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)
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
243 ("afs: Lost contact with volume location server ", sa, "", 1, code);
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);
304 afs_PutConn(tc, rxconn, SHARED_LOCK);
306 * If probe worked, or probe call not yet defined (for compatibility
307 * with old vlsevers), then we treat this server as running again
309 if (code == 0 || (code <= -450 && code >= -470)) {
310 if (tc->parent->srvr == sa) {
311 afs_MarkServerUpOrDown(sa, 0);
312 print_internet_address("afs: volume location server ", sa,
313 " is back up", 2, code);
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);
521 CkSrv_MarkUpDown(struct afs_conn **conns, int nconns, afs_int32 *results)
527 for(i = 0; i < nconns; i++){
529 sa = tc->parent->srvr;
531 if (( results[i] >= 0 ) && (sa->sa_flags & SRVADDR_ISDOWN) &&
532 (tc->parent->srvr == sa)) {
534 print_internet_address("afs: file server ", sa, " is back up", 2,
537 ObtainWriteLock(&afs_xserver, 244);
538 ObtainWriteLock(&afs_xsrvAddr, 245);
539 afs_MarkServerUpOrDown(sa, 0);
540 ReleaseWriteLock(&afs_xsrvAddr);
541 ReleaseWriteLock(&afs_xserver);
543 if (afs_waitForeverCount) {
544 afs_osi_Wakeup(&afs_waitForever);
547 if (results[i] < 0) {
549 afs_ServerDown(sa, results[i]);
550 ForceNewConnections(sa); /* multi homed clients */
557 CkSrv_GetCaps(int nconns, struct rx_connection **rxconns,
558 struct afs_conn **conns)
565 caps = afs_osi_Alloc(nconns * sizeof (Capabilities));
566 osi_Assert(caps != NULL);
567 memset(caps, 0, nconns * sizeof(Capabilities));
569 results = afs_osi_Alloc(nconns * sizeof (afs_int32));
570 osi_Assert(results != NULL);
573 multi_Rx(rxconns,nconns)
575 multi_RXAFS_GetCapabilities(&caps[multi_i]);
576 results[multi_i] = multi_error;
580 for ( i = 0 ; i < nconns ; i++ ) {
581 ts = conns[i]->parent->srvr->server;
584 ts->capabilities = 0;
585 ts->flags |= SCAPS_KNOWN;
586 if ( results[i] == RXGEN_OPCODE ) {
587 /* Mark server as up - it responded */
591 if ( results[i] >= 0 )
592 /* we currently handle 32-bits of capabilities */
593 if (caps[i].Capabilities_len > 0) {
594 ts->capabilities = caps[i].Capabilities_val[0];
595 xdr_free((xdrproc_t)xdr_Capabilities, &caps[i]);
596 caps[i].Capabilities_val = NULL;
597 caps[i].Capabilities_len = 0;
600 CkSrv_MarkUpDown(conns, nconns, results);
602 afs_osi_Free(caps, nconns * sizeof(Capabilities));
603 afs_osi_Free(results, nconns * sizeof(afs_int32));
606 /* check down servers (if adown), or running servers (if !adown) */
608 afs_CheckServers(int adown, struct cell *acellp)
610 afs_LoopServers(adown?AFS_LS_DOWN:AFS_LS_UP, acellp, 1, CkSrv_GetCaps, NULL);
613 /* adown: AFS_LS_UP - check only up
614 * AFS_LS_DOWN - check only down.
615 * AFS_LS_ALL - check all */
617 afs_LoopServers(int adown, struct cell *acellp, int vlalso,
618 void (*func1) (int nservers, struct rx_connection **rxconns,
619 struct afs_conn **conns),
620 void (*func2) (int nservers, struct rx_connection **rxconns,
621 struct afs_conn **conns))
623 struct vrequest treq;
626 struct afs_conn *tc = NULL;
631 struct srvAddr **addrs;
632 struct afs_conn **conns;
634 struct rx_connection **rxconns;
635 afs_int32 *conntimer;
637 AFS_STATCNT(afs_CheckServers);
640 * No sense in doing the server checks if we are running in disconnected
643 if (AFS_IS_DISCONNECTED)
646 if ((code = afs_InitReq(&treq, afs_osi_credp)))
648 ObtainReadLock(&afs_xserver); /* Necessary? */
649 ObtainReadLock(&afs_xsrvAddr);
652 for (i = 0; i < NSERVERS; i++) {
653 for (sa = afs_srvAddrs[i]; sa; sa = sa->next_bkt) {
658 addrs = afs_osi_Alloc(srvAddrCount * sizeof(*addrs));
659 osi_Assert(addrs != NULL);
661 for (i = 0; i < NSERVERS; i++) {
662 for (sa = afs_srvAddrs[i]; sa; sa = sa->next_bkt) {
663 if (j >= srvAddrCount)
669 ReleaseReadLock(&afs_xsrvAddr);
670 ReleaseReadLock(&afs_xserver);
672 conns = afs_osi_Alloc(j * sizeof(struct afs_conn *));
673 osi_Assert(conns != NULL);
674 rxconns = afs_osi_Alloc(j * sizeof(struct rx_connection *));
675 osi_Assert(rxconns != NULL);
676 conntimer = afs_osi_Alloc(j * sizeof (afs_int32));
677 osi_Assert(conntimer != NULL);
680 for (i = 0; i < j; i++) {
681 struct rx_connection *rxconn;
687 /* See if a cell to check was specified. If it is spec'd and not
688 * this server's cell, just skip the server.
690 if (acellp && acellp != ts->cell)
693 if (((adown==AFS_LS_DOWN) && !(sa->sa_flags & SRVADDR_ISDOWN))
694 || ((adown==AFS_LS_UP) && (sa->sa_flags & SRVADDR_ISDOWN)))
697 /* check vlserver with special code */
698 if (sa->sa_portal == AFS_VLPORT) {
700 CheckVLServer(sa, &treq);
704 if (!ts->cell) /* not really an active server, anyway, it must */
705 continue; /* have just been added by setsprefs */
707 /* get a connection, even if host is down; bumps conn ref count */
708 tu = afs_GetUser(treq.uid, ts->cell->cellNum, SHARED_LOCK);
709 tc = afs_ConnBySA(sa, ts->cell->fsport, ts->cell->cellNum, tu,
710 1 /*force */ , 1 /*create */ , SHARED_LOCK, 0,
712 afs_PutUser(tu, SHARED_LOCK);
716 if ((sa->sa_flags & SRVADDR_ISDOWN) || afs_HaveCallBacksFrom(sa->server)) {
718 rxconns[nconns]=rxconn;
719 if (sa->sa_flags & SRVADDR_ISDOWN) {
720 rx_SetConnDeadTime(rxconn, 3);
726 } else /* not holding, kill ref */
727 afs_PutConn(tc, rxconn, SHARED_LOCK);
728 } /* Outer loop over addrs */
730 afs_osi_Free(addrs, srvAddrCount * sizeof(*addrs));
733 (*func1)(nconns, rxconns, conns);
736 (*func2)(nconns, rxconns, conns);
739 for (i = 0; i < nconns; i++) {
740 if (conntimer[i] == 1)
741 rx_SetConnDeadTime(rxconns[i], afs_rx_deadtime);
742 afs_PutConn(conns[i], rxconns[i], SHARED_LOCK); /* done with it now */
745 afs_osi_Free(conns, j * sizeof(struct afs_conn *));
746 afs_osi_Free(rxconns, j * sizeof(struct rx_connection *));
747 afs_osi_Free(conntimer, j * sizeof(afs_int32));
749 } /*afs_CheckServers*/
752 /* find a server structure given the host address */
754 afs_FindServer(afs_int32 aserver, afs_uint16 aport, afsUUID * uuidp,
761 AFS_STATCNT(afs_FindServer);
763 i = afs_uuid_hash(uuidp) % NSERVERS;
764 for (ts = afs_servers[i]; ts; ts = ts->next) {
765 if ((ts->flags & SRVR_MULTIHOMED)
767 (memcmp((char *)uuidp, (char *)&ts->sr_uuid, sizeof(*uuidp))
768 == 0) && (!ts->addr || (ts->addr->sa_portal == aport)))
773 for (sa = afs_srvAddrs[i]; sa; sa = sa->next_bkt) {
774 if ((sa->sa_ip == aserver) && (sa->sa_portal == aport)) {
781 } /*afs_FindServer */
784 /* some code for creating new server structs and setting preferences follows
785 * in the next few lines...
788 #define MAXDEFRANK 60000
789 #define DEFRANK 40000
791 /* Random number generator and constants from KnuthV2 2d ed, p170 */
797 a is 0.73m should be 0.01m .. 0.99m
798 c is more or less immaterial. 1 or a is suggested.
800 NB: LOW ORDER BITS are not very random. To get small random numbers,
801 treat result as <1, with implied binary point, and multiply by
803 NB: Has to be unsigned, since shifts on signed quantities may preserve
806 /* added rxi_getaddr() to try to get as much initial randomness as
807 possible, since at least one customer reboots ALL their clients
808 simultaneously -- so osi_Time is bound to be the same on some of the
809 clients. This is probably OK, but I don't want to see too much of it.
812 #define ranstage(x) (x)= (afs_uint32) (3141592621U*((afs_uint32)x)+1)
817 static afs_int32 state = 0;
820 AFS_STATCNT(afs_random);
825 * 0xfffffff0 was changed to (~0 << 4) since it works no matter how many
826 * bits are in a tv_usec
828 state = (t.tv_usec & (~0 << 4)) + (rxi_getaddr() & 0xff);
829 state += (t.tv_sec & 0xff);
830 for (i = 0; i < 30; i++) {
840 /* returns int 0..14 using the high bits of a pseudo-random number instead of
841 the low bits, as the low bits are "less random" than the high ones...
842 slight roundoff error exists, an excercise for the reader.
843 need to multiply by something with lots of ones in it, so multiply by
844 8 or 16 is right out.
847 afs_randomMod15(void)
851 temp = afs_random() >> 4;
852 temp = (temp * 15) >> 28;
858 afs_randomMod127(void)
862 temp = afs_random() >> 7;
863 temp = (temp * 127) >> 25;
868 /* afs_SortOneServer()
869 * Sort all of the srvAddrs, of a server struct, by rank from low to high.
872 afs_SortOneServer(struct server *asp)
874 struct srvAddr **rootsa, *lowsa, *tsa, *lowprev;
877 for (rootsa = &(asp->addr); *rootsa; rootsa = &(lowsa->next_sa)) {
879 lowsa = *rootsa; /* lowest sa is the first one */
880 lowrank = lowsa->sa_iprank;
882 for (tsa = *rootsa; tsa->next_sa; tsa = tsa->next_sa) {
883 rank = tsa->next_sa->sa_iprank;
884 if (rank < lowrank) {
886 lowsa = tsa->next_sa;
887 lowrank = lowsa->sa_iprank;
890 if (lowprev) { /* found one lower, so rearrange them */
891 lowprev->next_sa = lowsa->next_sa;
892 lowsa->next_sa = *rootsa;
899 * Sort the pointer to servers by the server's rank (its lowest rank).
900 * It is assumed that the server already has its IP addrs sorted (the
901 * first being its lowest rank: afs_GetServer() calls afs_SortOneServer()).
904 afs_SortServers(struct server *aservers[], int count)
909 AFS_STATCNT(afs_SortServers);
911 for (i = 0; i < count; i++) {
914 for (low = i, j = i + 1; j <= count; j++) {
915 if ((!aservers[j]) || (!aservers[j]->addr))
917 if ((!aservers[low]) || (!aservers[low]->addr))
919 if (aservers[j]->addr->sa_iprank < aservers[low]->addr->sa_iprank) {
925 aservers[i] = aservers[low];
929 } /*afs_SortServers */
931 /* afs_SetServerPrefs is rather system-dependent. It pokes around in kernel
932 data structures to determine what the local IP addresses and subnet masks
933 are in order to choose which server(s) are on the local subnet.
935 As I see it, there are several cases:
936 1. The server address is one of this host's local addresses. In this case
937 this server is to be preferred over all others.
938 2. The server is on the same subnet as one of the this host's local
939 addresses. (ie, an odd-sized subnet, not class A,B,orC)
940 3. The server is on the same net as this host (class A,B or C)
941 4. The server is on a different logical subnet or net than this host, but
942 this host is a 'metric 0 gateway' to it. Ie, two address-spaces share
944 5. This host has a direct (point-to-point, ie, PPP or SLIP) link to the
946 6. This host and the server are disjoint.
948 That is a rough order of preference. If a point-to-point link has a high
949 metric, I'm assuming that it is a very slow link, and putting it at the
950 bottom of the list (at least until RX works better over slow links). If
951 its metric is 1, I'm assuming that it's relatively fast (T1) and putting
953 It's not easy to check for case #4, so I'm ignoring it for the time being.
955 BSD "if" code keeps track of some rough network statistics (cf 'netstat -i')
956 That could be used to prefer certain servers fairly easily. Maybe some
959 NOTE: this code is very system-dependent, and very dependent on the TCP/IP
960 protocols (well, addresses that are stored in uint32s, at any rate).
963 #define IA_DST(ia)((struct sockaddr_in *)(&((struct in_ifaddr *)ia)->ia_dstaddr))
964 #define IA_BROAD(ia)((struct sockaddr_in *)(&((struct in_ifaddr *)ia)->ia_broadaddr))
966 /* SA2ULONG takes a sockaddr_in, not a sockaddr (same thing, just cast it!) */
967 #define SA2ULONG(sa) ((sa)->sin_addr.s_addr)
972 #define PPWEIGHT 4096
976 #ifdef AFS_USERSPACE_IP_ADDR
978 #define afs_min(A,B) ((A)<(B)) ? (A) : (B)
981 * The IP addresses and ranks are determined by afsd (in user space) and
982 * passed into the kernel at startup time through the AFSOP_ADVISEADDR
983 * system call. These are stored in the data structure
984 * called 'afs_cb_interface'.
986 * struct srvAddr *sa; remote server
987 * afs_int32 addr; one of my local addr in net order
988 * afs_uint32 subnetmask; subnet mask of local addr in net order
992 afsi_SetServerIPRank(struct srvAddr *sa, afs_int32 addr,
993 afs_uint32 subnetmask)
995 afs_uint32 myAddr, myNet, mySubnet, netMask;
996 afs_uint32 serverAddr;
998 myAddr = ntohl(addr); /* one of my IP addr in host order */
999 serverAddr = ntohl(sa->sa_ip); /* server's IP addr in host order */
1000 subnetmask = ntohl(subnetmask); /* subnet mask in host order */
1002 if (IN_CLASSA(myAddr))
1003 netMask = IN_CLASSA_NET;
1004 else if (IN_CLASSB(myAddr))
1005 netMask = IN_CLASSB_NET;
1006 else if (IN_CLASSC(myAddr))
1007 netMask = IN_CLASSC_NET;
1011 myNet = myAddr & netMask;
1012 mySubnet = myAddr & subnetmask;
1014 if ((serverAddr & netMask) == myNet) {
1015 if ((serverAddr & subnetmask) == mySubnet) {
1016 if (serverAddr == myAddr) { /* same machine */
1017 sa->sa_iprank = afs_min(sa->sa_iprank, TOPR);
1018 } else { /* same subnet */
1019 sa->sa_iprank = afs_min(sa->sa_iprank, HI);
1021 } else { /* same net */
1022 sa->sa_iprank = afs_min(sa->sa_iprank, MED);
1027 #else /* AFS_USERSPACE_IP_ADDR */
1028 #if (! defined(AFS_SUN5_ENV)) && (! defined(AFS_DARWIN_ENV)) && (! defined(AFS_OBSD47_ENV)) && (! defined(AFS_FBSD_ENV)) && defined(USEIFADDR)
1030 afsi_SetServerIPRank(struct srvAddr *sa, struct in_ifaddr *ifa)
1032 struct sockaddr_in *sin;
1035 if ((ntohl(sa->sa_ip) & ifa->ia_netmask) == ifa->ia_net) {
1036 if ((ntohl(sa->sa_ip) & ifa->ia_subnetmask) == ifa->ia_subnet) {
1038 if (SA2ULONG(sin) == ntohl(sa->sa_ip)) { /* ie, ME!!! */
1039 sa->sa_iprank = TOPR;
1041 t = HI + ifa->ia_ifp->if_metric; /* case #2 */
1042 if (sa->sa_iprank > t)
1046 t = MED + ifa->ia_ifp->if_metric; /* case #3 */
1047 if (sa->sa_iprank > t)
1051 #if defined(IFF_POINTOPOINT) && !defined(UKERNEL)
1052 /* check for case #4 -- point-to-point link */
1053 if ((ifa->ia_ifp->if_flags & IFF_POINTOPOINT)
1054 && (SA2ULONG(IA_DST(ifa)) == ntohl(sa->sa_ip))) {
1055 if (ifa->ia_ifp->if_metric >= (MAXDEFRANK - MED) / PPWEIGHT)
1058 t = MED + (PPWEIGHT << ifa->ia_ifp->if_metric);
1059 if (sa->sa_iprank > t)
1062 #endif /* IFF_POINTOPOINT */
1064 #endif /*(!defined(AFS_SUN5_ENV)) && defined(USEIFADDR) */
1065 #if (defined(AFS_DARWIN_ENV) || defined(AFS_OBSD47_ENV) || defined(AFS_FBSD_ENV)) && defined(USEIFADDR)
1067 #define afs_min(A,B) ((A)<(B)) ? (A) : (B)
1070 afsi_SetServerIPRank(struct srvAddr *sa, rx_ifaddr_t ifa)
1072 struct sockaddr sout;
1073 struct sockaddr_in *sin;
1074 #if defined(AFS_DARWIN80_ENV) && !defined(UKERNEL)
1080 afs_uint32 subnetmask, myAddr, myNet, myDstaddr, mySubnet, netMask;
1081 afs_uint32 serverAddr;
1083 if (rx_ifaddr_address_family(ifa) != AF_INET)
1085 t = rx_ifaddr_address(ifa, &sout, sizeof(sout));
1087 sin = (struct sockaddr_in *)&sout;
1088 myAddr = ntohl(sin->sin_addr.s_addr); /* one of my IP addr in host order */
1092 serverAddr = ntohl(sa->sa_ip); /* server's IP addr in host order */
1093 t = rx_ifaddr_netmask(ifa, &sout, sizeof(sout));
1095 sin = (struct sockaddr_in *)&sout;
1096 subnetmask = ntohl(sin->sin_addr.s_addr); /* subnet mask in host order */
1100 t = rx_ifaddr_dstaddress(ifa, &sout, sizeof(sout));
1102 sin = (struct sockaddr_in *)&sout;
1103 myDstaddr = ntohl(sin->sin_addr.s_addr);
1108 if (IN_CLASSA(myAddr))
1109 netMask = IN_CLASSA_NET;
1110 else if (IN_CLASSB(myAddr))
1111 netMask = IN_CLASSB_NET;
1112 else if (IN_CLASSC(myAddr))
1113 netMask = IN_CLASSC_NET;
1117 myNet = myAddr & netMask;
1118 mySubnet = myAddr & subnetmask;
1120 if ((serverAddr & netMask) == myNet) {
1121 if ((serverAddr & subnetmask) == mySubnet) {
1122 if (serverAddr == myAddr) { /* same machine */
1123 sa->sa_iprank = afs_min(sa->sa_iprank, TOPR);
1124 } else { /* same subnet */
1125 sa->sa_iprank = afs_min(sa->sa_iprank, HI + rx_ifnet_metric(rx_ifaddr_ifnet(ifa)));
1127 } else { /* same net */
1128 sa->sa_iprank = afs_min(sa->sa_iprank, MED + rx_ifnet_metric(rx_ifaddr_ifnet(ifa)));
1131 #ifdef IFF_POINTTOPOINT
1132 /* check for case #4 -- point-to-point link */
1133 if ((rx_ifnet_flags(rx_ifaddr_ifnet(ifa)) & IFF_POINTOPOINT)
1134 && (myDstaddr == serverAddr)) {
1135 if (rx_ifnet_metric(rx_ifaddr_ifnet(ifa)) >= (MAXDEFRANK - MED) / PPWEIGHT)
1138 t = MED + (PPWEIGHT << rx_ifnet_metric(rx_ifaddr_ifnet(ifa)));
1139 if (sa->sa_iprank > t)
1142 #endif /* IFF_POINTTOPOINT */
1144 #endif /*(!defined(AFS_SUN5_ENV)) && defined(USEIFADDR) */
1145 #endif /* else AFS_USERSPACE_IP_ADDR */
1147 #ifdef AFS_SGI62_ENV
1149 afsi_enum_set_rank(struct hashbucket *h, caddr_t mkey, caddr_t arg1,
1152 afsi_SetServerIPRank((struct srvAddr *)arg1, (struct in_ifaddr *)h);
1153 return 0; /* Never match, so we enumerate everyone */
1155 #endif /* AFS_SGI62_ENV */
1157 afs_SetServerPrefs(struct srvAddr *sa)
1159 #if defined(AFS_USERSPACE_IP_ADDR)
1163 for (i = 0; i < afs_cb_interface.numberOfInterfaces; i++) {
1164 afsi_SetServerIPRank(sa, afs_cb_interface.addr_in[i],
1165 afs_cb_interface.subnetmask[i]);
1167 #else /* AFS_USERSPACE_IP_ADDR */
1168 #if defined(AFS_SUN5_ENV)
1169 #ifdef AFS_SUN510_ENV
1172 extern struct ill_s *ill_g_headp;
1173 long *addr = (long *)ill_g_headp;
1177 int subnet, subnetmask, net, netmask;
1181 #ifdef AFS_SUN510_ENV
1182 rw_enter(&afsifinfo_lock, RW_READER);
1184 for (i = 0; (afsifinfo[i].ipaddr != NULL) && (i < ADDRSPERSITE); i++) {
1186 if (IN_CLASSA(afsifinfo[i].ipaddr)) {
1187 netmask = IN_CLASSA_NET;
1188 } else if (IN_CLASSB(afsifinfo[i].ipaddr)) {
1189 netmask = IN_CLASSB_NET;
1190 } else if (IN_CLASSC(afsifinfo[i].ipaddr)) {
1191 netmask = IN_CLASSC_NET;
1195 net = afsifinfo[i].ipaddr & netmask;
1198 /* XXXXXX Do the individual ip ranking below XXXXX */
1199 if ((sa->sa_ip & netmask) == net) {
1200 if ((sa->sa_ip & subnetmask) == subnet) {
1201 if (afsifinfo[i].ipaddr == sa->sa_ip) { /* ie, ME! */
1202 sa->sa_iprank = TOPR;
1204 sa->sa_iprank = HI + afsifinfo[i].metric; /* case #2 */
1207 sa->sa_iprank = MED + afsifinfo[i].metric; /* case #3 */
1210 sa->sa_iprank = LO + afsifinfo[i].metric; /* case #4 */
1212 /* check for case #5 -- point-to-point link */
1213 if ((afsifinfo[i].flags & IFF_POINTOPOINT)
1214 && (afsifinfo[i].dstaddr == sa->sa_ip)) {
1216 if (afsifinfo[i].metric >= (MAXDEFRANK - MED) / PPWEIGHT)
1217 sa->sa_iprank = MAXDEFRANK;
1219 sa->sa_iprank = MED + (PPWEIGHT << afsifinfo[i].metric);
1224 rw_exit(&afsifinfo_lock);
1226 for (ill = (struct ill_s *)*addr /*ill_g_headp */ ; ill;
1227 ill = ill->ill_next) {
1228 /* Make sure this is an IPv4 ILL */
1231 for (ipif = ill->ill_ipif; ipif; ipif = ipif->ipif_next) {
1232 subnet = ipif->ipif_local_addr & ipif->ipif_net_mask;
1233 subnetmask = ipif->ipif_net_mask;
1235 * Generate the local net using the local address and
1236 * whate we know about Class A, B and C networks.
1238 if (IN_CLASSA(ipif->ipif_local_addr)) {
1239 netmask = IN_CLASSA_NET;
1240 } else if (IN_CLASSB(ipif->ipif_local_addr)) {
1241 netmask = IN_CLASSB_NET;
1242 } else if (IN_CLASSC(ipif->ipif_local_addr)) {
1243 netmask = IN_CLASSC_NET;
1247 net = ipif->ipif_local_addr & netmask;
1249 /* XXXXXX Do the individual ip ranking below XXXXX */
1250 if ((sa->sa_ip & netmask) == net) {
1251 if ((sa->sa_ip & subnetmask) == subnet) {
1252 if (ipif->ipif_local_addr == sa->sa_ip) { /* ie, ME! */
1253 sa->sa_iprank = TOPR;
1255 sa->sa_iprank = HI + ipif->ipif_metric; /* case #2 */
1258 sa->sa_iprank = MED + ipif->ipif_metric; /* case #3 */
1261 sa->sa_iprank = LO + ipif->ipif_metric; /* case #4 */
1263 /* check for case #5 -- point-to-point link */
1264 if ((ipif->ipif_flags & IFF_POINTOPOINT)
1265 && (ipif->ipif_pp_dst_addr == sa->sa_ip)) {
1267 if (ipif->ipif_metric >= (MAXDEFRANK - MED) / PPWEIGHT)
1268 sa->sa_iprank = MAXDEFRANK;
1270 sa->sa_iprank = MED + (PPWEIGHT << ipif->ipif_metric);
1275 #endif /* AFS_SUN510_ENV */
1278 rx_ifnet_t ifn = NULL;
1279 struct in_ifaddr *ifad = (struct in_ifaddr *)0;
1280 struct sockaddr_in *sin;
1287 ifn = rxi_FindIfnet(sa->sa_ip, &ifad);
1289 if (ifn) { /* local, more or less */
1291 if (ifn->if_flags & IFF_LOOPBACK) {
1292 sa->sa_iprank = TOPR;
1295 #endif /* IFF_LOOPBACK */
1296 sin = (struct sockaddr_in *)IA_SIN(ifad);
1297 if (SA2ULONG(sin) == sa->sa_ip) {
1298 sa->sa_iprank = TOPR;
1301 #ifdef IFF_BROADCAST
1302 if (ifn->if_flags & IFF_BROADCAST) {
1303 if (sa->sa_ip == (sa->sa_ip & SA2ULONG(IA_BROAD(ifad)))) {
1308 #endif /* IFF_BROADCAST */
1309 #ifdef IFF_POINTOPOINT
1310 if (ifn->if_flags & IFF_POINTOPOINT) {
1311 if (sa->sa_ip == SA2ULONG(IA_DST(ifad))) {
1312 if (ifn->if_metric > 4) {
1316 sa->sa_iprank = ifn->if_metric;
1319 #endif /* IFF_POINTOPOINT */
1320 sa->sa_iprank += MED + ifn->if_metric; /* couldn't find anything better */
1322 #else /* USEIFADDR */
1326 #ifdef AFS_SGI62_ENV
1327 (void)hash_enum(&hashinfo_inaddr, afsi_enum_set_rank, HTF_INET, NULL,
1328 (caddr_t) sa, NULL);
1329 #elif defined(AFS_DARWIN80_ENV)
1337 if (!ifnet_list_get(AF_INET, &ifns, &count)) {
1338 for (m = 0; m < count; m++) {
1339 if (!ifnet_get_address_list(ifns[m], &ifads)) {
1340 for (j = 0; ifads[j] != NULL && cnt < ADDRSPERSITE; j++) {
1341 afsi_SetServerIPRank(sa, ifads[j]);
1344 ifnet_free_address_list(ifads);
1347 ifnet_list_free(ifns);
1350 #elif defined(AFS_DARWIN_ENV)
1354 TAILQ_FOREACH(ifn, &ifnet, if_link) {
1355 TAILQ_FOREACH(ifa, &ifn->if_addrhead, ifa_link) {
1356 afsi_SetServerIPRank(sa, ifa);
1358 #elif defined(AFS_FBSD_ENV)
1360 struct in_ifaddr *ifa;
1361 #if defined(AFS_FBSD80_ENV)
1362 TAILQ_FOREACH(ifa, &V_in_ifaddrhead, ia_link) {
1364 TAILQ_FOREACH(ifa, &in_ifaddrhead, ia_link) {
1366 afsi_SetServerIPRank(sa, &ifa->ia_ifa);
1368 #elif defined(AFS_OBSD_ENV)
1370 extern struct in_ifaddrhead in_ifaddr;
1371 struct in_ifaddr *ifa;
1372 for (ifa = in_ifaddr.tqh_first; ifa; ifa = ifa->ia_list.tqe_next)
1373 afsi_SetServerIPRank(sa, ifa);
1375 #elif defined(AFS_NBSD40_ENV)
1377 extern struct in_ifaddrhead in_ifaddrhead;
1378 struct in_ifaddr *ifa;
1379 for (ifa = in_ifaddrhead.tqh_first; ifa; ifa = ifa->ia_list.tqe_next)
1380 afsi_SetServerIPRank(sa, ifa);
1384 struct in_ifaddr *ifa;
1385 for (ifa = in_ifaddr; ifa; ifa = ifa->ia_next) {
1386 afsi_SetServerIPRank(sa, ifa);
1389 #endif /* USEIFADDR */
1393 #endif /* AFS_SUN5_ENV */
1394 #endif /* else AFS_USERSPACE_IP_ADDR */
1396 sa->sa_iprank += afs_randomMod15();
1399 } /* afs_SetServerPrefs */
1407 /* afs_FlushServer()
1408 * The addresses on this server struct has changed in some way and will
1409 * clean up all other structures that may reference it.
1410 * The afs_xserver, afs_xvcb and afs_xsrvAddr locks are assumed taken.
1413 afs_FlushServer(struct server *srvp, struct volume *tv)
1416 struct server *ts, **pts;
1418 /* Find any volumes residing on this server and flush their state */
1419 afs_ResetVolumes(srvp, tv);
1421 /* Flush all callbacks in the all vcaches for this specific server */
1422 afs_FlushServerCBs(srvp);
1424 /* Remove all the callbacks structs */
1426 struct afs_cbr *cb, *cbnext;
1428 for (cb = srvp->cbrs; cb; cb = cbnext) {
1431 } srvp->cbrs = (struct afs_cbr *)0;
1434 /* If no more srvAddr structs hanging off of this server struct,
1438 /* Remove the server structure from the cell list - if there */
1439 afs_RemoveCellEntry(srvp);
1441 /* Remove from the afs_servers hash chain */
1442 for (i = 0; i < NSERVERS; i++) {
1443 for (pts = &(afs_servers[i]), ts = *pts; ts;
1444 pts = &(ts->next), ts = *pts) {
1452 *pts = ts->next; /* Found it. Remove it */
1453 afs_osi_Free(ts, sizeof(struct server)); /* Free it */
1459 /* afs_RemoveSrvAddr()
1460 * This removes a SrvAddr structure from its server structure.
1461 * The srvAddr struct is not free'd because it connections may still
1462 * be open to it. It is up to the calling process to make sure it
1463 * remains connected to a server struct.
1464 * The afs_xserver and afs_xsrvAddr locks are assumed taken.
1465 * It is not removed from the afs_srvAddrs hash chain.
1466 * If resetting volumes, do not reset volume tv
1469 afs_RemoveSrvAddr(struct srvAddr *sap, struct volume *tv)
1471 struct srvAddr **psa, *sa;
1478 /* Find the srvAddr in the server's list and remove it */
1479 for (psa = &(srv->addr), sa = *psa; sa; psa = &(sa->next_sa), sa = *psa) {
1487 /* Flush the server struct since it's IP address has changed */
1488 afs_FlushServer(srv, tv);
1492 /* afs_GetCapabilities
1493 * Try and retrieve capabilities of a given file server. Carps on actual
1494 * failure. Servers are not expected to support this RPC. */
1496 afs_GetCapabilities(struct server *ts)
1498 Capabilities caps = {0, NULL};
1499 struct vrequest treq;
1500 struct afs_conn *tc;
1501 struct unixuser *tu;
1502 struct rx_connection *rxconn;
1505 if ( !ts || !ts->cell )
1507 if ( !afs_osi_credp )
1510 if ((code = afs_InitReq(&treq, afs_osi_credp)))
1512 tu = afs_GetUser(treq.uid, ts->cell->cellNum, SHARED_LOCK);
1515 tc = afs_ConnBySA(ts->addr, ts->cell->fsport, ts->cell->cellNum, tu, 0, 1,
1516 SHARED_LOCK, 0, &rxconn);
1517 afs_PutUser(tu, SHARED_LOCK);
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);
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! */
1536 ts->flags |= SCAPS_KNOWN;
1538 if ( caps.Capabilities_len > 0 ) {
1539 ts->capabilities = caps.Capabilities_val[0];
1540 xdr_free((xdrproc_t)xdr_Capabilities, &caps);
1541 caps.Capabilities_len = 0;
1542 caps.Capabilities_val = NULL;
1547 static struct server *
1548 afs_SearchServer(u_short aport, afsUUID * uuidp, afs_int32 locktype,
1549 struct server **oldts, afs_int32 addr_uniquifier)
1551 struct server *ts = afs_FindServer(0, aport, uuidp, locktype);
1552 if (ts && (ts->sr_addr_uniquifier == addr_uniquifier) && ts->addr) {
1553 /* Found a server struct that is multihomed and same
1554 * uniqufier (same IP addrs). The above if statement is the
1555 * same as in InstallUVolumeEntry().
1560 *oldts = ts; /* Will reuse if same uuid */
1565 * Return an updated and properly initialized server structure.
1567 * Takes a server ID, cell, and port.
1568 * If server does not exist, then one will be created.
1569 * @param[in] aserverp
1570 * The server address in network byte order
1571 * @param[in] nservers
1572 * The number of IP addresses claimed by the server
1574 * The cell the server is in
1576 * The port for the server (fileserver or vlserver) in network byte order
1577 * @param[in] locktype
1578 * The type of lock to hold when iterating server hash (unused).
1580 * The uuid for servers supporting one.
1581 * @param[in] addr_uniquifier
1582 * The vldb-provider per-instantiated-server uniquifer counter.
1584 * A volume not to reset information for if the server addresses
1588 * A server structure matching the request.
1591 afs_GetServer(afs_uint32 *aserverp, afs_int32 nservers, afs_int32 acell,
1592 u_short aport, afs_int32 locktype, afsUUID * uuidp,
1593 afs_int32 addr_uniquifier, struct volume *tv)
1595 struct server *oldts = 0, *ts, *newts, *orphts = 0;
1596 struct srvAddr *oldsa, *newsa, *nextsa, *orphsa;
1597 afs_int32 iphash, k, srvcount = 0;
1598 unsigned int srvhash;
1600 AFS_STATCNT(afs_GetServer);
1602 ObtainSharedLock(&afs_xserver, 13);
1604 /* Check if the server struct exists and is up to date */
1607 panic("afs_GetServer: incorrect count of servers");
1608 ObtainReadLock(&afs_xsrvAddr);
1609 ts = afs_FindServer(aserverp[0], aport, NULL, locktype);
1610 ReleaseReadLock(&afs_xsrvAddr);
1611 if (ts && !(ts->flags & SRVR_MULTIHOMED)) {
1612 /* Found a server struct that is not multihomed and has the
1613 * IP address associated with it. A correct match.
1615 ReleaseSharedLock(&afs_xserver);
1620 panic("afs_GetServer: incorrect count of servers");
1622 ts = afs_SearchServer(aport, uuidp, locktype, &oldts, addr_uniquifier);
1624 ReleaseSharedLock(&afs_xserver);
1630 * Lock hierarchy requires xvcb, then xserver. We *have* xserver.
1631 * Do a little dance and see if we can grab xvcb. If not, we
1632 * need to recheck that oldts is still right after a drop and reobtain.
1634 if (EWOULDBLOCK == NBObtainWriteLock(&afs_xvcb, 300)) {
1635 ReleaseSharedLock(&afs_xserver);
1636 ObtainWriteLock(&afs_xvcb, 299);
1637 ObtainWriteLock(&afs_xserver, 35);
1639 /* we don't know what changed while we didn't hold the lock */
1641 ts = afs_SearchServer(aport, uuidp, locktype, &oldts,
1644 ReleaseWriteLock(&afs_xserver);
1645 ReleaseWriteLock(&afs_xvcb);
1649 UpgradeSToWLock(&afs_xserver, 36);
1651 ObtainWriteLock(&afs_xsrvAddr, 116);
1652 srvcount = afs_totalServers;
1654 /* Reuse/allocate a new server structure */
1658 newts = afs_osi_Alloc(sizeof(struct server));
1660 panic("malloc of server struct");
1662 memset(newts, 0, sizeof(struct server));
1664 /* Add the server struct to the afs_servers[] hash chain */
1666 (uuidp ? (afs_uuid_hash(uuidp) % NSERVERS) : SHash(aserverp[0]));
1667 newts->next = afs_servers[srvhash];
1668 afs_servers[srvhash] = newts;
1671 /* Initialize the server structure */
1672 if (uuidp) { /* Multihomed */
1673 newts->sr_uuid = *uuidp;
1674 newts->sr_addr_uniquifier = addr_uniquifier;
1675 newts->flags |= SRVR_MULTIHOMED;
1678 newts->cell = afs_GetCell(acell, 0);
1680 /* For each IP address we are registering */
1681 for (k = 0; k < nservers; k++) {
1682 iphash = SHash(aserverp[k]);
1684 /* Check if the srvAddr structure already exists. If so, remove
1685 * it from its server structure and add it to the new one.
1687 for (oldsa = afs_srvAddrs[iphash]; oldsa; oldsa = oldsa->next_bkt) {
1688 if ((oldsa->sa_ip == aserverp[k]) && (oldsa->sa_portal == aport))
1691 if (oldsa && (oldsa->server != newts)) {
1692 afs_RemoveSrvAddr(oldsa, tv); /* Remove from its server struct */
1693 oldsa->next_sa = newts->addr; /* Add to the new server struct */
1694 newts->addr = oldsa;
1697 /* Reuse/allocate a new srvAddr structure */
1701 newsa = afs_osi_Alloc(sizeof(struct srvAddr));
1703 panic("malloc of srvAddr struct");
1704 afs_totalSrvAddrs++;
1705 memset(newsa, 0, sizeof(struct srvAddr));
1707 /* Add the new srvAddr to the afs_srvAddrs[] hash chain */
1708 newsa->next_bkt = afs_srvAddrs[iphash];
1709 afs_srvAddrs[iphash] = newsa;
1711 /* Hang off of the server structure */
1712 newsa->next_sa = newts->addr;
1713 newts->addr = newsa;
1715 /* Initialize the srvAddr Structure */
1716 newsa->sa_ip = aserverp[k];
1717 newsa->sa_portal = aport;
1720 /* Update the srvAddr Structure */
1721 newsa->server = newts;
1722 if (newts->flags & SRVR_ISDOWN)
1723 newsa->sa_flags |= SRVADDR_ISDOWN;
1725 newsa->sa_flags |= SRVADDR_MH;
1727 newsa->sa_flags &= ~SRVADDR_MH;
1729 /* Compute preference values and resort */
1730 if (!newsa->sa_iprank) {
1731 afs_SetServerPrefs(newsa); /* new server rank */
1734 afs_SortOneServer(newts); /* Sort by rank */
1736 /* If we reused the server struct, remove any of its srvAddr
1737 * structs that will no longer be associated with this server.
1739 if (oldts) { /* reused the server struct */
1740 for (orphsa = newts->addr; orphsa; orphsa = nextsa) {
1741 nextsa = orphsa->next_sa;
1742 for (k = 0; k < nservers; k++) {
1743 if (orphsa->sa_ip == aserverp[k])
1744 break; /* belongs */
1747 continue; /* belongs */
1749 /* Have a srvAddr struct. Now get a server struct (if not already) */
1751 orphts = afs_osi_Alloc(sizeof(struct server));
1753 panic("malloc of lo server struct");
1754 memset(orphts, 0, sizeof(struct server));
1757 /* Add the orphaned server to the afs_servers[] hash chain.
1758 * Its iphash does not matter since we never look up the server
1759 * in the afs_servers table by its ip address (only by uuid -
1760 * which this has none).
1762 iphash = SHash(aserverp[k]);
1763 orphts->next = afs_servers[iphash];
1764 afs_servers[iphash] = orphts;
1767 orphts->cell = afs_GetCell(acell, 0);
1770 /* Hang the srvAddr struct off of the server structure. The server
1771 * may have multiple srvAddrs, but it won't be marked multihomed.
1773 afs_RemoveSrvAddr(orphsa, tv); /* remove */
1774 orphsa->next_sa = orphts->addr; /* hang off server struct */
1775 orphts->addr = orphsa;
1776 orphsa->server = orphts;
1777 orphsa->sa_flags |= SRVADDR_NOUSE; /* flag indicating not in use */
1778 orphsa->sa_flags &= ~SRVADDR_MH; /* Not multihomed */
1781 /* We can't need this below, and won't reacquire */
1782 ReleaseWriteLock(&afs_xvcb);
1784 srvcount = afs_totalServers - srvcount; /* # servers added and removed */
1786 struct afs_stats_SrvUpDownInfo *upDownP;
1787 /* With the introduction of this new record, we need to adjust the
1788 * proper individual & global server up/down info.
1790 upDownP = GetUpDownStats(newts);
1791 upDownP->numTtlRecords += srvcount;
1792 afs_stats_cmperf.srvRecords += srvcount;
1793 if (afs_stats_cmperf.srvRecords > afs_stats_cmperf.srvRecordsHWM)
1794 afs_stats_cmperf.srvRecordsHWM = afs_stats_cmperf.srvRecords;
1796 /* We can't need this below, and won't reacquire */
1797 ReleaseWriteLock(&afs_xvcb);
1799 ReleaseWriteLock(&afs_xsrvAddr);
1801 if ( aport == AFS_FSPORT && !(newts->flags & SCAPS_KNOWN))
1802 afs_GetCapabilities(newts);
1804 ReleaseWriteLock(&afs_xserver);
1806 } /* afs_GetServer */
1809 afs_ActivateServer(struct srvAddr *sap)
1811 osi_timeval_t currTime; /*Filled with current time */
1812 osi_timeval_t *currTimeP; /*Ptr to above */
1813 struct afs_stats_SrvUpDownInfo *upDownP; /*Ptr to up/down info record */
1814 struct server *aserver = sap->server;
1816 if (!(aserver->flags & AFS_SERVER_FLAG_ACTIVATED)) {
1818 * This server record has not yet been activated. Go for it,
1819 * recording its ``birth''.
1821 aserver->flags |= AFS_SERVER_FLAG_ACTIVATED;
1822 currTimeP = &currTime;
1823 osi_GetuTime(currTimeP);
1824 aserver->activationTime = currTime.tv_sec;
1825 upDownP = GetUpDownStats(aserver);
1826 if (aserver->flags & SRVR_ISDOWN) {
1827 upDownP->numDownRecords++;
1829 upDownP->numUpRecords++;
1830 upDownP->numRecordsNeverDown++;
1836 afs_RemoveAllConns(void)
1839 struct server *ts, *nts;
1842 ObtainReadLock(&afs_xserver);
1843 ObtainWriteLock(&afs_xconn, 1001);
1845 /*printf("Destroying connections ... ");*/
1846 for (i = 0; i < NSERVERS; i++) {
1847 for (ts = afs_servers[i]; ts; ts = nts) {
1849 for (sa = ts->addr; sa; sa = sa->next_sa) {
1851 afs_ReleaseConns(sa->conns);
1858 /*printf("done\n");*/
1860 ReleaseWriteLock(&afs_xconn);
1861 ReleaseReadLock(&afs_xserver);
1866 afs_MarkAllServersUp(void)
1872 ObtainWriteLock(&afs_xserver, 721);
1873 ObtainWriteLock(&afs_xsrvAddr, 722);
1874 for (i = 0; i< NSERVERS; i++) {
1875 for (ts = afs_servers[i]; ts; ts = ts->next) {
1876 for (sa = ts->addr; sa; sa = sa->next_sa) {
1877 afs_MarkServerUpOrDown(sa, 0);
1881 ReleaseWriteLock(&afs_xsrvAddr);
1882 ReleaseWriteLock(&afs_xserver);