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"
39 #include "afs/sysincludes.h" /* Standard vendor system headers */
42 #if !defined(AFS_LINUX20_ENV)
45 #include <netinet/in.h>
48 #include "h/hashing.h"
50 #if !defined(AFS_HPUX110_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_DARWIN60_ENV)
51 #include <netinet/in_var.h>
52 #endif /* AFS_HPUX110_ENV */
53 #ifdef AFS_DARWIN60_ENV
54 #include <net/if_var.h>
56 #endif /* !defined(UKERNEL) */
58 #include "afsincludes.h" /* Afs-based standard headers */
59 #include "afs/afs_stats.h" /* afs statistics */
60 #include "rx/rx_multi.h"
62 #if defined(AFS_SUN56_ENV)
64 #include <inet/common.h>
65 #if defined(AFS_SUN58_ENV)
66 # include <netinet/ip6.h>
67 # define ipif_local_addr ipif_lcl_addr
68 # ifndef V4_PART_OF_V6
69 # define V4_PART_OF_V6(v6) v6.s6_addr32[3]
75 /* Exported variables */
76 afs_rwlock_t afs_xserver; /* allocation lock for servers */
77 struct server *afs_setTimeHost = 0; /* last host we used for time */
78 struct server *afs_servers[NSERVERS]; /* Hashed by server`s uuid & 1st ip */
79 afs_rwlock_t afs_xsrvAddr; /* allocation lock for srvAddrs */
80 struct srvAddr *afs_srvAddrs[NSERVERS]; /* Hashed by server's ip */
83 /* debugging aids - number of alloc'd server and srvAddr structs. */
84 int afs_reuseServers = 0;
85 int afs_reuseSrvAddrs = 0;
86 int afs_totalServers = 0;
87 int afs_totalSrvAddrs = 0;
91 static struct afs_stats_SrvUpDownInfo *
92 GetUpDownStats(struct server *srv)
94 struct afs_stats_SrvUpDownInfo *upDownP;
95 u_short fsport = AFS_FSPORT;
98 fsport = srv->cell->fsport;
100 if (srv->addr->sa_portal == fsport)
101 upDownP = afs_stats_cmperf.fs_UpDown;
103 upDownP = afs_stats_cmperf.vl_UpDown;
105 if (srv->cell && afs_IsPrimaryCell(srv->cell))
106 return &upDownP[AFS_STATS_UPDOWN_IDX_SAME_CELL];
108 return &upDownP[AFS_STATS_UPDOWN_IDX_DIFF_CELL];
112 /*------------------------------------------------------------------------
113 * afs_MarkServerUpOrDown
116 * Mark the given server up or down, and track its uptime stats.
119 * a_serverP : Ptr to server record to fiddle with.
120 * a_isDown : Is the server is to be marked down?
126 * The CM server structures must be write-locked.
130 *------------------------------------------------------------------------*/
133 afs_MarkServerUpOrDown(struct srvAddr *sa, int a_isDown)
135 register struct server *a_serverP = sa->server;
136 register struct srvAddr *sap;
137 osi_timeval_t currTime, *currTimeP; /*Current time */
138 afs_int32 downTime; /*Computed downtime, in seconds */
139 struct afs_stats_SrvUpDownInfo *upDownP; /*Ptr to up/down info record */
142 * If the server record is marked the same as the new status we've
143 * been fed, then there isn't much to be done.
145 if ((a_isDown && (sa->sa_flags & SRVADDR_ISDOWN))
146 || (!a_isDown && !(sa->sa_flags & SRVADDR_ISDOWN)))
150 sa->sa_flags |= SRVADDR_ISDOWN;
151 for (sap = a_serverP->addr; sap; sap = sap->next_sa) {
152 if (!(sap->sa_flags & SRVADDR_ISDOWN)) {
153 /* Not all ips are up so don't bother with the
154 * server's up/down stats */
159 * All ips are down we treat the whole server down
161 a_serverP->flags |= SRVR_ISDOWN;
163 * If this was our time server, search for another time server
165 if (a_serverP == afs_setTimeHost)
168 sa->sa_flags &= ~SRVADDR_ISDOWN;
169 /* If any ips are up, the server is also marked up */
170 a_serverP->flags &= ~SRVR_ISDOWN;
171 for (sap = a_serverP->addr; sap; sap = sap->next_sa) {
172 if (sap->sa_flags & SRVADDR_ISDOWN) {
173 /* Not all ips are up so don't bother with the
174 * server's up/down stats */
181 * Compute the current time and which overall stats record is to be
182 * updated; we'll need them one way or another.
184 currTimeP = &currTime;
185 osi_GetuTime(currTimeP);
187 upDownP = GetUpDownStats(a_serverP);
191 * Server going up -> down; remember the beginning of this
194 a_serverP->lastDowntimeStart = currTime.tv_sec;
196 (upDownP->numDownRecords)++;
197 (upDownP->numUpRecords)--;
198 } /*Server being marked down */
201 * Server going down -> up; remember everything about this
202 * newly-completed downtime incident.
204 downTime = currTime.tv_sec - a_serverP->lastDowntimeStart;
205 (a_serverP->numDowntimeIncidents)++;
206 a_serverP->sumOfDowntimes += downTime;
208 (upDownP->numUpRecords)++;
209 (upDownP->numDownRecords)--;
210 (upDownP->numDowntimeIncidents)++;
211 if (a_serverP->numDowntimeIncidents == 1)
212 (upDownP->numRecordsNeverDown)--;
213 upDownP->sumOfDowntimes += downTime;
214 if ((upDownP->shortestDowntime == 0)
215 || (downTime < upDownP->shortestDowntime))
216 upDownP->shortestDowntime = downTime;
217 if ((upDownP->longestDowntime == 0)
218 || (downTime > upDownP->longestDowntime))
219 upDownP->longestDowntime = downTime;
222 if (downTime <= AFS_STATS_MAX_DOWNTIME_DURATION_BUCKET0)
223 (upDownP->downDurations[0])++;
224 else if (downTime <= AFS_STATS_MAX_DOWNTIME_DURATION_BUCKET1)
225 (upDownP->downDurations[1])++;
226 else if (downTime <= AFS_STATS_MAX_DOWNTIME_DURATION_BUCKET2)
227 (upDownP->downDurations[2])++;
228 else if (downTime <= AFS_STATS_MAX_DOWNTIME_DURATION_BUCKET3)
229 (upDownP->downDurations[3])++;
230 else if (downTime <= AFS_STATS_MAX_DOWNTIME_DURATION_BUCKET4)
231 (upDownP->downDurations[4])++;
232 else if (downTime <= AFS_STATS_MAX_DOWNTIME_DURATION_BUCKET5)
233 (upDownP->downDurations[5])++;
235 (upDownP->downDurations[6])++;
237 } /*Server being marked up */
239 } /*MarkServerUpOrDown */
243 afs_ServerDown(struct srvAddr *sa)
245 register struct server *aserver = sa->server;
247 AFS_STATCNT(ServerDown);
248 if (aserver->flags & SRVR_ISDOWN || sa->sa_flags & SRVADDR_ISDOWN)
250 afs_MarkServerUpOrDown(sa, SRVR_ISDOWN);
251 if (sa->sa_portal == aserver->cell->vlport)
252 print_internet_address
253 ("afs: Lost contact with volume location server ", sa, "", 1);
255 print_internet_address("afs: Lost contact with file server ", sa, "",
261 /* return true if we have any callback promises from this server */
263 afs_HaveCallBacksFrom(struct server *aserver)
265 register afs_int32 now;
267 register struct vcache *tvc;
269 AFS_STATCNT(HaveCallBacksFrom);
270 now = osi_Time(); /* for checking for expired callbacks */
271 for (i = 0; i < VCSIZE; i++) { /* for all guys in the hash table */
272 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
274 * Check to see if this entry has an unexpired callback promise
275 * from the required host
277 if (aserver == tvc->callback && tvc->cbExpires >= now
278 && ((tvc->states & CRO) == 0))
284 } /*HaveCallBacksFrom */
288 CheckVLServer(register struct srvAddr *sa, struct vrequest *areq)
290 register struct server *aserver = sa->server;
291 register struct conn *tc;
292 register afs_int32 code;
294 AFS_STATCNT(CheckVLServer);
295 /* Ping dead servers to see if they're back */
296 if (!((aserver->flags & SRVR_ISDOWN) || (sa->sa_flags & SRVADDR_ISDOWN))
297 || (aserver->flags & SRVR_ISGONE))
300 return; /* can't do much */
302 tc = afs_ConnByHost(aserver, aserver->cell->vlport,
303 aserver->cell->cellNum, areq, 1, SHARED_LOCK);
306 rx_SetConnDeadTime(tc->id, 3);
309 code = VL_ProbeServer(tc->id);
311 rx_SetConnDeadTime(tc->id, afs_rx_deadtime);
312 afs_PutConn(tc, SHARED_LOCK);
314 * If probe worked, or probe call not yet defined (for compatibility
315 * with old vlsevers), then we treat this server as running again
317 if (code == 0 || (code <= -450 && code >= -470)) {
318 if (tc->srvr == sa) {
319 afs_MarkServerUpOrDown(sa, 0);
320 print_internet_address("afs: volume location server ", sa,
328 #ifndef AFS_MINCHANGE /* So that some can increase it in param.h */
329 #define AFS_MINCHANGE 2 /* min change we'll bother with */
331 #ifndef AFS_MAXCHANGEBACK
332 #define AFS_MAXCHANGEBACK 10 /* max seconds we'll set a clock back at once */
336 /*------------------------------------------------------------------------
337 * EXPORTED afs_CountServers
340 * Originally meant to count the number of servers and determining
341 * up/down info, this routine will now simply sum up all of the
342 * server record ages. All other up/down information is kept on the
352 * This routine locks afs_xserver for write for the duration.
355 * Set CM perf stats field sumOfRecordAges for all server record
357 *------------------------------------------------------------------------*/
360 afs_CountServers(void)
362 int currIdx; /*Curr idx into srv table */
363 struct server *currSrvP; /*Ptr to curr server record */
364 afs_int32 currChainLen; /*Length of curr hash chain */
365 osi_timeval_t currTime; /*Current time */
366 osi_timeval_t *currTimeP; /*Ptr to above */
367 afs_int32 srvRecordAge; /*Age of server record, in secs */
368 struct afs_stats_SrvUpDownInfo *upDownP; /*Ptr to current up/down
369 * info being manipulated */
372 * Write-lock the server table so we don't get any interference.
374 ObtainReadLock(&afs_xserver);
377 * Iterate over each hash index in the server table, walking down each
378 * chain and tallying what we haven't computed from the records there on
379 * the fly. First, though, initialize the tallies that will change.
381 afs_stats_cmperf.srvMaxChainLength = 0;
383 afs_stats_cmperf.fs_UpDown[0].sumOfRecordAges = 0;
384 afs_stats_cmperf.fs_UpDown[0].ageOfYoungestRecord = 0;
385 afs_stats_cmperf.fs_UpDown[0].ageOfOldestRecord = 0;
386 memset((char *)afs_stats_cmperf.fs_UpDown[0].downIncidents, 0,
387 AFS_STATS_NUM_DOWNTIME_INCIDENTS_BUCKETS * sizeof(afs_int32));
389 afs_stats_cmperf.fs_UpDown[1].sumOfRecordAges = 0;
390 afs_stats_cmperf.fs_UpDown[1].ageOfYoungestRecord = 0;
391 afs_stats_cmperf.fs_UpDown[1].ageOfOldestRecord = 0;
392 memset((char *)afs_stats_cmperf.fs_UpDown[1].downIncidents, 0,
393 AFS_STATS_NUM_DOWNTIME_INCIDENTS_BUCKETS * sizeof(afs_int32));
395 afs_stats_cmperf.vl_UpDown[0].sumOfRecordAges = 0;
396 afs_stats_cmperf.vl_UpDown[0].ageOfYoungestRecord = 0;
397 afs_stats_cmperf.vl_UpDown[0].ageOfOldestRecord = 0;
398 memset((char *)afs_stats_cmperf.vl_UpDown[0].downIncidents, 0,
399 AFS_STATS_NUM_DOWNTIME_INCIDENTS_BUCKETS * sizeof(afs_int32));
401 afs_stats_cmperf.vl_UpDown[1].sumOfRecordAges = 0;
402 afs_stats_cmperf.vl_UpDown[1].ageOfYoungestRecord = 0;
403 afs_stats_cmperf.vl_UpDown[1].ageOfOldestRecord = 0;
404 memset((char *)afs_stats_cmperf.vl_UpDown[1].downIncidents, 0,
405 AFS_STATS_NUM_DOWNTIME_INCIDENTS_BUCKETS * sizeof(afs_int32));
408 * Compute the current time, used to figure out server record ages.
410 currTimeP = &currTime;
411 osi_GetuTime(currTimeP);
414 * Sweep the server hash table, tallying all we need to know.
416 for (currIdx = 0; currIdx < NSERVERS; currIdx++) {
418 for (currSrvP = afs_servers[currIdx]; currSrvP;
419 currSrvP = currSrvP->next) {
421 * Bump the current chain length.
426 * Any further tallying for this record will only be done if it has
429 if ((currSrvP->flags & AFS_SERVER_FLAG_ACTIVATED)
430 && currSrvP->addr && currSrvP->cell) {
433 * Compute the current server record's age, then remember it
434 * in the appropriate places.
436 srvRecordAge = currTime.tv_sec - currSrvP->activationTime;
437 upDownP = GetUpDownStats(currSrvP);
438 upDownP->sumOfRecordAges += srvRecordAge;
439 if ((upDownP->ageOfYoungestRecord == 0)
440 || (srvRecordAge < upDownP->ageOfYoungestRecord))
441 upDownP->ageOfYoungestRecord = srvRecordAge;
442 if ((upDownP->ageOfOldestRecord == 0)
443 || (srvRecordAge > upDownP->ageOfOldestRecord))
444 upDownP->ageOfOldestRecord = srvRecordAge;
446 if (currSrvP->numDowntimeIncidents <=
447 AFS_STATS_MAX_DOWNTIME_INCIDENTS_BUCKET0)
448 (upDownP->downIncidents[0])++;
449 else if (currSrvP->numDowntimeIncidents <=
450 AFS_STATS_MAX_DOWNTIME_INCIDENTS_BUCKET1)
451 (upDownP->downIncidents[1])++;
452 else if (currSrvP->numDowntimeIncidents <=
453 AFS_STATS_MAX_DOWNTIME_INCIDENTS_BUCKET2)
454 (upDownP->downIncidents[2])++;
455 else if (currSrvP->numDowntimeIncidents <=
456 AFS_STATS_MAX_DOWNTIME_INCIDENTS_BUCKET3)
457 (upDownP->downIncidents[3])++;
458 else if (currSrvP->numDowntimeIncidents <=
459 AFS_STATS_MAX_DOWNTIME_INCIDENTS_BUCKET4)
460 (upDownP->downIncidents[4])++;
462 (upDownP->downIncidents[5])++;
465 } /*Current server has been active */
466 } /*Walk this chain */
469 * Before advancing to the next chain, remember facts about this one.
471 if (currChainLen > afs_stats_cmperf.srvMaxChainLength) {
473 * We beat out the former champion (which was initially set to 0
474 * here). Mark down the new winner, and also remember if it's an
477 afs_stats_cmperf.srvMaxChainLength = currChainLen;
478 if (currChainLen > afs_stats_cmperf.srvMaxChainLengthHWM)
479 afs_stats_cmperf.srvMaxChainLengthHWM = currChainLen;
480 } /*Update chain length maximum */
481 } /*For each hash chain */
484 * We're done. Unlock the server table before returning to our caller.
486 ReleaseReadLock(&afs_xserver);
488 } /*afs_CountServers */
491 /* check down servers (if adown), or running servers (if !adown) */
493 afs_CheckServers(int adown, struct cell *acellp)
495 struct vrequest treq;
501 afs_int32 start, end, delta;
507 struct srvAddr **addrs;
510 struct rx_connection **rxconns;
512 afs_int32 *conntimer, *deltas;
515 AFS_STATCNT(afs_CheckServers);
517 conns = (struct conn **)0;
518 rxconns = (struct rx_connection **) 0;
522 if ((code = afs_InitReq(&treq, afs_osi_credp)))
524 ObtainReadLock(&afs_xserver); /* Necessary? */
525 ObtainReadLock(&afs_xsrvAddr);
528 for (i = 0; i < NSERVERS; i++) {
529 for (sa = afs_srvAddrs[i]; sa; sa = sa->next_bkt) {
534 addrs = afs_osi_Alloc(srvAddrCount * sizeof(*addrs));
536 for (i = 0; i < NSERVERS; i++) {
537 for (sa = afs_srvAddrs[i]; sa; sa = sa->next_bkt) {
538 if (j >= srvAddrCount)
544 ReleaseReadLock(&afs_xsrvAddr);
545 ReleaseReadLock(&afs_xserver);
547 conns = (struct conn **)afs_osi_Alloc(j * sizeof(struct conn *));
548 rxconns = (struct rx_connection **)afs_osi_Alloc(j * sizeof(struct rx_connection *));
549 conntimer = (afs_int32 *)afs_osi_Alloc(j * sizeof (afs_int32));
550 deltas = (afs_int32 *)afs_osi_Alloc(j * sizeof (afs_int32));
552 for (i = 0; i < j; i++) {
559 /* See if a cell to check was specified. If it is spec'd and not
560 * this server's cell, just skip the server.
562 if (acellp && acellp != ts->cell)
565 if ((!adown && (sa->sa_flags & SRVADDR_ISDOWN))
566 || (adown && !(sa->sa_flags & SRVADDR_ISDOWN)))
569 /* check vlserver with special code */
570 if (sa->sa_portal == AFS_VLPORT) {
571 CheckVLServer(sa, &treq);
575 if (!ts->cell) /* not really an active server, anyway, it must */
576 continue; /* have just been added by setsprefs */
578 /* get a connection, even if host is down; bumps conn ref count */
579 tu = afs_GetUser(treq.uid, ts->cell->cellNum, SHARED_LOCK);
580 tc = afs_ConnBySA(sa, ts->cell->fsport, ts->cell->cellNum, tu,
581 1 /*force */ , 1 /*create */ , SHARED_LOCK);
582 afs_PutUser(tu, SHARED_LOCK);
586 if ((sa->sa_flags & SRVADDR_ISDOWN) || afs_HaveCallBacksFrom(sa->server)
587 || (tc->srvr->server == afs_setTimeHost)) {
589 rxconns[nconns]=tc->id;
590 if (sa->sa_flags & SRVADDR_ISDOWN) {
591 rx_SetConnDeadTime(tc->id, 3);
598 } /* Outer loop over addrs */
600 start = osi_Time(); /* time the gettimeofday call */
602 multi_Rx(rxconns,nconns)
604 tv.tv_sec = tv.tv_usec = 0;
605 multi_RXAFS_GetTime(&tv.tv_sec, &tv.tv_usec);
608 if (conntimer[multi_i] == 0)
609 rx_SetConnDeadTime(tc->id, AFS_RXDEADTIME);
611 if ((start == end) && !multi_error)
612 deltas[multi_i] = end - tv.tv_sec;
613 if (( multi_error >= 0 ) && (sa->sa_flags & SRVADDR_ISDOWN) && (tc->srvr == sa)) {
615 print_internet_address("afs: file server ", sa, " is back up", 2);
617 ObtainWriteLock(&afs_xserver, 244);
618 ObtainWriteLock(&afs_xsrvAddr, 245);
619 afs_MarkServerUpOrDown(sa, 0);
620 ReleaseWriteLock(&afs_xsrvAddr);
621 ReleaseWriteLock(&afs_xserver);
623 if (afs_waitForeverCount) {
624 afs_osi_Wakeup(&afs_waitForever);
627 if (multi_error < 0) {
630 ForceNewConnections(sa); /* multi homed clients */
638 * If we're supposed to set the time, and the call worked
639 * quickly (same second response) and this is the host we
640 * use for the time and the time is really different, then
641 * really set the time
643 if (afs_setTime != 0) {
644 for (i=0; i<nconns; i++) {
649 if ((tc->srvr->server == afs_setTimeHost ||
650 /* Sync only to a server in the local cell */
651 (afs_setTimeHost == (struct server *)0 &&
652 afs_IsPrimaryCell(sa->server->cell)))) {
654 char msgbuf[90]; /* strlen("afs: setting clock...") + slop */
655 delta = end - tv.tv_sec; /* how many secs fast we are */
657 afs_setTimeHost = tc->srvr->server;
658 /* see if clock has changed enough to make it worthwhile */
659 if (delta >= AFS_MINCHANGE || delta <= -AFS_MINCHANGE) {
661 if (delta > AFS_MAXCHANGEBACK) {
662 /* setting clock too far back, just do it a little */
663 tv.tv_sec = end - AFS_MAXCHANGEBACK;
665 tv.tv_sec = end - delta;
667 afs_osi_SetTime(&tv);
669 strcpy(msgbuf, "afs: setting clock back ");
670 if (delta > AFS_MAXCHANGEBACK) {
672 afs_cv2string(&tbuffer[CVBS],
674 afs_strcat(msgbuf, " seconds (of ");
676 afs_cv2string(&tbuffer[CVBS],
679 afs_strcat(msgbuf, ", via ");
680 print_internet_address(msgbuf, sa,
681 "); clock is still fast.",
685 afs_cv2string(&tbuffer[CVBS], delta));
686 afs_strcat(msgbuf, " seconds (via ");
687 print_internet_address(msgbuf, sa, ").", 0);
690 strcpy(msgbuf, "afs: setting clock ahead ");
692 afs_cv2string(&tbuffer[CVBS], -delta));
693 afs_strcat(msgbuf, " seconds (via ");
694 print_internet_address(msgbuf, sa, ").", 0);
696 /* We're only going to set it once; why bother looping? */
702 for (i = 0; i < nconns; i++) {
703 afs_PutConn(conns[i], SHARED_LOCK); /* done with it now */
706 afs_osi_Free(addrs, srvAddrCount * sizeof(*addrs));
707 afs_osi_Free(conns, j * sizeof(struct conn *));
708 afs_osi_Free(rxconns, j * sizeof(struct rx_connection *));
709 afs_osi_Free(conntimer, j * sizeof(afs_int32));
710 afs_osi_Free(deltas, j * sizeof(afs_int32));
712 } /*afs_CheckServers*/
715 /* find a server structure given the host address */
717 afs_FindServer(afs_int32 aserver, afs_uint16 aport, afsUUID * uuidp,
724 AFS_STATCNT(afs_FindServer);
726 i = afs_uuid_hash(uuidp) % NSERVERS;
727 for (ts = afs_servers[i]; ts; ts = ts->next) {
728 if ((ts->flags & SRVR_MULTIHOMED)
730 (memcmp((char *)uuidp, (char *)&ts->sr_uuid, sizeof(*uuidp))
731 == 0) && (!ts->addr || (ts->addr->sa_portal == aport)))
736 for (sa = afs_srvAddrs[i]; sa; sa = sa->next_bkt) {
737 if ((sa->sa_ip == aserver) && (sa->sa_portal == aport)) {
744 } /*afs_FindServer */
747 /* some code for creating new server structs and setting preferences follows
748 * in the next few lines...
751 #define MAXDEFRANK 60000
752 #define DEFRANK 40000
754 /* Random number generator and constants from KnuthV2 2d ed, p170 */
760 a is 0.73m should be 0.01m .. 0.99m
761 c is more or less immaterial. 1 or a is suggested.
763 NB: LOW ORDER BITS are not very random. To get small random numbers,
764 treat result as <1, with implied binary point, and multiply by
766 NB: Has to be unsigned, since shifts on signed quantities may preserve
769 /* added rxi_getaddr() to try to get as much initial randomness as
770 possible, since at least one customer reboots ALL their clients
771 simultaneously -- so osi_Time is bound to be the same on some of the
772 clients. This is probably OK, but I don't want to see too much of it.
775 #define ranstage(x) (x)= (afs_uint32) (3141592621U*((afs_uint32)x)+1)
780 static afs_int32 state = 0;
783 AFS_STATCNT(afs_random);
788 * 0xfffffff0 was changed to (~0 << 4) since it works no matter how many
789 * bits are in a tv_usec
791 state = (t.tv_usec & (~0 << 4)) + (rxi_getaddr() & 0xff);
792 state += (t.tv_sec & 0xff);
793 for (i = 0; i < 30; i++) {
803 /* returns int 0..14 using the high bits of a pseudo-random number instead of
804 the low bits, as the low bits are "less random" than the high ones...
805 slight roundoff error exists, an excercise for the reader.
806 need to multiply by something with lots of ones in it, so multiply by
807 8 or 16 is right out.
810 afs_randomMod15(void)
814 temp = afs_random() >> 4;
815 temp = (temp * 15) >> 28;
821 afs_randomMod127(void)
825 temp = afs_random() >> 7;
826 temp = (temp * 127) >> 25;
831 /* afs_SortOneServer()
832 * Sort all of the srvAddrs, of a server struct, by rank from low to high.
835 afs_SortOneServer(struct server *asp)
837 struct srvAddr **rootsa, *lowsa, *tsa, *lowprev;
840 for (rootsa = &(asp->addr); *rootsa; rootsa = &(lowsa->next_sa)) {
842 lowsa = *rootsa; /* lowest sa is the first one */
843 lowrank = lowsa->sa_iprank;
845 for (tsa = *rootsa; tsa->next_sa; tsa = tsa->next_sa) {
846 rank = tsa->next_sa->sa_iprank;
847 if (rank < lowrank) {
849 lowsa = tsa->next_sa;
850 lowrank = lowsa->sa_iprank;
853 if (lowprev) { /* found one lower, so rearrange them */
854 lowprev->next_sa = lowsa->next_sa;
855 lowsa->next_sa = *rootsa;
862 * Sort the pointer to servers by the server's rank (its lowest rank).
863 * It is assumed that the server already has its IP addrs sorted (the
864 * first being its lowest rank: afs_GetServer() calls afs_SortOneServer()).
867 afs_SortServers(struct server *aservers[], int count)
872 AFS_STATCNT(afs_SortServers);
874 for (i = 0; i < count; i++) {
877 for (low = i, j = i + 1; j <= count; j++) {
878 if ((!aservers[j]) || (!aservers[j]->addr))
880 if ((!aservers[low]) || (!aservers[low]->addr))
882 if (aservers[j]->addr->sa_iprank < aservers[low]->addr->sa_iprank) {
888 aservers[i] = aservers[low];
892 } /*afs_SortServers */
894 /* afs_SetServerPrefs is rather system-dependent. It pokes around in kernel
895 data structures to determine what the local IP addresses and subnet masks
896 are in order to choose which server(s) are on the local subnet.
898 As I see it, there are several cases:
899 1. The server address is one of this host's local addresses. In this case
900 this server is to be preferred over all others.
901 2. The server is on the same subnet as one of the this host's local
902 addresses. (ie, an odd-sized subnet, not class A,B,orC)
903 3. The server is on the same net as this host (class A,B or C)
904 4. The server is on a different logical subnet or net than this host, but
905 this host is a 'metric 0 gateway' to it. Ie, two address-spaces share
907 5. This host has a direct (point-to-point, ie, PPP or SLIP) link to the
909 6. This host and the server are disjoint.
911 That is a rough order of preference. If a point-to-point link has a high
912 metric, I'm assuming that it is a very slow link, and putting it at the
913 bottom of the list (at least until RX works better over slow links). If
914 its metric is 1, I'm assuming that it's relatively fast (T1) and putting
916 It's not easy to check for case #4, so I'm ignoring it for the time being.
918 BSD "if" code keeps track of some rough network statistics (cf 'netstat -i')
919 That could be used to prefer certain servers fairly easily. Maybe some
922 NOTE: this code is very system-dependent, and very dependent on the TCP/IP
923 protocols (well, addresses that are stored in uint32s, at any rate).
926 #define IA_DST(ia)((struct sockaddr_in *)(&((struct in_ifaddr *)ia)->ia_dstaddr))
927 #define IA_BROAD(ia)((struct sockaddr_in *)(&((struct in_ifaddr *)ia)->ia_broadaddr))
929 /* SA2ULONG takes a sockaddr_in, not a sockaddr (same thing, just cast it!) */
930 #define SA2ULONG(sa) ((sa)->sin_addr.s_addr)
935 #define PPWEIGHT 4096
940 #if defined(AFS_SUN5_ENV) && ! defined(AFS_SUN56_ENV)
941 #include <inet/common.h>
942 /* IP interface structure, one per local address */
943 typedef struct ipif_s {
944 /**/ struct ipif_s *ipif_next;
945 struct ill_s *ipif_ill; /* Back pointer to our ill */
946 long ipif_id; /* Logical unit number */
947 u_int ipif_mtu; /* Starts at ipif_ill->ill_max_frag */
948 afs_int32 ipif_local_addr; /* Local IP address for this if. */
949 afs_int32 ipif_net_mask; /* Net mask for this interface. */
950 afs_int32 ipif_broadcast_addr; /* Broadcast addr for this interface. */
951 afs_int32 ipif_pp_dst_addr; /* Point-to-point dest address. */
952 u_int ipif_flags; /* Interface flags. */
953 u_int ipif_metric; /* BSD if metric, for compatibility. */
954 u_int ipif_ire_type; /* LOCAL or LOOPBACK */
955 mblk_t *ipif_arp_down_mp; /* Allocated at time arp comes up to
956 * prevent awkward out of mem condition
959 mblk_t *ipif_saved_ire_mp; /* Allocated for each extra IRE_SUBNET/
960 * RESOLVER on this interface so that
961 * they can survive ifconfig down.
964 * The packet counts in the ipif contain the sum of the
965 * packet counts in dead IREs that were affiliated with
968 u_long ipif_fo_pkt_count; /* Forwarded thru our dead IREs */
969 u_long ipif_ib_pkt_count; /* Inbound packets for our dead IREs */
970 u_long ipif_ob_pkt_count; /* Outbound packets to our dead IREs */
972 ipif_multicast_up:1, /* We have joined the allhosts group */
976 typedef struct ipfb_s {
977 /**/ struct ipf_s *ipfb_ipf; /* List of ... */
978 kmutex_t ipfb_lock; /* Protect all ipf in list */
981 typedef struct ilm_s {
982 /**/ afs_int32 ilm_addr;
984 u_int ilm_timer; /* IGMP */
985 struct ipif_s *ilm_ipif; /* Back pointer to ipif */
986 struct ilm_s *ilm_next; /* Linked list for each ill */
989 typedef struct ill_s {
990 /**/ struct ill_s *ill_next; /* Chained in at ill_g_head. */
991 struct ill_s **ill_ptpn; /* Pointer to previous next. */
992 queue_t *ill_rq; /* Read queue. */
993 queue_t *ill_wq; /* Write queue. */
995 int ill_error; /* Error value sent up by device. */
997 ipif_t *ill_ipif; /* Interface chain for this ILL. */
998 u_int ill_ipif_up_count; /* Number of IPIFs currently up. */
999 u_int ill_max_frag; /* Max IDU. */
1000 char *ill_name; /* Our name. */
1001 u_int ill_name_length; /* Name length, incl. terminator. */
1002 u_int ill_subnet_type; /* IRE_RESOLVER or IRE_SUBNET. */
1003 u_int ill_ppa; /* Physical Point of Attachment num. */
1005 int ill_sap_length; /* Including sign (for position) */
1006 u_int ill_phys_addr_length; /* Excluding the sap. */
1007 mblk_t *ill_frag_timer_mp; /* Reassembly timer state. */
1008 ipfb_t *ill_frag_hash_tbl; /* Fragment hash list head. */
1010 queue_t *ill_bind_pending_q; /* Queue waiting for DL_BIND_ACK. */
1011 ipif_t *ill_ipif_pending; /* IPIF waiting for DL_BIND_ACK. */
1013 /* ill_hdr_length and ill_hdr_mp will be non zero if
1014 * the underlying device supports the M_DATA fastpath
1018 ilm_t *ill_ilm; /* Multicast mebership for lower ill */
1020 /* All non-nil cells between 'ill_first_mp_to_free' and
1021 * 'ill_last_mp_to_free' are freed in ill_delete.
1023 #define ill_first_mp_to_free ill_hdr_mp
1024 mblk_t *ill_hdr_mp; /* Contains fastpath template */
1025 mblk_t *ill_bcast_mp; /* DLPI header for broadcasts. */
1026 mblk_t *ill_bind_pending; /* T_BIND_REQ awaiting completion. */
1027 mblk_t *ill_resolver_mp; /* Resolver template. */
1028 mblk_t *ill_attach_mp;
1029 mblk_t *ill_bind_mp;
1030 mblk_t *ill_unbind_mp;
1031 mblk_t *ill_detach_mp;
1032 #define ill_last_mp_to_free ill_detach_mp
1034 u_int ill_frag_timer_running:1, ill_needs_attach:1, ill_is_ptp:1,
1035 ill_priv_stream:1, ill_unbind_pending:1, ill_pad_to_bit_31:27;
1036 MI_HRT_DCL(ill_rtime)
1037 MI_HRT_DCL(ill_rtmp)
1041 #ifdef AFS_USERSPACE_IP_ADDR
1043 #define afs_min(A,B) ((A)<(B)) ? (A) : (B)
1046 * The IP addresses and ranks are determined by afsd (in user space) and
1047 * passed into the kernel at startup time through the AFSOP_ADVISEADDR
1048 * system call. These are stored in the data structure
1049 * called 'afs_cb_interface'.
1051 * struct srvAddr *sa; remote server
1052 * afs_int32 addr; one of my local addr in net order
1053 * afs_uint32 subnetmask; subnet mask of local addr in net order
1057 afsi_SetServerIPRank(struct srvAddr *sa, afs_int32 addr,
1058 afs_uint32 subnetmask)
1060 afs_uint32 myAddr, myNet, mySubnet, netMask;
1061 afs_uint32 serverAddr;
1063 myAddr = ntohl(addr); /* one of my IP addr in host order */
1064 serverAddr = ntohl(sa->sa_ip); /* server's IP addr in host order */
1065 subnetmask = ntohl(subnetmask); /* subnet mask in host order */
1067 if (IN_CLASSA(myAddr))
1068 netMask = IN_CLASSA_NET;
1069 else if (IN_CLASSB(myAddr))
1070 netMask = IN_CLASSB_NET;
1071 else if (IN_CLASSC(myAddr))
1072 netMask = IN_CLASSC_NET;
1076 myNet = myAddr & netMask;
1077 mySubnet = myAddr & subnetmask;
1079 if ((serverAddr & netMask) == myNet) {
1080 if ((serverAddr & subnetmask) == mySubnet) {
1081 if (serverAddr == myAddr) { /* same machine */
1082 sa->sa_iprank = afs_min(sa->sa_iprank, TOPR);
1083 } else { /* same subnet */
1084 sa->sa_iprank = afs_min(sa->sa_iprank, HI);
1086 } else { /* same net */
1087 sa->sa_iprank = afs_min(sa->sa_iprank, MED);
1091 #else /* AFS_USERSPACE_IP_ADDR */
1092 #if (! defined(AFS_SUN5_ENV)) && !defined(AFS_DARWIN60_ENV) && defined(USEIFADDR)
1094 afsi_SetServerIPRank(struct srvAddr *sa, struct in_ifaddr *ifa)
1096 struct sockaddr_in *sin;
1099 if ((ntohl(sa->sa_ip) & ifa->ia_netmask) == ifa->ia_net) {
1100 if ((ntohl(sa->sa_ip) & ifa->ia_subnetmask) == ifa->ia_subnet) {
1102 if (SA2ULONG(sin) == ntohl(sa->sa_ip)) { /* ie, ME!!! */
1103 sa->sa_iprank = TOPR;
1105 t = HI + ifa->ia_ifp->if_metric; /* case #2 */
1106 if (sa->sa_iprank > t)
1110 t = MED + ifa->ia_ifp->if_metric; /* case #3 */
1111 if (sa->sa_iprank > t)
1115 #ifdef IFF_POINTTOPOINT
1116 /* check for case #4 -- point-to-point link */
1117 if ((ifa->ia_ifp->if_flags & IFF_POINTOPOINT)
1118 && (SA2ULONG(IA_DST(ifa)) == ntohl(sa->sa_ip))) {
1119 if (ifa->ia_ifp->if_metric >= (MAXDEFRANK - MED) / PPWEIGHT)
1122 t = MED + (PPWEIGHT << ifa->ia_ifp->if_metric);
1123 if (sa->sa_iprank > t)
1126 #endif /* IFF_POINTTOPOINT */
1128 #endif /*(!defined(AFS_SUN5_ENV)) && defined(USEIFADDR) */
1129 #if defined(AFS_DARWIN60_ENV) && defined(USEIFADDR)
1131 #define afs_min(A,B) ((A)<(B)) ? (A) : (B)
1134 afsi_SetServerIPRank(sa, ifa)
1138 struct sockaddr_in *sin;
1141 afs_uint32 subnetmask, myAddr, myNet, myDstaddr, mySubnet, netMask;
1142 afs_uint32 serverAddr;
1144 if (ifa->ifa_addr->sa_family != AF_INET)
1146 sin = (struct sockaddr_in *)ifa->ifa_addr;
1147 myAddr = ntohl(sin->sin_addr.s_addr); /* one of my IP addr in host order */
1148 serverAddr = ntohl(sa->sa_ip); /* server's IP addr in host order */
1149 sin = (struct sockaddr_in *)ifa->ifa_netmask;
1150 subnetmask = ntohl(sin->sin_addr.s_addr); /* subnet mask in host order */
1151 sin = (struct sockaddr_in *)ifa->ifa_dstaddr;
1153 myDstaddr = sin->sin_addr.s_addr;
1155 if (IN_CLASSA(myAddr))
1156 netMask = IN_CLASSA_NET;
1157 else if (IN_CLASSB(myAddr))
1158 netMask = IN_CLASSB_NET;
1159 else if (IN_CLASSC(myAddr))
1160 netMask = IN_CLASSC_NET;
1164 myNet = myAddr & netMask;
1165 mySubnet = myAddr & subnetmask;
1167 if ((serverAddr & netMask) == myNet) {
1168 if ((serverAddr & subnetmask) == mySubnet) {
1169 if (serverAddr == myAddr) { /* same machine */
1170 sa->sa_iprank = afs_min(sa->sa_iprank, TOPR);
1171 } else { /* same subnet */
1172 sa->sa_iprank = afs_min(sa->sa_iprank, HI + ifa->ifa_metric);
1174 } else { /* same net */
1175 sa->sa_iprank = afs_min(sa->sa_iprank, MED + ifa->ifa_metric);
1178 #ifdef IFF_POINTTOPOINT
1179 /* check for case #4 -- point-to-point link */
1180 if ((ifa->ia_ifp->if_flags & IFF_POINTOPOINT)
1181 && (myDstaddr == serverAddr)) {
1182 if (ifa->ia_ifp->if_metric >= (MAXDEFRANK - MED) / PPWEIGHT)
1185 t = MED + (PPWEIGHT << ifa->->ifa_metric);
1186 if (sa->sa_iprank > t)
1189 #endif /* IFF_POINTTOPOINT */
1191 #endif /*(!defined(AFS_SUN5_ENV)) && defined(USEIFADDR) */
1192 #endif /* else AFS_USERSPACE_IP_ADDR */
1194 #ifdef AFS_SGI62_ENV
1197 afsi_enum_set_rank(struct hashbucket *h, caddr_t mkey, caddr_t arg1,
1199 afsi_SetServerIPRank((struct srvAddr *)arg1, (struct in_ifaddr *)h);
1200 return 0; /* Never match, so we enumerate everyone */
1202 #endif /* AFS_SGI62_ENV */
1203 static int afs_SetServerPrefs(struct srvAddr *sa) {
1204 #if defined(AFS_USERSPACE_IP_ADDR)
1208 for (i = 0; i < afs_cb_interface.numberOfInterfaces; i++) {
1209 afsi_SetServerIPRank(sa, afs_cb_interface.addr_in[i],
1210 afs_cb_interface.subnetmask[i]);
1212 #else /* AFS_USERSPACE_IP_ADDR */
1213 #if defined(AFS_SUN5_ENV)
1214 #ifdef AFS_SUN510_ENV
1215 ill_walk_context_t ctx;
1217 extern struct ill_s *ill_g_headp;
1218 long *addr = (long *)ill_g_headp;
1222 int subnet, subnetmask, net, netmask;
1226 #ifdef AFS_SUN510_ENV
1227 for (ill = ILL_START_WALK_ALL(&ctx) ; ill ; ill = ill_next(&ctx, ill)) {
1229 for (ill = (struct ill_s *)*addr /*ill_g_headp */ ; ill;
1230 ill = ill->ill_next) {
1232 #ifdef AFS_SUN58_ENV
1233 /* Make sure this is an IPv4 ILL */
1237 for (ipif = ill->ill_ipif; ipif; ipif = ipif->ipif_next) {
1238 subnet = ipif->ipif_local_addr & ipif->ipif_net_mask;
1239 subnetmask = ipif->ipif_net_mask;
1241 * Generate the local net using the local address and
1242 * whate we know about Class A, B and C networks.
1244 if (IN_CLASSA(ipif->ipif_local_addr)) {
1245 netmask = IN_CLASSA_NET;
1246 } else if (IN_CLASSB(ipif->ipif_local_addr)) {
1247 netmask = IN_CLASSB_NET;
1248 } else if (IN_CLASSC(ipif->ipif_local_addr)) {
1249 netmask = IN_CLASSC_NET;
1253 net = ipif->ipif_local_addr & netmask;
1256 if (ipif->ipif_local_addr != 0x7f000001) { /* ignore loopback */
1260 *addrp++ = ipif->ipif_local_addr;
1265 /* XXXXXX Do the individual ip ranking below XXXXX */
1266 if ((sa->sa_ip & netmask) == net) {
1267 if ((sa->sa_ip & subnetmask) == subnet) {
1268 if (ipif->ipif_local_addr == sa->sa_ip) { /* ie, ME! */
1269 sa->sa_iprank = TOPR;
1271 sa->sa_iprank = HI + ipif->ipif_metric; /* case #2 */
1274 sa->sa_iprank = MED + ipif->ipif_metric; /* case #3 */
1277 sa->sa_iprank = LO + ipif->ipif_metric; /* case #4 */
1279 /* check for case #5 -- point-to-point link */
1280 if ((ipif->ipif_flags & IFF_POINTOPOINT)
1281 && (ipif->ipif_pp_dst_addr == sa->sa_ip)) {
1283 if (ipif->ipif_metric >= (MAXDEFRANK - MED) / PPWEIGHT)
1284 sa->sa_iprank = MAXDEFRANK;
1286 sa->sa_iprank = MED + (PPWEIGHT << ipif->ipif_metric);
1293 struct ifnet *ifn = NULL;
1294 struct in_ifaddr *ifad = (struct in_ifaddr *)0;
1295 struct sockaddr_in *sin;
1298 #ifdef notdef /* clean up, remove this */
1299 for (ifn = ifnet; ifn != NULL; ifn = ifn->if_next) {
1300 for (ifad = ifn->if_addrlist; ifad != NULL; ifad = ifad->ifa_next) {
1301 if ((IFADDR2SA(ifad)->sa_family == AF_INET)
1302 && !(ifn->if_flags & IFF_LOOPBACK)) {
1307 ((struct sockaddr_in *)IFADDR2SA(ifad))->sin_addr.
1316 ifn = rxi_FindIfnet(sa->sa_ip, &ifad);
1318 if (ifn) { /* local, more or less */
1320 if (ifn->if_flags & IFF_LOOPBACK) {
1321 sa->sa_iprank = TOPR;
1324 #endif /* IFF_LOOPBACK */
1325 sin = (struct sockaddr_in *)IA_SIN(ifad);
1326 if (SA2ULONG(sin) == sa->sa_ip) {
1327 sa->sa_iprank = TOPR;
1330 #ifdef IFF_BROADCAST
1331 if (ifn->if_flags & IFF_BROADCAST) {
1332 if (sa->sa_ip == (sa->sa_ip & SA2ULONG(IA_BROAD(ifad)))) {
1337 #endif /* IFF_BROADCAST */
1338 #ifdef IFF_POINTOPOINT
1339 if (ifn->if_flags & IFF_POINTOPOINT) {
1340 if (sa->sa_ip == SA2ULONG(IA_DST(ifad))) {
1341 if (ifn->if_metric > 4) {
1345 sa->sa_iprank = ifn->if_metric;
1348 #endif /* IFF_POINTOPOINT */
1349 sa->sa_iprank += MED + ifn->if_metric; /* couldn't find anything better */
1351 #else /* USEIFADDR */
1355 #ifdef AFS_SGI62_ENV
1356 (void)hash_enum(&hashinfo_inaddr, afsi_enum_set_rank, HTF_INET, NULL,
1357 (caddr_t) sa, NULL);
1358 #elif defined(AFS_DARWIN60_ENV)
1362 TAILQ_FOREACH(ifn, &ifnet, if_link) {
1363 TAILQ_FOREACH(ifa, &ifn->if_addrhead, ifa_link) {
1364 afsi_SetServerIPRank(sa, ifa);
1366 #elif defined(AFS_DARWIN_ENV) || defined(AFS_FBSD_ENV)
1368 struct in_ifaddr *ifa;
1369 TAILQ_FOREACH(ifa, &in_ifaddrhead, ia_link) {
1370 afsi_SetServerIPRank(sa, ifa);
1372 #elif defined(AFS_OBSD_ENV)
1374 extern struct in_ifaddrhead in_ifaddr;
1375 struct in_ifaddr *ifa;
1376 for (ifa = in_ifaddr.tqh_first; ifa; ifa = ifa->ia_list.tqe_next)
1377 afsi_SetServerIPRank(sa, ifa);
1381 struct in_ifaddr *ifa;
1382 for (ifa = in_ifaddr; ifa; ifa = ifa->ia_next) {
1383 afsi_SetServerIPRank(sa, ifa);
1387 #endif /* USEIFADDR */
1388 #endif /* AFS_SUN5_ENV */
1389 #endif /* else AFS_USERSPACE_IP_ADDR */
1391 sa->sa_iprank += afs_randomMod15();
1394 } /* afs_SetServerPrefs */
1402 /* afs_FlushServer()
1403 * The addresses on this server struct has changed in some way and will
1404 * clean up all other structures that may reference it.
1405 * The afs_xserver and afs_xsrvAddr locks are assumed taken.
1407 void afs_FlushServer(struct server *srvp) {
1409 struct server *ts, **pts;
1411 /* Find any volumes residing on this server and flush their state */
1412 afs_ResetVolumes(srvp);
1414 /* Flush all callbacks in the all vcaches for this specific server */
1415 afs_FlushServerCBs(srvp);
1417 /* Remove all the callbacks structs */
1419 struct afs_cbr *cb, *cbnext;
1421 MObtainWriteLock(&afs_xvcb, 300);
1422 for (cb = srvp->cbrs; cb; cb = cbnext) {
1425 } srvp->cbrs = (struct afs_cbr *)0;
1426 ReleaseWriteLock(&afs_xvcb);
1429 /* If no more srvAddr structs hanging off of this server struct,
1433 /* Remove the server structure from the cell list - if there */
1434 afs_RemoveCellEntry(srvp);
1436 /* Remove from the afs_servers hash chain */
1437 for (i = 0; i < NSERVERS; i++) {
1438 for (pts = &(afs_servers[i]), ts = *pts; ts;
1439 pts = &(ts->next), ts = *pts) {
1447 *pts = ts->next; /* Found it. Remove it */
1448 afs_osi_Free(ts, sizeof(struct server)); /* Free it */
1454 /* afs_RemoveSrvAddr()
1455 * This removes a SrvAddr structure from its server structure.
1456 * The srvAddr struct is not free'd because it connections may still
1457 * be open to it. It is up to the calling process to make sure it
1458 * remains connected to a server struct.
1459 * The afs_xserver and afs_xsrvAddr locks are assumed taken.
1460 * It is not removed from the afs_srvAddrs hash chain.
1462 void afs_RemoveSrvAddr(struct srvAddr *sap) {
1463 struct srvAddr **psa, *sa;
1470 /* Find the srvAddr in the server's list and remove it */
1471 for (psa = &(srv->addr), sa = *psa; sa; psa = &(sa->next_sa), sa = *psa) {
1479 /* Flush the server struct since it's IP address has changed */
1480 afs_FlushServer(srv);
1485 * Return an updated and properly initialized server structure
1486 * corresponding to the server ID, cell, and port specified.
1487 * If one does not exist, then one will be created.
1488 * aserver and aport must be in NET byte order.
1490 struct server *afs_GetServer(afs_uint32 * aserverp, afs_int32 nservers,
1491 afs_int32 acell, u_short aport,
1492 afs_int32 locktype, afsUUID * uuidp,
1493 afs_int32 addr_uniquifier) {
1494 struct server *oldts = 0, *ts, *newts, *orphts = 0;
1495 struct srvAddr *oldsa, *newsa, *nextsa, *orphsa;
1497 afs_int32 iphash, k, srvcount = 0;
1498 unsigned int srvhash;
1500 AFS_STATCNT(afs_GetServer);
1502 ObtainSharedLock(&afs_xserver, 13);
1504 /* Check if the server struct exists and is up to date */
1507 panic("afs_GetServer: incorect count of servers");
1508 ObtainReadLock(&afs_xsrvAddr);
1509 ts = afs_FindServer(aserverp[0], aport, NULL, locktype);
1510 ReleaseReadLock(&afs_xsrvAddr);
1511 if (ts && !(ts->flags & SRVR_MULTIHOMED)) {
1512 /* Found a server struct that is not multihomed and has the
1513 * IP address associated with it. A correct match.
1515 ReleaseSharedLock(&afs_xserver);
1520 panic("afs_GetServer: incorrect count of servers");
1521 ts = afs_FindServer(0, aport, uuidp, locktype);
1522 if (ts && (ts->sr_addr_uniquifier == addr_uniquifier) && ts->addr) {
1523 /* Found a server struct that is multihomed and same
1524 * uniqufier (same IP addrs). The above if statement is the
1525 * same as in InstallUVolumeEntry().
1527 ReleaseSharedLock(&afs_xserver);
1531 oldts = ts; /* Will reuse if same uuid */
1534 UpgradeSToWLock(&afs_xserver, 36);
1535 ObtainWriteLock(&afs_xsrvAddr, 116);
1537 srvcount = afs_totalServers;
1539 /* Reuse/allocate a new server structure */
1543 newts = (struct server *)afs_osi_Alloc(sizeof(struct server));
1545 panic("malloc of server struct");
1547 memset((char *)newts, 0, sizeof(struct server));
1549 /* Add the server struct to the afs_servers[] hash chain */
1551 (uuidp ? (afs_uuid_hash(uuidp) % NSERVERS) : SHash(aserverp[0]));
1552 newts->next = afs_servers[srvhash];
1553 afs_servers[srvhash] = newts;
1556 /* Initialize the server structure */
1557 if (uuidp) { /* Multihomed */
1558 newts->sr_uuid = *uuidp;
1559 newts->sr_addr_uniquifier = addr_uniquifier;
1560 newts->flags |= SRVR_MULTIHOMED;
1563 newts->cell = afs_GetCell(acell, 0);
1565 fsport = (newts->cell ? newts->cell->fsport : AFS_FSPORT);
1567 /* For each IP address we are registering */
1568 for (k = 0; k < nservers; k++) {
1569 iphash = SHash(aserverp[k]);
1571 /* Check if the srvAddr structure already exists. If so, remove
1572 * it from its server structure and add it to the new one.
1574 for (oldsa = afs_srvAddrs[iphash]; oldsa; oldsa = oldsa->next_bkt) {
1575 if ((oldsa->sa_ip == aserverp[k]) && (oldsa->sa_portal == aport))
1578 if (oldsa && (oldsa->server != newts)) {
1579 afs_RemoveSrvAddr(oldsa); /* Remove from its server struct */
1580 oldsa->next_sa = newts->addr; /* Add to the new server struct */
1581 newts->addr = oldsa;
1584 /* Reuse/allocate a new srvAddr structure */
1588 newsa = (struct srvAddr *)afs_osi_Alloc(sizeof(struct srvAddr));
1590 panic("malloc of srvAddr struct");
1591 afs_totalSrvAddrs++;
1592 memset((char *)newsa, 0, sizeof(struct srvAddr));
1594 /* Add the new srvAddr to the afs_srvAddrs[] hash chain */
1595 newsa->next_bkt = afs_srvAddrs[iphash];
1596 afs_srvAddrs[iphash] = newsa;
1598 /* Hang off of the server structure */
1599 newsa->next_sa = newts->addr;
1600 newts->addr = newsa;
1602 /* Initialize the srvAddr Structure */
1603 newsa->sa_ip = aserverp[k];
1604 newsa->sa_portal = aport;
1607 /* Update the srvAddr Structure */
1608 newsa->server = newts;
1609 if (newts->flags & SRVR_ISDOWN)
1610 newsa->sa_flags |= SRVADDR_ISDOWN;
1612 newsa->sa_flags |= SRVADDR_MH;
1614 newsa->sa_flags &= ~SRVADDR_MH;
1616 /* Compute preference values and resort */
1617 if (!newsa->sa_iprank) {
1618 if (aport == fsport) {
1619 afs_SetServerPrefs(newsa); /* new fileserver rank */
1621 newsa->sa_iprank = 10000 + afs_randomMod127(); /* new vlserver rank */
1625 afs_SortOneServer(newts); /* Sort by rank */
1627 /* If we reused the server struct, remove any of its srvAddr
1628 * structs that will no longer be associated with this server.
1630 if (oldts) { /* reused the server struct */
1631 for (orphsa = newts->addr; orphsa; orphsa = nextsa) {
1632 nextsa = orphsa->next_sa;
1633 for (k = 0; k < nservers; k++) {
1634 if (orphsa->sa_ip == aserverp[k])
1635 break; /* belongs */
1638 continue; /* belongs */
1640 /* Have a srvAddr struct. Now get a server struct (if not already) */
1643 (struct server *)afs_osi_Alloc(sizeof(struct server));
1645 panic("malloc of lo server struct");
1646 memset((char *)orphts, 0, sizeof(struct server));
1649 /* Add the orphaned server to the afs_servers[] hash chain.
1650 * Its iphash does not matter since we never look up the server
1651 * in the afs_servers table by its ip address (only by uuid -
1652 * which this has none).
1654 iphash = SHash(aserverp[k]);
1655 orphts->next = afs_servers[iphash];
1656 afs_servers[iphash] = orphts;
1659 orphts->cell = afs_GetCell(acell, 0);
1662 /* Hang the srvAddr struct off of the server structure. The server
1663 * may have multiple srvAddrs, but it won't be marked multihomed.
1665 afs_RemoveSrvAddr(orphsa); /* remove */
1666 orphsa->next_sa = orphts->addr; /* hang off server struct */
1667 orphts->addr = orphsa;
1668 orphsa->server = orphts;
1669 orphsa->sa_flags |= SRVADDR_NOUSE; /* flag indicating not in use */
1670 orphsa->sa_flags &= ~SRVADDR_MH; /* Not multihomed */
1674 srvcount = afs_totalServers - srvcount; /* # servers added and removed */
1676 struct afs_stats_SrvUpDownInfo *upDownP;
1677 /* With the introduction of this new record, we need to adjust the
1678 * proper individual & global server up/down info.
1680 upDownP = GetUpDownStats(newts);
1681 upDownP->numTtlRecords += srvcount;
1682 afs_stats_cmperf.srvRecords += srvcount;
1683 if (afs_stats_cmperf.srvRecords > afs_stats_cmperf.srvRecordsHWM)
1684 afs_stats_cmperf.srvRecordsHWM = afs_stats_cmperf.srvRecords;
1687 ReleaseWriteLock(&afs_xsrvAddr);
1688 ReleaseWriteLock(&afs_xserver);
1690 } /* afs_GetServer */
1692 void afs_ActivateServer(struct srvAddr *sap) {
1693 osi_timeval_t currTime; /*Filled with current time */
1694 osi_timeval_t *currTimeP; /*Ptr to above */
1695 struct afs_stats_SrvUpDownInfo *upDownP; /*Ptr to up/down info record */
1696 struct server *aserver = sap->server;
1698 if (!(aserver->flags & AFS_SERVER_FLAG_ACTIVATED)) {
1700 * This server record has not yet been activated. Go for it,
1701 * recording its ``birth''.
1703 aserver->flags |= AFS_SERVER_FLAG_ACTIVATED;
1704 currTimeP = &currTime;
1705 osi_GetuTime(currTimeP);
1706 aserver->activationTime = currTime.tv_sec;
1707 upDownP = GetUpDownStats(aserver);
1708 if (aserver->flags & SRVR_ISDOWN) {
1709 upDownP->numDownRecords++;
1711 upDownP->numUpRecords++;
1712 upDownP->numRecordsNeverDown++;
1718 void shutdown_server()
1722 for (i = 0; i < NSERVERS; i++) {
1723 struct server *ts, *next;
1725 ts = afs_servers[i];
1728 afs_osi_Free(ts, sizeof(struct server));
1733 for (i = 0; i < NSERVERS; i++) {
1734 struct srvAddr *sa, *next;
1736 sa = afs_srvAddrs[i];
1738 next = sa->next_bkt;
1739 afs_osi_Free(sa, sizeof(struct srvAddr));