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 */
51 # ifdef AFS_DARWIN_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)
61 # include <inet/led.h>
62 # include <inet/common.h>
63 # include <netinet/ip6.h>
64 # define ipif_local_addr ipif_lcl_addr
65 # ifndef V4_PART_OF_V6
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_timeval32_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_GetTime(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_timeval32_t currTime; /*Current time */
358 osi_timeval32_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_GetTime(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 * Clear the low nybble of tv_usec in a size-independent manner before adding
830 * in the rest of the state.
833 state ^= (state & 0x0f);
834 state += rxi_getaddr() & 0xff;
835 state += (t.tv_sec & 0xff);
836 for (i = 0; i < 30; i++) {
846 /* returns int 0..14 using the high bits of a pseudo-random number instead of
847 the low bits, as the low bits are "less random" than the high ones...
848 slight roundoff error exists, an excercise for the reader.
849 need to multiply by something with lots of ones in it, so multiply by
850 8 or 16 is right out.
853 afs_randomMod15(void)
857 temp = afs_random() >> 4;
858 temp = (temp * 15) >> 28;
864 afs_randomMod127(void)
868 temp = afs_random() >> 7;
869 temp = (temp * 127) >> 25;
874 /* afs_SortOneServer()
875 * Sort all of the srvAddrs, of a server struct, by rank from low to high.
878 afs_SortOneServer(struct server *asp)
880 struct srvAddr **rootsa, *lowsa, *tsa, *lowprev;
883 for (rootsa = &(asp->addr); *rootsa; rootsa = &(lowsa->next_sa)) {
885 lowsa = *rootsa; /* lowest sa is the first one */
886 lowrank = lowsa->sa_iprank;
888 for (tsa = *rootsa; tsa->next_sa; tsa = tsa->next_sa) {
889 rank = tsa->next_sa->sa_iprank;
890 if (rank < lowrank) {
892 lowsa = tsa->next_sa;
893 lowrank = lowsa->sa_iprank;
896 if (lowprev) { /* found one lower, so rearrange them */
897 lowprev->next_sa = lowsa->next_sa;
898 lowsa->next_sa = *rootsa;
905 * Sort the pointer to servers by the server's rank (its lowest rank).
906 * It is assumed that the server already has its IP addrs sorted (the
907 * first being its lowest rank: afs_GetServer() calls afs_SortOneServer()).
910 afs_SortServers(struct server *aservers[], int count)
915 AFS_STATCNT(afs_SortServers);
917 for (i = 0; i < count; i++) {
920 for (low = i, j = i + 1; j <= count; j++) {
921 if ((!aservers[j]) || (!aservers[j]->addr))
923 if ((!aservers[low]) || (!aservers[low]->addr))
925 if (aservers[j]->addr->sa_iprank < aservers[low]->addr->sa_iprank) {
931 aservers[i] = aservers[low];
935 } /*afs_SortServers */
937 /* afs_SetServerPrefs is rather system-dependent. It pokes around in kernel
938 data structures to determine what the local IP addresses and subnet masks
939 are in order to choose which server(s) are on the local subnet.
941 As I see it, there are several cases:
942 1. The server address is one of this host's local addresses. In this case
943 this server is to be preferred over all others.
944 2. The server is on the same subnet as one of the this host's local
945 addresses. (ie, an odd-sized subnet, not class A,B,orC)
946 3. The server is on the same net as this host (class A,B or C)
947 4. The server is on a different logical subnet or net than this host, but
948 this host is a 'metric 0 gateway' to it. Ie, two address-spaces share
950 5. This host has a direct (point-to-point, ie, PPP or SLIP) link to the
952 6. This host and the server are disjoint.
954 That is a rough order of preference. If a point-to-point link has a high
955 metric, I'm assuming that it is a very slow link, and putting it at the
956 bottom of the list (at least until RX works better over slow links). If
957 its metric is 1, I'm assuming that it's relatively fast (T1) and putting
959 It's not easy to check for case #4, so I'm ignoring it for the time being.
961 BSD "if" code keeps track of some rough network statistics (cf 'netstat -i')
962 That could be used to prefer certain servers fairly easily. Maybe some
965 NOTE: this code is very system-dependent, and very dependent on the TCP/IP
966 protocols (well, addresses that are stored in uint32s, at any rate).
969 #define IA_DST(ia)((struct sockaddr_in *)(&((struct in_ifaddr *)ia)->ia_dstaddr))
970 #define IA_BROAD(ia)((struct sockaddr_in *)(&((struct in_ifaddr *)ia)->ia_broadaddr))
972 /* SA2ULONG takes a sockaddr_in, not a sockaddr (same thing, just cast it!) */
973 #define SA2ULONG(sa) ((sa)->sin_addr.s_addr)
978 #define PPWEIGHT 4096
982 #ifdef AFS_USERSPACE_IP_ADDR
984 # define afs_min(A,B) ((A)<(B)) ? (A) : (B)
987 * The IP addresses and ranks are determined by afsd (in user space) and
988 * passed into the kernel at startup time through the AFSOP_ADVISEADDR
989 * system call. These are stored in the data structure
990 * called 'afs_cb_interface'.
992 * struct srvAddr *sa; remote server
993 * afs_int32 addr; one of my local addr in net order
994 * afs_uint32 subnetmask; subnet mask of local addr in net order
998 afsi_SetServerIPRank(struct srvAddr *sa, afs_int32 addr,
999 afs_uint32 subnetmask)
1001 afs_uint32 myAddr, myNet, mySubnet, netMask;
1002 afs_uint32 serverAddr;
1004 myAddr = ntohl(addr); /* one of my IP addr in host order */
1005 serverAddr = ntohl(sa->sa_ip); /* server's IP addr in host order */
1006 subnetmask = ntohl(subnetmask); /* subnet mask in host order */
1008 if (IN_CLASSA(myAddr))
1009 netMask = IN_CLASSA_NET;
1010 else if (IN_CLASSB(myAddr))
1011 netMask = IN_CLASSB_NET;
1012 else if (IN_CLASSC(myAddr))
1013 netMask = IN_CLASSC_NET;
1017 myNet = myAddr & netMask;
1018 mySubnet = myAddr & subnetmask;
1020 if ((serverAddr & netMask) == myNet) {
1021 if ((serverAddr & subnetmask) == mySubnet) {
1022 if (serverAddr == myAddr) { /* same machine */
1023 sa->sa_iprank = afs_min(sa->sa_iprank, TOPR);
1024 } else { /* same subnet */
1025 sa->sa_iprank = afs_min(sa->sa_iprank, HI);
1027 } else { /* same net */
1028 sa->sa_iprank = afs_min(sa->sa_iprank, MED);
1034 #else /* AFS_USERSPACE_IP_ADDR */
1036 # if !defined(AFS_SUN5_ENV) && !defined(AFS_DARWIN_ENV) && !defined(AFS_OBSD47_ENV) && !defined(AFS_FBSD_ENV) && defined(USEIFADDR)
1038 afsi_SetServerIPRank(struct srvAddr *sa, struct in_ifaddr *ifa)
1040 struct sockaddr_in *sin;
1043 if ((ntohl(sa->sa_ip) & ifa->ia_netmask) == ifa->ia_net) {
1044 if ((ntohl(sa->sa_ip) & ifa->ia_subnetmask) == ifa->ia_subnet) {
1046 if (SA2ULONG(sin) == ntohl(sa->sa_ip)) { /* ie, ME!!! */
1047 sa->sa_iprank = TOPR;
1049 t = HI + ifa->ia_ifp->if_metric; /* case #2 */
1050 if (sa->sa_iprank > t)
1054 t = MED + ifa->ia_ifp->if_metric; /* case #3 */
1055 if (sa->sa_iprank > t)
1059 # if defined(IFF_POINTOPOINT) && !defined(UKERNEL)
1060 /* check for case #4 -- point-to-point link */
1061 if ((ifa->ia_ifp->if_flags & IFF_POINTOPOINT)
1062 && (SA2ULONG(IA_DST(ifa)) == ntohl(sa->sa_ip))) {
1063 if (ifa->ia_ifp->if_metric >= (MAXDEFRANK - MED) / PPWEIGHT)
1066 t = MED + (PPWEIGHT << ifa->ia_ifp->if_metric);
1067 if (sa->sa_iprank > t)
1070 # endif /* IFF_POINTOPOINT */
1072 # endif /* !SUN5 && !DARWIN && !OBSD47 && !FBSD && USEIFADDR */
1073 # if (defined(AFS_DARWIN_ENV) || defined(AFS_OBSD47_ENV) || defined(AFS_FBSD_ENV)) && defined(USEIFADDR)
1075 # define afs_min(A,B) ((A)<(B)) ? (A) : (B)
1078 afsi_SetServerIPRank(struct srvAddr *sa, rx_ifaddr_t ifa)
1080 struct sockaddr sout;
1081 struct sockaddr_in *sin;
1082 # if defined(AFS_DARWIN80_ENV) && !defined(UKERNEL)
1088 afs_uint32 subnetmask, myAddr, myNet, myDstaddr, mySubnet, netMask;
1089 afs_uint32 serverAddr;
1091 if (rx_ifaddr_address_family(ifa) != AF_INET)
1093 t = rx_ifaddr_address(ifa, &sout, sizeof(sout));
1095 sin = (struct sockaddr_in *)&sout;
1096 myAddr = ntohl(sin->sin_addr.s_addr); /* one of my IP addr in host order */
1100 serverAddr = ntohl(sa->sa_ip); /* server's IP addr in host order */
1101 t = rx_ifaddr_netmask(ifa, &sout, sizeof(sout));
1103 sin = (struct sockaddr_in *)&sout;
1104 subnetmask = ntohl(sin->sin_addr.s_addr); /* subnet mask in host order */
1108 t = rx_ifaddr_dstaddress(ifa, &sout, sizeof(sout));
1110 sin = (struct sockaddr_in *)&sout;
1111 myDstaddr = ntohl(sin->sin_addr.s_addr);
1116 if (IN_CLASSA(myAddr))
1117 netMask = IN_CLASSA_NET;
1118 else if (IN_CLASSB(myAddr))
1119 netMask = IN_CLASSB_NET;
1120 else if (IN_CLASSC(myAddr))
1121 netMask = IN_CLASSC_NET;
1125 myNet = myAddr & netMask;
1126 mySubnet = myAddr & subnetmask;
1128 if ((serverAddr & netMask) == myNet) {
1129 if ((serverAddr & subnetmask) == mySubnet) {
1130 if (serverAddr == myAddr) { /* same machine */
1131 sa->sa_iprank = afs_min(sa->sa_iprank, TOPR);
1132 } else { /* same subnet */
1133 sa->sa_iprank = afs_min(sa->sa_iprank, HI + rx_ifnet_metric(rx_ifaddr_ifnet(ifa)));
1135 } else { /* same net */
1136 sa->sa_iprank = afs_min(sa->sa_iprank, MED + rx_ifnet_metric(rx_ifaddr_ifnet(ifa)));
1139 # ifdef IFF_POINTTOPOINT
1140 /* check for case #4 -- point-to-point link */
1141 if ((rx_ifnet_flags(rx_ifaddr_ifnet(ifa)) & IFF_POINTOPOINT)
1142 && (myDstaddr == serverAddr)) {
1143 if (rx_ifnet_metric(rx_ifaddr_ifnet(ifa)) >= (MAXDEFRANK - MED) / PPWEIGHT)
1146 t = MED + (PPWEIGHT << rx_ifnet_metric(rx_ifaddr_ifnet(ifa)));
1147 if (sa->sa_iprank > t)
1150 # endif /* IFF_POINTTOPOINT */
1152 # endif /* (DARWIN || OBSD47 || FBSD) && USEIFADDR */
1153 #endif /* AFS_USERSPACE_IP_ADDR */
1155 #ifdef AFS_SGI62_ENV
1157 afsi_enum_set_rank(struct hashbucket *h, caddr_t mkey, caddr_t arg1,
1160 afsi_SetServerIPRank((struct srvAddr *)arg1, (struct in_ifaddr *)h);
1161 return 0; /* Never match, so we enumerate everyone */
1163 #endif /* AFS_SGI62_ENV */
1165 afs_SetServerPrefs(struct srvAddr *const sa)
1167 #if defined(AFS_USERSPACE_IP_ADDR)
1171 for (i = 0; i < afs_cb_interface.numberOfInterfaces; i++) {
1172 afsi_SetServerIPRank(sa, afs_cb_interface.addr_in[i],
1173 afs_cb_interface.subnetmask[i]);
1175 #else /* AFS_USERSPACE_IP_ADDR */
1176 # if defined(AFS_SUN5_ENV)
1177 # ifdef AFS_SUN510_ENV
1180 extern struct ill_s *ill_g_headp;
1181 long *addr = (long *)ill_g_headp;
1185 int subnet, subnetmask, net, netmask;
1188 # ifdef AFS_SUN510_ENV
1189 rw_enter(&afsifinfo_lock, RW_READER);
1191 for (i = 0; (afsifinfo[i].ipaddr != NULL) && (i < ADDRSPERSITE); i++) {
1193 if (IN_CLASSA(afsifinfo[i].ipaddr)) {
1194 netmask = IN_CLASSA_NET;
1195 } else if (IN_CLASSB(afsifinfo[i].ipaddr)) {
1196 netmask = IN_CLASSB_NET;
1197 } else if (IN_CLASSC(afsifinfo[i].ipaddr)) {
1198 netmask = IN_CLASSC_NET;
1202 net = afsifinfo[i].ipaddr & netmask;
1205 /* XXXXXX Do the individual ip ranking below XXXXX */
1206 if ((sa->sa_ip & netmask) == net) {
1207 if ((sa->sa_ip & subnetmask) == subnet) {
1208 if (afsifinfo[i].ipaddr == sa->sa_ip) { /* ie, ME! */
1209 sa->sa_iprank = TOPR;
1211 sa->sa_iprank = HI + afsifinfo[i].metric; /* case #2 */
1214 sa->sa_iprank = MED + afsifinfo[i].metric; /* case #3 */
1217 sa->sa_iprank = LO + afsifinfo[i].metric; /* case #4 */
1219 /* check for case #5 -- point-to-point link */
1220 if ((afsifinfo[i].flags & IFF_POINTOPOINT)
1221 && (afsifinfo[i].dstaddr == sa->sa_ip)) {
1223 if (afsifinfo[i].metric >= (MAXDEFRANK - MED) / PPWEIGHT)
1224 sa->sa_iprank = MAXDEFRANK;
1226 sa->sa_iprank = MED + (PPWEIGHT << afsifinfo[i].metric);
1231 rw_exit(&afsifinfo_lock);
1233 # else /* AFS_SUN510_ENV */
1235 for (ill = (struct ill_s *)*addr /*ill_g_headp */ ; ill;
1236 ill = ill->ill_next) {
1237 /* Make sure this is an IPv4 ILL */
1240 for (ipif = ill->ill_ipif; ipif; ipif = ipif->ipif_next) {
1241 subnet = ipif->ipif_local_addr & ipif->ipif_net_mask;
1242 subnetmask = ipif->ipif_net_mask;
1244 * Generate the local net using the local address and
1245 * whate we know about Class A, B and C networks.
1247 if (IN_CLASSA(ipif->ipif_local_addr)) {
1248 netmask = IN_CLASSA_NET;
1249 } else if (IN_CLASSB(ipif->ipif_local_addr)) {
1250 netmask = IN_CLASSB_NET;
1251 } else if (IN_CLASSC(ipif->ipif_local_addr)) {
1252 netmask = IN_CLASSC_NET;
1256 net = ipif->ipif_local_addr & netmask;
1258 /* XXXXXX Do the individual ip ranking below XXXXX */
1259 if ((sa->sa_ip & netmask) == net) {
1260 if ((sa->sa_ip & subnetmask) == subnet) {
1261 if (ipif->ipif_local_addr == sa->sa_ip) { /* ie, ME! */
1262 sa->sa_iprank = TOPR;
1264 sa->sa_iprank = HI + ipif->ipif_metric; /* case #2 */
1267 sa->sa_iprank = MED + ipif->ipif_metric; /* case #3 */
1270 sa->sa_iprank = LO + ipif->ipif_metric; /* case #4 */
1272 /* check for case #5 -- point-to-point link */
1273 if ((ipif->ipif_flags & IFF_POINTOPOINT)
1274 && (ipif->ipif_pp_dst_addr == sa->sa_ip)) {
1276 if (ipif->ipif_metric >= (MAXDEFRANK - MED) / PPWEIGHT)
1277 sa->sa_iprank = MAXDEFRANK;
1279 sa->sa_iprank = MED + (PPWEIGHT << ipif->ipif_metric);
1284 # endif /* AFS_SUN510_ENV */
1285 # else /* AFS_SUN5_ENV */
1287 rx_ifnet_t ifn = NULL;
1288 struct in_ifaddr *ifad = (struct in_ifaddr *)0;
1289 struct sockaddr_in *sin;
1292 ifn = rxi_FindIfnet(sa->sa_ip, &ifad);
1293 if (ifn) { /* local, more or less */
1294 # ifdef IFF_LOOPBACK
1295 if (ifn->if_flags & IFF_LOOPBACK) {
1296 sa->sa_iprank = TOPR;
1299 # endif /* IFF_LOOPBACK */
1300 sin = (struct sockaddr_in *)IA_SIN(ifad);
1301 if (SA2ULONG(sin) == sa->sa_ip) {
1302 sa->sa_iprank = TOPR;
1305 # ifdef IFF_BROADCAST
1306 if (ifn->if_flags & IFF_BROADCAST) {
1307 if (sa->sa_ip == (sa->sa_ip & SA2ULONG(IA_BROAD(ifad)))) {
1312 # endif /* IFF_BROADCAST */
1313 # ifdef IFF_POINTOPOINT
1314 if (ifn->if_flags & IFF_POINTOPOINT) {
1315 if (sa->sa_ip == SA2ULONG(IA_DST(ifad))) {
1316 if (ifn->if_metric > 4) {
1320 sa->sa_iprank = ifn->if_metric;
1323 # endif /* IFF_POINTOPOINT */
1324 sa->sa_iprank += MED + ifn->if_metric; /* couldn't find anything better */
1326 # else /* USEIFADDR */
1329 # ifdef AFS_SGI62_ENV
1330 (void)hash_enum(&hashinfo_inaddr, afsi_enum_set_rank, HTF_INET, NULL,
1331 (caddr_t) sa, NULL);
1332 # elif defined(AFS_DARWIN80_ENV)
1340 if (!ifnet_list_get(AF_INET, &ifns, &count)) {
1341 for (m = 0; m < count; m++) {
1342 if (!ifnet_get_address_list(ifns[m], &ifads)) {
1343 for (j = 0; ifads[j] != NULL && cnt < ADDRSPERSITE; j++) {
1344 afsi_SetServerIPRank(sa, ifads[j]);
1347 ifnet_free_address_list(ifads);
1350 ifnet_list_free(ifns);
1353 # elif defined(AFS_DARWIN_ENV)
1357 TAILQ_FOREACH(ifn, &ifnet, if_link) {
1358 TAILQ_FOREACH(ifa, &ifn->if_addrhead, ifa_link) {
1359 afsi_SetServerIPRank(sa, ifa);
1361 # elif defined(AFS_FBSD_ENV)
1363 struct in_ifaddr *ifa;
1364 CURVNET_SET(rx_socket->so_vnet);
1365 TAILQ_FOREACH(ifa, &V_in_ifaddrhead, ia_link) {
1366 afsi_SetServerIPRank(sa, &ifa->ia_ifa);
1370 # elif defined(AFS_OBSD_ENV)
1372 extern struct in_ifaddrhead in_ifaddr;
1373 struct in_ifaddr *ifa;
1374 for (ifa = in_ifaddr.tqh_first; ifa; ifa = ifa->ia_list.tqe_next)
1375 afsi_SetServerIPRank(sa, ifa);
1377 # elif defined(AFS_NBSD40_ENV)
1379 extern struct in_ifaddrhead in_ifaddrhead;
1380 struct in_ifaddr *ifa;
1381 for (ifa = in_ifaddrhead.tqh_first; ifa; ifa = ifa->ia_list.tqe_next)
1382 afsi_SetServerIPRank(sa, ifa);
1386 struct in_ifaddr *ifa;
1387 for (ifa = in_ifaddr; ifa; ifa = ifa->ia_next) {
1388 afsi_SetServerIPRank(sa, ifa);
1391 # endif /* USEIFADDR */
1395 # endif /* AFS_SUN5_ENV */
1396 #endif /* AFS_USERSPACE_IP_ADDR */
1397 sa->sa_iprank += afs_randomMod15();
1400 } /* afs_SetServerPrefs */
1408 /* afs_FlushServer()
1409 * The addresses on this server struct has changed in some way and will
1410 * clean up all other structures that may reference it.
1411 * The afs_xserver, afs_xvcb and afs_xsrvAddr locks are assumed taken.
1414 afs_FlushServer(struct server *srvp, struct volume *tv)
1417 struct server *ts, **pts;
1419 /* Find any volumes residing on this server and flush their state */
1420 afs_ResetVolumes(srvp, tv);
1422 /* Flush all callbacks in the all vcaches for this specific server */
1423 afs_FlushServerCBs(srvp);
1425 /* Remove all the callbacks structs */
1427 struct afs_cbr *cb, *cbnext;
1429 for (cb = srvp->cbrs; cb; cb = cbnext) {
1432 } srvp->cbrs = (struct afs_cbr *)0;
1435 /* If no more srvAddr structs hanging off of this server struct,
1439 /* Remove the server structure from the cell list - if there */
1440 afs_RemoveCellEntry(srvp);
1442 /* Remove from the afs_servers hash chain */
1443 for (i = 0; i < NSERVERS; i++) {
1444 for (pts = &(afs_servers[i]), ts = *pts; ts;
1445 pts = &(ts->next), ts = *pts) {
1453 *pts = ts->next; /* Found it. Remove it */
1454 afs_osi_Free(ts, sizeof(struct server)); /* Free it */
1460 /* afs_RemoveSrvAddr()
1461 * This removes a SrvAddr structure from its server structure.
1462 * The srvAddr struct is not free'd because it connections may still
1463 * be open to it. It is up to the calling process to make sure it
1464 * remains connected to a server struct.
1465 * The afs_xserver and afs_xsrvAddr locks are assumed taken.
1466 * It is not removed from the afs_srvAddrs hash chain.
1467 * If resetting volumes, do not reset volume tv
1470 afs_RemoveSrvAddr(struct srvAddr *sap, struct volume *tv)
1472 struct srvAddr **psa, *sa;
1479 /* Find the srvAddr in the server's list and remove it */
1480 for (psa = &(srv->addr), sa = *psa; sa; psa = &(sa->next_sa), sa = *psa) {
1488 /* Flush the server struct since it's IP address has changed */
1489 afs_FlushServer(srv, tv);
1493 /* afs_GetCapabilities
1494 * Try and retrieve capabilities of a given file server. Carps on actual
1495 * failure. Servers are not expected to support this RPC. */
1497 afs_GetCapabilities(struct server *ts)
1499 Capabilities caps = {0, NULL};
1500 struct vrequest *treq = NULL;
1501 struct afs_conn *tc;
1502 struct unixuser *tu;
1503 struct rx_connection *rxconn;
1506 if ( !ts || !ts->cell )
1508 if ( !afs_osi_credp )
1511 if ((code = afs_CreateReq(&treq, afs_osi_credp)))
1513 tu = afs_GetUser(treq->uid, ts->cell->cellNum, SHARED_LOCK);
1515 afs_DestroyReq(treq);
1518 tc = afs_ConnBySA(ts->addr, ts->cell->fsport, ts->cell->cellNum, tu, 0, 1,
1519 SHARED_LOCK, 0, &rxconn);
1520 afs_PutUser(tu, SHARED_LOCK);
1522 afs_DestroyReq(treq);
1525 /* InitCallBackStateN, triggered by our RPC, may need this */
1526 ReleaseWriteLock(&afs_xserver);
1528 code = RXAFS_GetCapabilities(rxconn, &caps);
1530 ObtainWriteLock(&afs_xserver, 723);
1531 /* we forced a conn above; important we mark it down if needed */
1532 if ((code < 0) && (code != RXGEN_OPCODE)) {
1533 afs_ServerDown(tc->parent->srvr, code, rxconn);
1534 ForceNewConnections(tc->parent->srvr); /* multi homed clients */
1536 afs_PutConn(tc, rxconn, SHARED_LOCK);
1537 if ( code && code != RXGEN_OPCODE ) {
1538 afs_warn("RXAFS_GetCapabilities failed with code %d\n", code);
1539 /* better not be anything to free. we failed! */
1540 afs_DestroyReq(treq);
1544 ts->flags |= SCAPS_KNOWN;
1546 if ( caps.Capabilities_len > 0 ) {
1547 ts->capabilities = caps.Capabilities_val[0];
1548 xdr_free((xdrproc_t)xdr_Capabilities, &caps);
1549 caps.Capabilities_len = 0;
1550 caps.Capabilities_val = NULL;
1553 afs_DestroyReq(treq);
1556 static struct server *
1557 afs_SearchServer(u_short aport, afsUUID * uuidp, afs_int32 locktype,
1558 struct server **oldts, afs_int32 addr_uniquifier)
1560 struct server *ts = afs_FindServer(0, aport, uuidp, locktype);
1561 if (ts && (ts->sr_addr_uniquifier == addr_uniquifier) && ts->addr) {
1562 /* Found a server struct that is multihomed and same
1563 * uniqufier (same IP addrs). The above if statement is the
1564 * same as in InstallUVolumeEntry().
1569 *oldts = ts; /* Will reuse if same uuid */
1574 * Return an updated and properly initialized server structure.
1576 * Takes a server ID, cell, and port.
1577 * If server does not exist, then one will be created.
1578 * @param[in] aserverp
1579 * The server address in network byte order
1580 * @param[in] nservers
1581 * The number of IP addresses claimed by the server
1583 * The cell the server is in
1585 * The port for the server (fileserver or vlserver) in network byte order
1586 * @param[in] locktype
1587 * The type of lock to hold when iterating server hash (unused).
1589 * The uuid for servers supporting one.
1590 * @param[in] addr_uniquifier
1591 * The vldb-provider per-instantiated-server uniquifer counter.
1593 * A volume not to reset information for if the server addresses
1597 * A server structure matching the request.
1600 afs_GetServer(afs_uint32 *aserverp, afs_int32 nservers, afs_int32 acell,
1601 u_short aport, afs_int32 locktype, afsUUID * uuidp,
1602 afs_int32 addr_uniquifier, struct volume *tv)
1604 struct server *oldts = 0, *ts, *newts, *orphts = 0;
1605 struct srvAddr *oldsa, *newsa, *nextsa, *orphsa;
1606 afs_int32 iphash, k, srvcount = 0;
1607 unsigned int srvhash;
1609 AFS_STATCNT(afs_GetServer);
1611 ObtainSharedLock(&afs_xserver, 13);
1613 /* Check if the server struct exists and is up to date */
1616 panic("afs_GetServer: incorrect count of servers");
1617 ObtainReadLock(&afs_xsrvAddr);
1618 ts = afs_FindServer(aserverp[0], aport, NULL, locktype);
1619 ReleaseReadLock(&afs_xsrvAddr);
1620 if (ts && !(ts->flags & SRVR_MULTIHOMED)) {
1621 /* Found a server struct that is not multihomed and has the
1622 * IP address associated with it. A correct match.
1624 ReleaseSharedLock(&afs_xserver);
1629 panic("afs_GetServer: incorrect count of servers");
1631 ts = afs_SearchServer(aport, uuidp, locktype, &oldts, addr_uniquifier);
1633 ReleaseSharedLock(&afs_xserver);
1639 * Lock hierarchy requires xvcb, then xserver. We *have* xserver.
1640 * Do a little dance and see if we can grab xvcb. If not, we
1641 * need to recheck that oldts is still right after a drop and reobtain.
1643 if (EWOULDBLOCK == NBObtainWriteLock(&afs_xvcb, 300)) {
1644 ReleaseSharedLock(&afs_xserver);
1645 ObtainWriteLock(&afs_xvcb, 299);
1646 ObtainWriteLock(&afs_xserver, 35);
1648 /* we don't know what changed while we didn't hold the lock */
1650 ts = afs_SearchServer(aport, uuidp, locktype, &oldts,
1653 ReleaseWriteLock(&afs_xserver);
1654 ReleaseWriteLock(&afs_xvcb);
1658 UpgradeSToWLock(&afs_xserver, 36);
1660 ObtainWriteLock(&afs_xsrvAddr, 116);
1661 srvcount = afs_totalServers;
1663 /* Reuse/allocate a new server structure */
1667 newts = afs_osi_Alloc(sizeof(struct server));
1669 panic("malloc of server struct");
1671 memset(newts, 0, sizeof(struct server));
1673 /* Add the server struct to the afs_servers[] hash chain */
1675 (uuidp ? (afs_uuid_hash(uuidp) % NSERVERS) : SHash(aserverp[0]));
1676 newts->next = afs_servers[srvhash];
1677 afs_servers[srvhash] = newts;
1680 /* Initialize the server structure */
1681 if (uuidp) { /* Multihomed */
1682 newts->sr_uuid = *uuidp;
1683 newts->sr_addr_uniquifier = addr_uniquifier;
1684 newts->flags |= SRVR_MULTIHOMED;
1687 /* Use the afs_GetCellStale variant to avoid afs_GetServer recursion. */
1688 newts->cell = afs_GetCellStale(acell, 0);
1690 /* For each IP address we are registering */
1691 for (k = 0; k < nservers; k++) {
1692 iphash = SHash(aserverp[k]);
1694 /* Check if the srvAddr structure already exists. If so, remove
1695 * it from its server structure and add it to the new one.
1697 for (oldsa = afs_srvAddrs[iphash]; oldsa; oldsa = oldsa->next_bkt) {
1698 if ((oldsa->sa_ip == aserverp[k]) && (oldsa->sa_portal == aport))
1701 if (oldsa && (oldsa->server != newts)) {
1702 afs_RemoveSrvAddr(oldsa, tv); /* Remove from its server struct */
1703 oldsa->next_sa = newts->addr; /* Add to the new server struct */
1704 newts->addr = oldsa;
1707 /* Reuse/allocate a new srvAddr structure */
1711 newsa = afs_osi_Alloc(sizeof(struct srvAddr));
1713 panic("malloc of srvAddr struct");
1714 afs_totalSrvAddrs++;
1715 memset(newsa, 0, sizeof(struct srvAddr));
1717 /* Add the new srvAddr to the afs_srvAddrs[] hash chain */
1718 newsa->next_bkt = afs_srvAddrs[iphash];
1719 afs_srvAddrs[iphash] = newsa;
1721 /* Hang off of the server structure */
1722 newsa->next_sa = newts->addr;
1723 newts->addr = newsa;
1725 /* Initialize the srvAddr Structure */
1726 newsa->sa_ip = aserverp[k];
1727 newsa->sa_portal = aport;
1730 /* Update the srvAddr Structure */
1731 newsa->server = newts;
1732 if (newts->flags & SRVR_ISDOWN)
1733 newsa->sa_flags |= SRVADDR_ISDOWN;
1735 newsa->sa_flags |= SRVADDR_MH;
1737 newsa->sa_flags &= ~SRVADDR_MH;
1739 /* Compute preference values and resort */
1740 if (!newsa->sa_iprank) {
1741 afs_SetServerPrefs(newsa); /* new server rank */
1744 afs_SortOneServer(newts); /* Sort by rank */
1746 /* If we reused the server struct, remove any of its srvAddr
1747 * structs that will no longer be associated with this server.
1749 if (oldts) { /* reused the server struct */
1750 for (orphsa = newts->addr; orphsa; orphsa = nextsa) {
1751 nextsa = orphsa->next_sa;
1752 for (k = 0; k < nservers; k++) {
1753 if (orphsa->sa_ip == aserverp[k])
1754 break; /* belongs */
1757 continue; /* belongs */
1759 /* Have a srvAddr struct. Now get a server struct (if not already) */
1761 orphts = afs_osi_Alloc(sizeof(struct server));
1763 panic("malloc of lo server struct");
1764 memset(orphts, 0, sizeof(struct server));
1767 /* Add the orphaned server to the afs_servers[] hash chain.
1768 * Its iphash does not matter since we never look up the server
1769 * in the afs_servers table by its ip address (only by uuid -
1770 * which this has none).
1772 iphash = SHash(aserverp[k]);
1773 orphts->next = afs_servers[iphash];
1774 afs_servers[iphash] = orphts;
1777 /* Use the afs_GetCellStale variant to avoid afs_GetServer recursion. */
1778 orphts->cell = afs_GetCellStale(acell, 0);
1781 /* Hang the srvAddr struct off of the server structure. The server
1782 * may have multiple srvAddrs, but it won't be marked multihomed.
1784 afs_RemoveSrvAddr(orphsa, tv); /* remove */
1785 orphsa->next_sa = orphts->addr; /* hang off server struct */
1786 orphts->addr = orphsa;
1787 orphsa->server = orphts;
1788 orphsa->sa_flags |= SRVADDR_NOUSE; /* flag indicating not in use */
1789 orphsa->sa_flags &= ~SRVADDR_MH; /* Not multihomed */
1792 /* We can't need this below, and won't reacquire */
1793 ReleaseWriteLock(&afs_xvcb);
1795 srvcount = afs_totalServers - srvcount; /* # servers added and removed */
1797 struct afs_stats_SrvUpDownInfo *upDownP;
1798 /* With the introduction of this new record, we need to adjust the
1799 * proper individual & global server up/down info.
1801 upDownP = GetUpDownStats(newts);
1802 upDownP->numTtlRecords += srvcount;
1803 afs_stats_cmperf.srvRecords += srvcount;
1804 if (afs_stats_cmperf.srvRecords > afs_stats_cmperf.srvRecordsHWM)
1805 afs_stats_cmperf.srvRecordsHWM = afs_stats_cmperf.srvRecords;
1808 ReleaseWriteLock(&afs_xsrvAddr);
1810 if ( aport == AFS_FSPORT && !(newts->flags & SCAPS_KNOWN))
1811 afs_GetCapabilities(newts);
1813 ReleaseWriteLock(&afs_xserver);
1815 } /* afs_GetServer */
1818 afs_ActivateServer(struct srvAddr *sap)
1820 osi_timeval32_t currTime; /*Filled with current time */
1821 osi_timeval32_t *currTimeP; /*Ptr to above */
1822 struct afs_stats_SrvUpDownInfo *upDownP; /*Ptr to up/down info record */
1823 struct server *aserver = sap->server;
1825 if (!(aserver->flags & AFS_SERVER_FLAG_ACTIVATED)) {
1827 * This server record has not yet been activated. Go for it,
1828 * recording its ``birth''.
1830 aserver->flags |= AFS_SERVER_FLAG_ACTIVATED;
1831 currTimeP = &currTime;
1832 osi_GetTime(currTimeP);
1833 aserver->activationTime = currTime.tv_sec;
1834 upDownP = GetUpDownStats(aserver);
1835 if (aserver->flags & SRVR_ISDOWN) {
1836 upDownP->numDownRecords++;
1838 upDownP->numUpRecords++;
1839 upDownP->numRecordsNeverDown++;
1845 afs_RemoveAllConns(void)
1848 struct server *ts, *nts;
1851 ObtainReadLock(&afs_xserver);
1852 ObtainWriteLock(&afs_xconn, 1001);
1854 /*printf("Destroying connections ... ");*/
1855 for (i = 0; i < NSERVERS; i++) {
1856 for (ts = afs_servers[i]; ts; ts = nts) {
1858 for (sa = ts->addr; sa; sa = sa->next_sa) {
1860 afs_ReleaseConns(sa->conns);
1867 /*printf("done\n");*/
1869 ReleaseWriteLock(&afs_xconn);
1870 ReleaseReadLock(&afs_xserver);
1875 afs_MarkAllServersUp(void)
1881 ObtainWriteLock(&afs_xserver, 721);
1882 ObtainWriteLock(&afs_xsrvAddr, 722);
1883 for (i = 0; i< NSERVERS; i++) {
1884 for (ts = afs_servers[i]; ts; ts = ts->next) {
1885 for (sa = ts->addr; sa; sa = sa->next_sa) {
1886 afs_MarkServerUpOrDown(sa, 0);
1890 ReleaseWriteLock(&afs_xsrvAddr);
1891 ReleaseWriteLock(&afs_xserver);