERSPACE_IP_ADDR)
1167 for (i = 0; i < afs_cb_interface.numberOfInterfaces; i++) {
1168 afsi_SetServerIPRank(sa, afs_cb_interface.addr_in[i],
1169 afs_cb_interface.subnetmask[i]);
1171 #else /* AFS_USERSPACE_IP_ADDR */
1172 #if defined(AFS_SUN5_ENV)
1173 #ifdef AFS_SUN510_ENV
1176 extern struct ill_s *ill_g_headp;
1177 long *addr = (long *)ill_g_headp;
1181 int subnet, subnetmask, net, netmask;
1184 #ifdef AFS_SUN510_ENV
1185 rw_enter(&afsifinfo_lock, RW_READER);
1187 for (i = 0; (afsifinfo[i].ipaddr != NULL) && (i < ADDRSPERSITE); i++) {
1189 if (IN_CLASSA(afsifinfo[i].ipaddr)) {
1190 netmask = IN_CLASSA_NET;
1191 } else if (IN_CLASSB(afsifinfo[i].ipaddr)) {
1192 netmask = IN_CLASSB_NET;
1193 } else if (IN_CLASSC(afsifinfo[i].ipaddr)) {
1194 netmask = IN_CLASSC_NET;
1198 net = afsifinfo[i].ipaddr & netmask;
1201 /* XXXXXX Do the individual ip ranking below XXXXX */
1202 if ((sa->sa_ip & netmask) == net) {
1203 if ((sa->sa_ip & subnetmask) == subnet) {
1204 if (afsifinfo[i].ipaddr == sa->sa_ip) { /* ie, ME! */
1205 sa->sa_iprank = TOPR;
1207 sa->sa_iprank = HI + afsifinfo[i].metric; /* case #2 */
1210 sa->sa_iprank = MED + afsifinfo[i].metric; /* case #3 */
1213 sa->sa_iprank = LO + afsifinfo[i].metric; /* case #4 */
1215 /* check for case #5 -- point-to-point link */
1216 if ((afsifinfo[i].flags & IFF_POINTOPOINT)
1217 && (afsifinfo[i].dstaddr == sa->sa_ip)) {
1219 if (afsifinfo[i].metric >= (MAXDEFRANK - MED) / PPWEIGHT)
1220 sa->sa_iprank = MAXDEFRANK;
1222 sa->sa_iprank = MED + (PPWEIGHT << afsifinfo[i].metric);
1227 rw_exit(&afsifinfo_lock);
1229 for (ill = (struct ill_s *)*addr /*ill_g_headp */ ; ill;
1230 ill = ill->ill_next) {
1231 /* Make sure this is an IPv4 ILL */
1234 for (ipif = ill->ill_ipif; ipif; ipif = ipif->ipif_next) {
1235 subnet = ipif->ipif_local_addr & ipif->ipif_net_mask;
1236 subnetmask = ipif->ipif_net_mask;
1238 * Generate the local net using the local address and
1239 * whate we know about Class A, B and C networks.
1241 if (IN_CLASSA(ipif->ipif_local_addr)) {
1242 netmask = IN_CLASSA_NET;
1243 } else if (IN_CLASSB(ipif->ipif_local_addr)) {
1244 netmask = IN_CLASSB_NET;
1245 } else if (IN_CLASSC(ipif->ipif_local_addr)) {
1246 netmask = IN_CLASSC_NET;
1250 net = ipif->ipif_local_addr & netmask;
1252 /* XXXXXX Do the individual ip ranking below XXXXX */
1253 if ((sa->sa_ip & netmask) == net) {
1254 if ((sa->sa_ip & subnetmask) == subnet) {
1255 if (ipif->ipif_local_addr == sa->sa_ip) { /* ie, ME! */
1256 sa->sa_iprank = TOPR;
1258 sa->sa_iprank = HI + ipif->ipif_metric; /* case #2 */
1261 sa->sa_iprank = MED + ipif->ipif_metric; /* case #3 */
1264 sa->sa_iprank = LO + ipif->ipif_metric; /* case #4 */
1266 /* check for case #5 -- point-to-point link */
1267 if ((ipif->ipif_flags & IFF_POINTOPOINT)
1268 && (ipif->ipif_pp_dst_addr == sa->sa_ip)) {
1270 if (ipif->ipif_metric >= (MAXDEFRANK - MED) / PPWEIGHT)
1271 sa->sa_iprank = MAXDEFRANK;
1273 sa->sa_iprank = MED + (PPWEIGHT << ipif->ipif_metric);
1278 #endif /* AFS_SUN510_ENV */
1281 rx_ifnet_t ifn = NULL;
1282 struct in_ifaddr *ifad = (struct in_ifaddr *)0;
1283 struct sockaddr_in *sin;
1286 ifn = rxi_FindIfnet(sa->sa_ip, &ifad);
1287 if (ifn) { /* local, more or less */
1289 if (ifn->if_flags & IFF_LOOPBACK) {
1290 sa->sa_iprank = TOPR;
1293 #endif /* IFF_LOOPBACK */
1294 sin = (struct sockaddr_in *)IA_SIN(ifad);
1295 if (SA2ULONG(sin) == sa->sa_ip) {
1296 sa->sa_iprank = TOPR;
1299 #ifdef IFF_BROADCAST
1300 if (ifn->if_flags & IFF_BROADCAST) {
1301 if (sa->sa_ip == (sa->sa_ip & SA2ULONG(IA_BROAD(ifad)))) {
1306 #endif /* IFF_BROADCAST */
1307 #ifdef IFF_POINTOPOINT
1308 if (ifn->if_flags & IFF_POINTOPOINT) {
1309 if (sa->sa_ip == SA2ULONG(IA_DST(ifad))) {
1310 if (ifn->if_metric > 4) {
1314 sa->sa_iprank = ifn->if_metric;
1317 #endif /* IFF_POINTOPOINT */
1318 sa->sa_iprank += MED + ifn->if_metric; /* couldn't find anything better */
1320 #else /* USEIFADDR */
1323 #ifdef AFS_SGI62_ENV
1324 (void)hash_enum(&hashinfo_inaddr, afsi_enum_set_rank, HTF_INET, NULL,
1325 (caddr_t) sa, NULL);
1326 #elif defined(AFS_DARWIN80_ENV)
1334 if (!ifnet_list_get(AF_INET, &ifns, &count)) {
1335 for (m = 0; m < count; m++) {
1336 if (!ifnet_get_address_list(ifns[m], &ifads)) {
1337 for (j = 0; ifads[j] != NULL && cnt < ADDRSPERSITE; j++) {
1338 afsi_SetServerIPRank(sa, ifads[j]);
1341 ifnet_free_address_list(ifads);
1344 ifnet_list_free(ifns);
1347 #elif defined(AFS_DARWIN_ENV)
1351 TAILQ_FOREACH(ifn, &ifnet, if_link) {
1352 TAILQ_FOREACH(ifa, &ifn->if_addrhead, ifa_link) {
1353 afsi_SetServerIPRank(sa, ifa);
1355 #elif defined(AFS_FBSD_ENV)
1357 struct in_ifaddr *ifa;
1358 #if defined(AFS_FBSD80_ENV)
1359 TAILQ_FOREACH(ifa, &V_in_ifaddrhead, ia_link) {
1361 TAILQ_FOREACH(ifa, &in_ifaddrhead, ia_link) {
1363 afsi_SetServerIPRank(sa, &ifa->ia_ifa);
1365 #elif defined(AFS_OBSD_ENV)
1367 extern struct in_ifaddrhead in_ifaddr;
1368 struct in_ifaddr *ifa;
1369 for (ifa = in_ifaddr.tqh_first; ifa; ifa = ifa->ia_list.tqe_next)
1370 afsi_SetServerIPRank(sa, ifa);
1372 #elif defined(AFS_NBSD40_ENV)
1374 extern struct in_ifaddrhead in_ifaddrhead;
1375 struct in_ifaddr *ifa;
1376 for (ifa = in_ifaddrhead.tqh_first; ifa; ifa = ifa->ia_list.tqe_next)
1377 afsi_SetServerIPRank(sa, ifa);
1381 struct in_ifaddr *ifa;
1382 for (ifa = in_ifaddr; ifa; ifa = ifa->ia_next) {
1383 afsi_SetServerIPRank(sa, ifa);
1386 #endif /* USEIFADDR */
1390 #endif /* AFS_SUN5_ENV */
1391 #endif /* else AFS_USERSPACE_IP_ADDR */
1392 sa->sa_iprank += afs_randomMod15();
1395 } /* afs_SetServerPrefs */
1403 /* afs_FlushServer()
1404 * The addresses on this server struct has changed in some way and will
1405 * clean up all other structures that may reference it.
1406 * The afs_xserver, afs_xvcb and afs_xsrvAddr locks are assumed taken.
1409 afs_FlushServer(struct server *srvp, struct volume *tv)
1412 struct server *ts, **pts;
1414 /* Find any volumes residing on this server and flush their state */
1415 afs_ResetVolumes(srvp, tv);
1417 /* Flush all callbacks in the all vcaches for this specific server */
1418 afs_FlushServerCBs(srvp);
1420 /* Remove all the callbacks structs */
1422 struct afs_cbr *cb, *cbnext;
1424 for (cb = srvp->cbrs; cb; cb = cbnext) {
1427 } srvp->cbrs = (struct afs_cbr *)0;
1430 /* If no more srvAddr structs hanging off of this server struct,
1434 /* Remove the server structure from the cell list - if there */
1435 afs_RemoveCellEntry(srvp);
1437 /* Remove from the afs_servers hash chain */
1438 for (i = 0; i < NSERVERS; i++) {
1439 for (pts = &(afs_servers[i]), ts = *pts; ts;
1440 pts = &(ts->next), ts = *pts) {
1448 *pts = ts->next; /* Found it. Remove it */
1449 afs_osi_Free(ts, sizeof(struct server)); /* Free it */
1455 /* afs_RemoveSrvAddr()
1456 * This removes a SrvAddr structure from its server structure.
1457 * The srvAddr struct is not free'd because it connections may still
1458 * be open to it. It is up to the calling process to make sure it
1459 * remains connected to a server struct.
1460 * The afs_xserver and afs_xsrvAddr locks are assumed taken.
1461 * It is not removed from the afs_srvAddrs hash chain.
1462 * If resetting volumes, do not reset volume tv
1465 afs_RemoveSrvAddr(struct srvAddr *sap, struct volume *tv)
1467 struct srvAddr **psa, *sa;
1474 /* Find the srvAddr in the server's list and remove it */
1475 for (psa = &(srv->addr), sa = *psa; sa; psa = &(sa->next_sa), sa = *psa) {
1483 /* Flush the server struct since it's IP address has changed */
1484 afs_FlushServer(srv, tv);
1488 /* afs_GetCapabilities
1489 * Try and retrieve capabilities of a given file server. Carps on actual
1490 * failure. Servers are not expected to support this RPC. */
1492 afs_GetCapabilities(struct server *ts)
1494 Capabilities caps = {0, NULL};
1495 struct vrequest *treq = NULL;
1496 struct afs_conn *tc;
1497 struct unixuser *tu;
1498 struct rx_connection *rxconn;
1501 if ( !ts || !ts->cell )
1503 if ( !afs_osi_credp )
1506 if ((code = afs_CreateReq(&treq, afs_osi_credp)))
1508 tu = afs_GetUser(treq->uid, ts->cell->cellNum, SHARED_LOCK);
1510 afs_DestroyReq(treq);
1513 tc = afs_ConnBySA(ts->addr, ts->cell->fsport, ts->cell->cellNum, tu, 0, 1,
1514 SHARED_LOCK, 0, &rxconn);
1515 afs_PutUser(tu, SHARED_LOCK);
1517 afs_DestroyReq(treq);
1520 /* InitCallBackStateN, triggered by our RPC, may need this */
1521 ReleaseWriteLock(&afs_xserver);
1522 code = RXAFS_GetCapabilities(rxconn, &caps);
1523 ObtainWriteLock(&afs_xserver, 723);
1524 /* we forced a conn above; important we mark it down if needed */
1525 if ((code < 0) && (code != RXGEN_OPCODE)) {
1526 afs_ServerDown(tc->parent->srvr, code, rxconn);
1527 ForceNewConnections(tc->parent->srvr); /* multi homed clients */
1529 afs_PutConn(tc, rxconn, SHARED_LOCK);
1530 if ( code && code != RXGEN_OPCODE ) {
1531 afs_warn("RXAFS_GetCapabilities failed with code %d\n", code);
1532 /* better not be anything to free. we failed! */
1533 afs_DestroyReq(treq);
1537 ts->flags |= SCAPS_KNOWN;
1539 if ( caps.Capabilities_len > 0 ) {
1540 ts->capabilities = caps.Capabilities_val[0];
1541 xdr_free((xdrproc_t)xdr_Capabilities, &caps);
1542 caps.Capabilities_len = 0;
1543 caps.Capabilities_val = NULL;
1546 afs_DestroyReq(treq);
1549 static struct server *
1550 afs_SearchServer(u_short aport, afsUUID * uuidp, afs_int32 locktype,
1551 struct server **oldts, afs_int32 addr_uniquifier)
1553 struct server *ts = afs_FindServer(0, aport, uuidp, locktype);
1554 if (ts && (ts->sr_addr_uniquifier == addr_uniquifier) && ts->addr) {
1555 /* Found a server struct that is multihomed and same
1556 * uniqufier (same IP addrs). The above if statement is the
1557 * same as in InstallUVolumeEntry().
1562 *oldts = ts; /* Will reuse if same uuid */
1567 * Return an updated and properly initialized server structure.
1569 * Takes a server ID, cell, and port.
1570 * If server does not exist, then one will be created.
1571 * @param[in] aserverp
1572 * The server address in network byte order
1573 * @param[in] nservers
1574 * The number of IP addresses claimed by the server
1576 * The cell the server is in
1578 * The port for the server (fileserver or vlserver) in network byte order
1579 * @param[in] locktype
1580 * The type of lock to hold when iterating server hash (unused).
1582 * The uuid for servers supporting one.
1583 * @param[in] addr_uniquifier
1584 * The vldb-provider per-instantiated-server uniquifer counter.
1586 * A volume not to reset information for if the server addresses
1590 * A server structure matching the request.
1593 afs_GetServer(afs_uint32 *aserverp, afs_int32 nservers, afs_int32 acell,
1594 u_short aport, afs_int32 locktype, afsUUID * uuidp,
1595 afs_int32 addr_uniquifier, struct volume *tv)
1597 struct server *oldts = 0, *ts, *newts, *orphts = 0;
1598 struct srvAddr *oldsa, *newsa, *nextsa, *orphsa;
1599 afs_int32 iphash, k, srvcount = 0;
1600 unsigned int srvhash;
1602 AFS_STATCNT(afs_GetServer);
1604 ObtainSharedLock(&afs_xserver, 13);
1606 /* Check if the server struct exists and is up to date */
1609 panic("afs_GetServer: incorrect count of servers");
1610 ObtainReadLock(&afs_xsrvAddr);
1611 ts = afs_FindServer(aserverp[0], aport, NULL, locktype);
1612 ReleaseReadLock(&afs_xsrvAddr);
1613 if (ts && !(ts->flags & SRVR_MULTIHOMED)) {
1614 /* Found a server struct that is not multihomed and has the
1615 * IP address associated with it. A correct match.
1617 ReleaseSharedLock(&afs_xserver);
1622 panic("afs_GetServer: incorrect count of servers");
1624 ts = afs_SearchServer(aport, uuidp, locktype, &oldts, addr_uniquifier);
1626 ReleaseSharedLock(&afs_xserver);
1632 * Lock hierarchy requires xvcb, then xserver. We *have* xserver.
1633 * Do a little dance and see if we can grab xvcb. If not, we
1634 * need to recheck that oldts is still right after a drop and reobtain.
1636 if (EWOULDBLOCK == NBObtainWriteLock(&afs_xvcb, 300)) {
1637 ReleaseSharedLock(&afs_xserver);
1638 ObtainWriteLock(&afs_xvcb, 299);
1639 ObtainWriteLock(&afs_xserver, 35);
1641 /* we don't know what changed while we didn't hold the lock */
1643 ts = afs_SearchServer(aport, uuidp, locktype, &oldts,
1646 ReleaseWriteLock(&afs_xserver);
1647 ReleaseWriteLock(&afs_xvcb);
1651 UpgradeSToWLock(&afs_xserver, 36);
1653 ObtainWriteLock(&afs_xsrvAddr, 116);
1654 srvcount = afs_totalServers;
1656 /* Reuse/allocate a new server structure */
1660 newts = afs_osi_Alloc(sizeof(struct server));
1662 panic("malloc of server struct");
1664 memset(newts, 0, sizeof(struct server));
1666 /* Add the server struct to the afs_servers[] hash chain */
1668 (uuidp ? (afs_uuid_hash(uuidp) % NSERVERS) : SHash(aserverp[0]));
1669 newts->next = afs_servers[srvhash];
1670 afs_servers[srvhash] = newts;
1673 /* Initialize the server structure */
1674 if (uuidp) { /* Multihomed */
1675 newts->sr_uuid = *uuidp;
1676 newts->sr_addr_uniquifier = addr_uniquifier;
1677 newts->flags |= SRVR_MULTIHOMED;
1680 /* Use the afs_GetCellStale variant to avoid afs_GetServer recursion. */
1681 newts->cell = afs_GetCellStale(acell, 0);
1683 /* For each IP address we are registering */
1684 for (k = 0; k < nservers; k++) {
1685 iphash = SHash(aserverp[k]);
1687 /* Check if the srvAddr structure already exists. If so, remove
1688 * it from its server structure and add it to the new one.
1690 for (oldsa = afs_srvAddrs[iphash]; oldsa; oldsa = oldsa->next_bkt) {
1691 if ((oldsa->sa_ip == aserverp[k]) && (oldsa->sa_portal == aport))
1694 if (oldsa && (oldsa->server != newts)) {
1695 afs_RemoveSrvAddr(oldsa, tv); /* Remove from its server struct */
1696 oldsa->next_sa = newts->addr; /* Add to the new server struct */
1697 newts->addr = oldsa;
1700 /* Reuse/allocate a new srvAddr structure */
1704 newsa = afs_osi_Alloc(sizeof(struct srvAddr));
1706 panic("malloc of srvAddr struct");
1707 afs_totalSrvAddrs++;
1708 memset(newsa, 0, sizeof(struct srvAddr));
1710 /* Add the new srvAddr to the afs_srvAddrs[] hash chain */
1711 newsa->next_bkt = afs_srvAddrs[iphash];
1712 afs_srvAddrs[iphash] = newsa;
1714 /* Hang off of the server structure */
1715 newsa->next_sa = newts->addr;
1716 newts->addr = newsa;
1718 /* Initialize the srvAddr Structure */
1719 newsa->sa_ip = aserverp[k];
1720 newsa->sa_portal = aport;
1723 /* Update the srvAddr Structure */
1724 newsa->server = newts;
1725 if (newts->flags & SRVR_ISDOWN)
1726 newsa->sa_flags |= SRVADDR_ISDOWN;
1728 newsa->sa_flags |= SRVADDR_MH;
1730 newsa->sa_flags &= ~SRVADDR_MH;
1732 /* Compute preference values and resort */
1733 if (!newsa->sa_iprank) {
1734 afs_SetServerPrefs(newsa); /* new server rank */
1737 afs_SortOneServer(newts); /* Sort by rank */
1739 /* If we reused the server struct, remove any of its srvAddr
1740 * structs that will no longer be associated with this server.
1742 if (oldts) { /* reused the server struct */
1743 for (orphsa = newts->addr; orphsa; orphsa = nextsa) {
1744 nextsa = orphsa->next_sa;
1745 for (k = 0; k < nservers; k++) {
1746 if (orphsa->sa_ip == aserverp[k])
1747 break; /* belongs */
1750 continue; /* belongs */
1752 /* Have a srvAddr struct. Now get a server struct (if not already) */
1754 orphts = afs_osi_Alloc(sizeof(struct server));
1756 panic("malloc of lo server struct");
1757 memset(orphts, 0, sizeof(struct server));
1760 /* Add the orphaned server to the afs_servers[] hash chain.
1761 * Its iphash does not matter since we never look up the server
1762 * in the afs_servers table by its ip address (only by uuid -
1763 * which this has none).
1765 iphash = SHash(aserverp[k]);
1766 orphts->next = afs_servers[iphash];
1767 afs_servers[iphash] = orphts;
1770 /* Use the afs_GetCellStale variant to avoid afs_GetServer recursion. */
1771 orphts->cell = afs_GetCellStale(acell, 0);
1774 /* Hang the srvAddr struct off of the server structure. The server
1775 * may have multiple srvAddrs, but it won't be marked multihomed.
1777 afs_RemoveSrvAddr(orphsa, tv); /* remove */
1778 orphsa->next_sa = orphts->addr; /* hang off server struct */
1779 orphts->addr = orphsa;
1780 orphsa->server = orphts;
1781 orphsa->sa_flags |= SRVADDR_NOUSE; /* flag indicating not in use */
1782 orphsa->sa_flags &= ~SRVADDR_MH; /* Not multihomed */
1785 /* We can't need this below, and won't reacquire */
1786 ReleaseWriteLock(&afs_xvcb);
1788 srvcount = afs_totalServers - srvcount; /* # servers added and removed */
1790 struct afs_stats_SrvUpDownInfo *upDownP;
1791 /* With the introduction of this new record, we need to adjust the
1792 * proper individual & global server up/down info.
1794 upDownP = GetUpDownStats(newts);
1795 upDownP->numTtlRecords += srvcount;
1796 afs_stats_cmperf.srvRecords += srvcount;
1797 if (afs_stats_cmperf.srvRecords > afs_stats_cmperf.srvRecordsHWM)
1798 afs_stats_cmperf.srvRecordsHWM = afs_stats_cmperf.srvRecords;
1800 /* We can't need this below, and won't reacquire */
1801 ReleaseWriteLock(&afs_xvcb);
1803 ReleaseWriteLock(&afs_xsrvAddr);
1805 if ( aport == AFS_FSPORT && !(newts->flags & SCAPS_KNOWN))
1806 afs_GetCapabilities(newts);
1808 ReleaseWriteLock(&afs_xserver);
1810 } /* afs_GetServer */
1813 afs_ActivateServer(struct srvAddr *sap)
1815 osi_timeval_t currTime; /*Filled with current time */
1816 osi_timeval_t *currTimeP; /*Ptr to above */
1817 struct afs_stats_SrvUpDownInfo *upDownP; /*Ptr to up/down info record */
1818 struct server *aserver = sap->server;
1820 if (!(aserver->flags & AFS_SERVER_FLAG_ACTIVATED)) {
1822 * This server record has not yet been activated. Go for it,
1823 * recording its ``birth''.
1825 aserver->flags |= AFS_SERVER_FLAG_ACTIVATED;
1826 currTimeP = &currTime;
1827 osi_GetuTime(currTimeP);
1828 aserver->activationTime = currTime.tv_sec;
1829 upDownP = GetUpDownStats(aserver);
1830 if (aserver->flags & SRVR_ISDOWN) {
1831 upDownP->numDownRecords++;
1833 upDownP->numUpRecords++;
1834 upDownP->numRecordsNeverDown++;
1840 afs_RemoveAllConns(void)
1843 struct server *ts, *nts;
1846 ObtainReadLock(&afs_xserver);
1847 ObtainWriteLock(&afs_xconn, 1001);
1849 /*printf("Destroying connections ... ");*/
1850 for (i = 0; i < NSERVERS; i++) {
1851 for (ts = afs_servers[i]; ts; ts = nts) {
1853 for (sa = ts->addr; sa; sa = sa->next_sa) {
1855 afs_ReleaseConns(sa->conns);
1862 /*printf("done\n");*/
1864 ReleaseWriteLock(&afs_xconn);
1865 ReleaseReadLock(&afs_xserver);
1870 afs_MarkAllServersUp(void)
1876 ObtainWriteLock(&afs_xserver, 721);
1877 ObtainWriteLock(&afs_xsrvAddr, 722);
1878 for (i = 0; i< NSERVERS; i++) {
1879 for (ts = afs_servers[i]; ts; ts = ts->next) {
1880 for (sa = ts->addr; sa; sa = sa->next_sa) {
1881 afs_MarkServerUpOrDown(sa, 0);
1885 ReleaseWriteLock(&afs_xsrvAddr);
1886 ReleaseWriteLock(&afs_xserver);