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
9 * Portions Copyright (c) 2006 Sine Nomine Associates
13 * NEW callback package callback.c (replaces vicecb.c)
14 * Updated call back routines, NOW with:
16 * Faster DeleteVenus (Now called DeleteAllCallBacks)
17 * Call back breaking for volumes
18 * Adaptive timeouts on call backs
19 * Architected for Multi RPC
20 * No locks (currently implicit vnode locks--these will go, to)
21 * Delayed call back when rpc connection down.
22 * Bulk break of delayed call backs when rpc connection
24 * Strict limit on number of call backs.
26 * InitCallBack(nblocks)
27 * Initialize: nblocks is max number # of file entries + # of callback entries
28 * nblocks must be < 65536
29 * Space used is nblocks*16 bytes
30 * Note that space will be reclaimed by breaking callbacks of old hosts
32 * time = AddCallBack(host, fid)
34 * Returns the expiration time at the workstation.
36 * BreakCallBack(host, fid)
37 * Break all call backs for fid, except for the specified host.
40 * BreakVolumeCallBacks(volume)
41 * Break all call backs on volume, using single call to each host
42 * Delete all the call backs.
44 * DeleteCallBack(host,fid)
45 * Delete (do not break) single call back for fid.
47 * DeleteFileCallBacks(fid)
48 * Delete (do not break) all call backs for fid.
50 * DeleteAllCallBacks(host)
51 * Delete (do not break) all call backs for host.
53 * CleanupTimedOutCallBacks()
54 * Delete all timed out call back entries
55 * Must be called periodically by file server.
57 * BreakDelayedCallBacks(host)
58 * Break all delayed call backs for host.
59 * Returns 1: one or more failed, 0: success.
61 * PrintCallBackStats()
62 * Print statistics about call backs to stdout.
64 * DumpCallBacks() ---wishful thinking---
65 * Dump call back state to /tmp/callback.state.
66 * This is separately interpretable by the program pcb.
68 * Notes: In general, if a call back to a host doesn't get through,
69 * then HostDown, supplied elsewhere, is called. BreakDelayedCallBacks,
70 * however, does not call HostDown, but instead returns an indication of
71 * success if all delayed call backs were finally broken.
73 * BreakDelayedCallBacks MUST be called at the first sign of activity
74 * from the host after HostDown has been called (or a previous
75 * BreakDelayedCallBacks failed). The BreakDelayedCallBacks must be
76 * allowed to complete before any requests from that host are handled.
77 * If BreakDelayedCallBacks fails, then the host should remain
78 * down (and the request should be failed).
80 * CleanupCallBacks MUST be called periodically by the file server for
81 * this package to work correctly. Every 5 minutes is suggested.
84 #include <afsconfig.h>
85 #include <afs/param.h>
91 #include <stdlib.h> /* for malloc() */
92 #include <time.h> /* ANSI standard location for time stuff */
103 #ifdef HAVE_STRINGS_H
107 #include <afs/assert.h>
109 #include <afs/stds.h>
111 #include <afs/nfs.h> /* yuck. This is an abomination. */
114 #include <afscbint.h>
115 #include <afs/afsutil.h>
117 #include <afs/ihandle.h>
118 #include <afs/vnode.h>
119 #include <afs/volume.h>
120 #include "viced_prototypes.h"
123 #include <afs/ptclient.h> /* need definition of prlist for host.h */
125 #include "callback.h"
126 #ifdef AFS_DEMAND_ATTACH_FS
127 #include "../tviced/serialize_state.h"
128 #endif /* AFS_DEMAND_ATTACH_FS */
131 extern afsUUID FS_HostUUID;
132 extern int hostCount;
133 static int ShowProblems = 1;
135 struct cbcounters cbstuff;
137 static struct FileEntry * FE = NULL; /* don't use FE[0] */
138 static struct CallBack * CB = NULL; /* don't use CB[0] */
140 static struct CallBack * CBfree = NULL;
141 static struct FileEntry * FEfree = NULL;
144 /* Time to live for call backs depends upon number of users of the file.
145 * TimeOuts is indexed by this number/8 (using TimeOut macro). Times
146 * in this table are for the workstation; server timeouts, add
149 static int TimeOuts[] = {
150 /* Note: don't make the first entry larger than 4 hours (see above) */
151 4 * 60 * 60, /* 0-7 users */
152 1 * 60 * 60, /* 8-15 users */
153 30 * 60, /* 16-23 users */
154 15 * 60, /* 24-31 users */
155 15 * 60, /* 32-39 users */
156 10 * 60, /* 40-47 users */
157 10 * 60, /* 48-55 users */
158 10 * 60, /* 56-63 users */
159 }; /* Anything more: MinTimeOut */
161 /* minimum time given for a call back */
162 static int MinTimeOut = (7 * 60);
164 /* Heads of CB queues; a timeout index is 1+index into this array */
165 static afs_uint32 timeout[CB_NUM_TIMEOUT_QUEUES];
167 static afs_int32 tfirst; /* cbtime of oldest unexpired call back time queue */
170 /* 16 byte object get/free routines */
175 /* Prototypes for static routines */
176 static struct FileEntry *FindFE(register AFSFid * fid);
177 static struct CallBack *iGetCB(register int *nused);
178 static int iFreeCB(register struct CallBack *cb, register int *nused);
179 static struct FileEntry *iGetFE(register int *nused);
180 static int iFreeFE(register struct FileEntry *fe, register int *nused);
181 static int TAdd(register struct CallBack *cb, register afs_uint32 * thead);
182 static int TDel(register struct CallBack *cb);
183 static int HAdd(register struct CallBack *cb, register struct host *host);
184 static int HDel(register struct CallBack *cb);
185 static int CDel(struct CallBack *cb, int deletefe);
186 static int CDelPtr(register struct FileEntry *fe, register afs_uint32 * cbp,
188 static afs_uint32 *FindCBPtr(struct FileEntry *fe, struct host *host);
189 static int FDel(register struct FileEntry *fe);
190 static int AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead,
191 int type, int locked);
192 static void MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
193 struct AFSCBFids *afidp, struct host *xhost);
194 static int MultiBreakVolumeCallBack_r(struct host *host, int isheld,
195 struct VCBParams *parms, int deletefe);
196 static int MultiBreakVolumeCallBack(struct host *host, int isheld,
197 struct VCBParams *parms);
198 static int MultiBreakVolumeLaterCallBack(struct host *host, int isheld,
199 struct VCBParams *parms);
200 static int GetSomeSpace_r(struct host *hostp, int locked);
201 static int ClearHostCallbacks_r(struct host *hp, int locked);
203 #define GetCB() ((struct CallBack *)iGetCB(&cbstuff.nCBs))
204 #define GetFE() ((struct FileEntry *)iGetFE(&cbstuff.nFEs))
205 #define FreeCB(cb) iFreeCB((struct CallBack *)cb, &cbstuff.nCBs)
206 #define FreeFE(fe) iFreeFE((struct FileEntry *)fe, &cbstuff.nFEs)
209 /* Other protos - move out sometime */
210 void PrintCB(register struct CallBack *cb, afs_uint32 now);
212 static afs_uint32 HashTable[FEHASH_SIZE]; /* File entry hash table */
214 static struct FileEntry *
215 FindFE(register AFSFid * fid)
219 register struct FileEntry *fe;
221 hash = FEHash(fid->Volume, fid->Unique);
222 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
224 if (fe->volid == fid->Volume && fe->unique == fid->Unique
225 && fe->vnode == fid->Vnode && (fe->status & FE_LATER) != FE_LATER)
231 #ifndef INTERPRET_DUMP
233 static struct CallBack *
234 iGetCB(register int *nused)
236 register struct CallBack *ret;
238 if ((ret = CBfree)) {
239 CBfree = (struct CallBack *)(((struct object *)ret)->next);
246 iFreeCB(register struct CallBack *cb, register int *nused)
248 ((struct object *)cb)->next = (struct object *)CBfree;
254 static struct FileEntry *
255 iGetFE(register int *nused)
257 register struct FileEntry *ret;
259 if ((ret = FEfree)) {
260 FEfree = (struct FileEntry *)(((struct object *)ret)->next);
267 iFreeFE(register struct FileEntry *fe, register int *nused)
269 ((struct object *)fe)->next = (struct object *)FEfree;
275 /* Add cb to end of specified timeout list */
277 TAdd(register struct CallBack *cb, register afs_uint32 * thead)
280 (*thead) = cb->tnext = cb->tprev = cbtoi(cb);
282 register struct CallBack *thp = itocb(*thead);
284 cb->tprev = thp->tprev;
288 thp->tprev = (itocb(thp->tprev)->tnext = cbtoi(cb));
290 thp->tprev = cbtoi(cb);
293 cb->thead = ttoi(thead);
297 /* Delete call back entry from timeout list */
299 TDel(register struct CallBack *cb)
301 register afs_uint32 *thead = itot(cb->thead);
303 if (*thead == cbtoi(cb))
304 *thead = (*thead == cb->tnext ? 0 : cb->tnext);
305 if (itocb(cb->tprev))
306 itocb(cb->tprev)->tnext = cb->tnext;
307 if (itocb(cb->tnext))
308 itocb(cb->tnext)->tprev = cb->tprev;
312 /* Add cb to end of specified host list */
314 HAdd(register struct CallBack *cb, register struct host *host)
316 cb->hhead = h_htoi(host);
318 host->cblist = cb->hnext = cb->hprev = cbtoi(cb);
320 register struct CallBack *fcb = itocb(host->cblist);
322 cb->hprev = fcb->hprev;
323 cb->hnext = cbtoi(fcb);
324 fcb->hprev = (itocb(fcb->hprev)->hnext = cbtoi(cb));
329 /* Delete call back entry from host list */
331 HDel(register struct CallBack *cb)
333 register afs_uint32 *hhead = &h_itoh(cb->hhead)->cblist;
335 if (*hhead == cbtoi(cb))
336 *hhead = (*hhead == cb->hnext ? 0 : cb->hnext);
337 itocb(cb->hprev)->hnext = cb->hnext;
338 itocb(cb->hnext)->hprev = cb->hprev;
342 /* Delete call back entry from fid's chain of cb's */
343 /* N.B. This one also deletes the CB, and also possibly parent FE, so
344 * make sure that it is not on any other list before calling this
347 CDel(struct CallBack *cb, int deletefe)
350 struct FileEntry *fe = itofe(cb->fhead);
351 register afs_uint32 *cbp;
354 for (safety = 0, cbp = &fe->firstcb; *cbp && *cbp != cbi;
355 cbp = &itocb(*cbp)->cnext, safety++) {
356 if (safety > cbstuff.nblks + 10) {
359 ("CDel: Internal Error -- shutting down: wanted %d from %d, now at %d\n",
360 cbi, fe->firstcb, *cbp));
362 ShutDownAndCore(PANIC);
365 CDelPtr(fe, cbp, deletefe);
369 /* Same as CDel, but pointer to parent pointer to CB entry is passed,
370 * as well as file entry */
371 /* N.B. This one also deletes the CB, and also possibly parent FE, so
372 * make sure that it is not on any other list before calling this
374 static int Ccdelpt = 0, CcdelB = 0;
377 CDelPtr(register struct FileEntry *fe, register afs_uint32 * cbp,
380 register struct CallBack *cb;
390 if (deletefe && (--fe->ncbs == 0))
396 FindCBPtr(struct FileEntry *fe, struct host *host)
398 register afs_uint32 hostindex = h_htoi(host);
399 register struct CallBack *cb;
400 register afs_uint32 *cbp;
403 for (safety = 0, cbp = &fe->firstcb; *cbp; cbp = &cb->cnext, safety++) {
404 if (safety > cbstuff.nblks) {
405 ViceLog(0, ("FindCBPtr: Internal Error -- shutting down.\n"));
407 ShutDownAndCore(PANIC);
410 if (cb->hhead == hostindex)
416 /* Delete file entry from hash table */
418 FDel(register struct FileEntry *fe)
420 register int fei = fetoi(fe);
421 register afs_uint32 *p = &HashTable[FEHash(fe->volid, fe->unique)];
423 while (*p && *p != fei)
424 p = &itofe(*p)->fnext;
431 /* initialize the callback package */
433 InitCallBack(int nblks)
436 tfirst = CBtime(FT_ApproxTime());
437 /* N.B. The "-1", below, is because
438 * FE[0] and CB[0] are not used--and not allocated */
439 FE = ((struct FileEntry *)(calloc(nblks, sizeof(struct FileEntry))));
441 ViceLog(0, ("Failed malloc in InitCallBack\n"));
444 FE--; /* FE[0] is supposed to point to junk */
445 cbstuff.nFEs = nblks;
447 FreeFE(&FE[cbstuff.nFEs]); /* This is correct */
448 CB = ((struct CallBack *)(calloc(nblks, sizeof(struct CallBack))));
450 ViceLog(0, ("Failed malloc in InitCallBack\n"));
453 CB--; /* CB[0] is supposed to point to junk */
454 cbstuff.nCBs = nblks;
456 FreeCB(&CB[cbstuff.nCBs]); /* This is correct */
457 cbstuff.nblks = nblks;
458 cbstuff.nbreakers = 0;
464 XCallBackBulk_r(struct host * ahost, struct AFSFid * fids, afs_int32 nfids)
466 struct AFSCallBack tcbs[AFSCBMAX];
472 struct rx_connection *cb_conn = NULL;
475 rx_SetConnDeadTime(ahost->callback_rxcon, 4);
476 rx_SetConnHardDeadTime(ahost->callback_rxcon, AFS_HARDDEADTIME);
483 for (i = 0; i < nfids && i < AFSCBMAX; i++) {
484 tcbs[i].CallBackVersion = CALLBACK_VERSION;
485 tcbs[i].ExpirationTime = 0;
486 tcbs[i].CallBackType = CB_DROPPED;
488 tf.AFSCBFids_len = i;
489 tf.AFSCBFids_val = &(fids[j]);
493 tc.AFSCBs_val = tcbs;
495 cb_conn = ahost->callback_rxcon;
496 rx_GetConnection(cb_conn);
498 code |= RXAFSCB_CallBack(cb_conn, &tf, &tc);
499 rx_PutConnection(cb_conn);
507 /* the locked flag tells us if the host entry has already been locked
508 * by our parent. I don't think anybody actually calls us with the
509 * host locked, but here's how to make that work: GetSomeSpace has to
510 * change so that it doesn't attempt to lock any hosts < "host". That
511 * means that it might be unable to free any objects, so it has to
512 * return an exit status. If it fails, then AddCallBack1 might fail,
513 * as well. If so, the host->ResetDone should probably be set to 0,
514 * and we probably don't want to return a callback promise to the
515 * cache manager, either. */
517 AddCallBack1(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
525 retVal = AddCallBack1_r(host, fid, thead, type, 1);
535 AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
538 struct FileEntry *fe;
539 struct CallBack *cb = 0, *lastcb = 0;
540 struct FileEntry *newfe = 0;
542 afs_uint32 *Thead = thead;
543 struct CallBack *newcb = 0;
548 /* allocate these guys first, since we can't call the allocator with
549 * the host structure locked -- or we might deadlock. However, we have
550 * to avoid races with FindFE... */
551 while (!(newcb = GetCB())) {
552 GetSomeSpace_r(host, locked);
554 while (!(newfe = GetFE())) { /* Get it now, so we don't have to call */
555 /* GetSomeSpace with the host locked, later. This might turn out to */
556 /* have been unneccessary, but that's actually kind of unlikely, since */
557 /* most files are not shared. */
558 GetSomeSpace_r(host, locked);
562 h_Lock_r(host); /* this can yield, so do it before we get any */
567 if (type == CB_NORMAL) {
569 TimeCeiling(FT_ApproxTime() + TimeOut(fe ? fe->ncbs : 0) +
571 Thead = THead(CBtime(time_out));
572 } else if (type == CB_VOLUME) {
573 time_out = TimeCeiling((60 * 120 + FT_ApproxTime()) + ServerBias);
574 Thead = THead(CBtime(time_out));
575 } else if (type == CB_BULK) {
576 /* bulk status can get so many callbacks all at once, and most of them
577 * are probably not for things that will be used for long.
580 TimeCeiling(FT_ApproxTime() + ServerBias +
581 TimeOut(22 + (fe ? fe->ncbs : 0)));
582 Thead = THead(CBtime(time_out));
588 register afs_uint32 hash;
593 fe->volid = fid->Volume;
594 fe->vnode = fid->Vnode;
595 fe->unique = fid->Unique;
598 hash = FEHash(fid->Volume, fid->Unique);
599 fe->fnext = HashTable[hash];
600 HashTable[hash] = fetoi(fe);
602 for (safety = 0, lastcb = cb = itocb(fe->firstcb); cb;
603 lastcb = cb, cb = itocb(cb->cnext), safety++) {
604 if (safety > cbstuff.nblks) {
605 ViceLog(0, ("AddCallBack1: Internal Error -- shutting down.\n"));
607 ShutDownAndCore(PANIC);
609 if (cb->hhead == h_htoi(host))
612 if (cb) { /* Already have call back: move to new timeout list */
613 /* don't change delayed callbacks back to normal ones */
614 if (cb->status != CB_DELAYED)
616 /* Only move if new timeout is longer */
617 if (TNorm(ttoi(Thead)) > TNorm(cb->thead)) {
621 if (newfe == NULL) { /* we are using the new FE */
622 fe->firstcb = cbtoi(cb);
624 cb->fhead = fetoi(fe);
629 *(lastcb ? &lastcb->cnext : &fe->firstcb) = cbtoi(cb);
632 cb->fhead = fetoi(fe);
638 /* now free any still-unused callback or host entries */
644 if (!locked) /* freecb and freefe might(?) yield */
647 if (type == CB_NORMAL || type == CB_VOLUME || type == CB_BULK)
648 return time_out - ServerBias; /* Expires sooner at workstation */
653 /* Take an array full of hosts, all held. Break callbacks to them, and
654 * release the holds once you're done, except don't release xhost. xhost
655 * may be NULL. Currently only works for a single Fid in afidp array.
656 * If you want to make this work with multiple fids, you need to fix
657 * the error handling. One approach would be to force a reset if a
658 * multi-fid call fails, or you could add delayed callbacks for each
659 * fid. You probably also need to sort and remove duplicate hosts.
660 * When this is called from the BreakVolumeCallBacks path, it does NOT
661 * force a reset if the RPC fails, it just marks the host down and tries
662 * to create a delayed callback. */
663 /* N.B. be sure that code works when ncbas == 0 */
664 /* N.B. requires all the cba[*].hp pointers to be valid... */
665 /* This routine does not hold a lock on the host for the duration of
666 * the BreakCallBack RPC, which is a significant deviation from tradition.
667 * It _does_ get a lock on the host before setting VenusDown = 1,
668 * which is sufficient only if VenusDown = 0 only happens when the
669 * lock is held over the RPC and the subsequent VenusDown == 0
670 * wherever that is done. */
672 MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
673 struct AFSCBFids *afidp, struct host *xhost)
676 struct rx_connection *conns[MAX_CB_HOSTS];
677 static struct AFSCBs tc = { 0, 0 };
678 int multi_to_cba_map[MAX_CB_HOSTS];
680 assert(ncbas <= MAX_CB_HOSTS);
682 /* set up conns for multi-call */
683 for (i = 0, j = 0; i < ncbas; i++) {
684 struct host *thishost = cba[i].hp;
685 if (!thishost || (thishost->hostFlags & HOSTDELETED)) {
688 rx_GetConnection(thishost->callback_rxcon);
689 multi_to_cba_map[j] = i;
690 conns[j++] = thishost->callback_rxcon;
693 rx_SetConnDeadTime(thishost->callback_rxcon, 4);
694 rx_SetConnHardDeadTime(thishost->callback_rxcon, AFS_HARDDEADTIME);
698 if (j) { /* who knows what multi would do with 0 conns? */
702 multi_RXAFSCB_CallBack(afidp, &tc);
708 i = multi_to_cba_map[multi_i];
714 ("BCB: INTERNAL ERROR: hp=%x, cba=%x, thead=%u\n",
718 ** try breaking callbacks on alternate interface addresses
720 if (MultiBreakCallBackAlternateAddress(hp, afidp)) {
723 ("BCB: Failed on file %u.%u.%u, Host %x (%s:%d) is down\n",
724 afidp->AFSCBFids_val->Volume,
725 afidp->AFSCBFids_val->Vnode,
726 afidp->AFSCBFids_val->Unique,
728 afs_inet_ntoa_r(hp->host, hoststr),
734 hp->hostFlags |= VENUSDOWN;
736 * We always go into AddCallBack1_r with the host locked
738 AddCallBack1_r(hp, afidp->AFSCBFids_val, itot(idx),
751 for (i = 0; i < ncbas; i++) {
754 if (hp && xhost != hp) {
759 /* H_UNLOCK around this so h_FreeConnection does not deadlock.
760 h_FreeConnection should *never* be called on a callback connection,
761 but on 10/27/04 a deadlock occurred where it was, when we know why,
762 this should be reverted. -- shadow */
764 for (i = 0; i < j; i++) {
765 rx_PutConnection(conns[i]);
773 * Break all call backs for fid, except for the specified host (unless flag
774 * is true, in which case all get a callback message. Assumption: the specified
775 * host is h_Held, by the caller; the others aren't.
776 * Specified host may be bogus, that's ok. This used to check to see if the
777 * host was down in two places, once right after the host was h_held, and
778 * again after it was locked. That race condition is incredibly rare and
779 * relatively harmless even when it does occur, so we don't check for it now.
781 /* if flag is true, send a break callback msg to "host", too */
783 BreakCallBack(struct host *xhost, AFSFid * fid, int flag)
785 struct FileEntry *fe;
786 struct CallBack *cb, *nextcb;
787 struct cbstruct cba[MAX_CB_HOSTS];
794 ("BCB: BreakCallBack(Host %x all but %s:%d, (%u,%u,%u))\n",
795 xhost, afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
796 fid->Volume, fid->Vnode, fid->Unique));
799 cbstuff.BreakCallBacks++;
804 hostindex = h_htoi(xhost);
805 cb = itocb(fe->firstcb);
806 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
807 /* the most common case is what follows the || */
810 tf.AFSCBFids_len = 1;
811 tf.AFSCBFids_val = fid;
814 for (ncbas = 0; cb && ncbas < MAX_CB_HOSTS; cb = nextcb) {
815 nextcb = itocb(cb->cnext);
816 if ((cb->hhead != hostindex || flag)
817 && (cb->status == CB_BULK || cb->status == CB_NORMAL
818 || cb->status == CB_VOLUME)) {
819 struct host *thishost = h_itoh(cb->hhead);
821 ViceLog(0, ("BCB: BOGUS! cb->hhead is NULL!\n"));
822 } else if (thishost->hostFlags & VENUSDOWN) {
824 ("BCB: %x (%s:%d) is down; delaying break call back\n",
825 thishost, afs_inet_ntoa_r(thishost->host, hoststr),
826 ntohs(thishost->port)));
827 cb->status = CB_DELAYED;
830 cba[ncbas].hp = thishost;
831 cba[ncbas].thead = cb->thead;
835 CDel(cb, 1); /* Usually first; so this delete
836 * is reasonably inexpensive */
842 MultiBreakCallBack_r(cba, ncbas, &tf, xhost);
844 /* we need to to all these initializations again because MultiBreakCallBack may block */
849 cb = itocb(fe->firstcb);
850 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
851 /* the most common case is what follows the || */
862 /* Delete (do not break) single call back for fid */
864 DeleteCallBack(struct host *host, AFSFid * fid)
866 register struct FileEntry *fe;
867 register afs_uint32 *pcb;
870 cbstuff.DeleteCallBacks++;
879 ("DCB: No call backs for fid (%u, %u, %u)\n", fid->Volume,
880 fid->Vnode, fid->Unique));
883 pcb = FindCBPtr(fe, host);
886 ("DCB: No call back for host %x (%s:%d), (%u, %u, %u)\n",
887 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
888 fid->Volume, fid->Vnode, fid->Unique));
902 * Delete (do not break) all call backs for fid. This call doesn't
903 * set all of the various host locks, but it shouldn't really matter
904 * since we're not adding callbacks, but deleting them. I'm not sure
905 * why it doesn't set the lock, however; perhaps it should.
908 DeleteFileCallBacks(AFSFid * fid)
910 register struct FileEntry *fe;
911 register struct CallBack *cb;
912 register afs_uint32 cbi;
916 cbstuff.DeleteFiles++;
921 ("DF: No fid (%u,%u,%u) to delete\n", fid->Volume, fid->Vnode,
925 for (n = 0, cbi = fe->firstcb; cbi; n++) {
937 /* Delete (do not break) all call backs for host. The host should be
940 DeleteAllCallBacks_r(struct host *host, int deletefe)
942 register struct CallBack *cb;
943 register int cbi, first;
945 cbstuff.DeleteAllCallBacks++;
946 cbi = first = host->cblist;
948 ViceLog(8, ("DV: no call backs\n"));
956 } while (cbi != first);
962 * Break all delayed call backs for host. Returns 1 if all call backs
963 * successfully broken; 0 otherwise. Assumes host is h_Held and h_Locked.
964 * Must be called with VenusDown set for this host
967 BreakDelayedCallBacks(struct host *host)
971 retVal = BreakDelayedCallBacks_r(host);
977 BreakDelayedCallBacks_r(struct host *host)
979 struct AFSFid fids[AFSCBMAX];
980 u_byte thead[AFSCBMAX]; /* This should match thead in struct Callback */
981 int cbi, first, nfids;
985 struct rx_connection *cb_conn;
988 if (!(host->hostFlags & RESETDONE) && !(host->hostFlags & HOSTDELETED)) {
989 host->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
990 cb_conn = host->callback_rxcon;
991 rx_GetConnection(cb_conn);
992 if (host->interface) {
995 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
998 code = RXAFSCB_InitCallBackState(cb_conn);
1000 rx_PutConnection(cb_conn);
1003 host->hostFlags |= ALTADDR; /* alternate addresses are valid */
1007 ("CB: Call back connect back failed (in break delayed) for Host %x (%s:%d)\n",
1008 host, afs_inet_ntoa_r(host->host, hoststr),
1009 ntohs(host->port)));
1011 host->hostFlags |= VENUSDOWN;
1014 ("InitCallBackState success on %x (%s:%d)\n",
1015 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
1016 /* reset was done successfully */
1017 host->hostFlags |= RESETDONE;
1018 host->hostFlags &= ~VENUSDOWN;
1021 while (!(host->hostFlags & HOSTDELETED)) {
1023 host->hostFlags &= ~VENUSDOWN; /* presume up */
1024 cbi = first = host->cblist;
1028 first = host->cblist;
1031 if (cb->status == CB_DELAYED) {
1032 register struct FileEntry *fe = itofe(cb->fhead);
1033 thead[nfids] = cb->thead;
1034 fids[nfids].Volume = fe->volid;
1035 fids[nfids].Vnode = fe->vnode;
1036 fids[nfids].Unique = fe->unique;
1042 } while (cbi && cbi != first && nfids < AFSCBMAX);
1048 if (XCallBackBulk_r(host, fids, nfids)) {
1049 /* Failed, again: put them back, probably with old
1054 ("CB: XCallBackBulk failed, Host %x (%s:%d); callback list follows:\n",
1055 host, afs_inet_ntoa_r(host->host, hoststr),
1056 ntohs(host->port)));
1058 for (i = 0; i < nfids; i++) {
1061 ("CB: Host %x (%s:%d), file %u.%u.%u (part of bulk callback)\n",
1062 host, afs_inet_ntoa_r(host->host, hoststr),
1063 ntohs(host->port), fids[i].Volume,
1064 fids[i].Vnode, fids[i].Unique));
1067 * AddCallBack1_r(host, &fids[i], itot(thead[i]), CB_DELAYED, 1);
1068 * * but it turns out to cause too many tricky locking problems.
1069 * * now, if break delayed fails, screw it. */
1071 host->hostFlags |= VENUSDOWN; /* Failed */
1072 ClearHostCallbacks_r(host, 1 /* locked */ );
1076 if (nfids < AFSCBMAX)
1080 cbstuff.nbreakers--;
1081 /* If we succeeded it's always ok to unset HFE_LATER */
1082 if (!(host->hostFlags & VENUSDOWN))
1083 host->hostFlags &= ~HFE_LATER;
1084 return (host->hostFlags & VENUSDOWN);
1088 ** isheld is 0 if the host is held in h_Enumerate
1089 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1092 MultiBreakVolumeCallBack_r(struct host *host, int isheld,
1093 struct VCBParams *parms, int deletefe)
1098 return isheld; /* host is held only by h_Enumerate, do nothing */
1099 if (host->hostFlags & HOSTDELETED)
1100 return 0; /* host is deleted, release hold */
1102 if (host->hostFlags & VENUSDOWN) {
1104 if (host->hostFlags & HOSTDELETED) {
1106 return 0; /* Release hold */
1109 ("BVCB: volume call back for Host %x (%s:%d) failed\n",
1110 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
1113 ("CB: volume callback for Host %x (%s:%d) failed\n",
1114 host, afs_inet_ntoa_r(host->host, hoststr),
1115 ntohs(host->port)));
1117 DeleteAllCallBacks_r(host, deletefe); /* Delete all callback state
1118 * rather than attempting to
1119 * selectively remember to
1120 * delete the volume callbacks
1122 host->hostFlags &= ~RESETDONE; /* Do InitCallBackState when host returns */
1124 return 0; /* release hold */
1126 assert(parms->ncbas <= MAX_CB_HOSTS);
1128 /* Do not call MultiBreakCallBack on the current host structure
1129 ** because it would prematurely release the hold on the host
1131 if (parms->ncbas == MAX_CB_HOSTS) {
1132 struct AFSCBFids tf;
1134 tf.AFSCBFids_len = 1;
1135 tf.AFSCBFids_val = parms->fid;
1137 /* this releases all the hosts */
1138 MultiBreakCallBack_r(parms->cba, parms->ncbas, &tf, 0 /* xhost */ );
1142 parms->cba[parms->ncbas].hp = host;
1143 parms->cba[(parms->ncbas)++].thead = parms->thead;
1144 return 1; /* DON'T release hold, because we still need it. */
1148 ** isheld is 0 if the host is held in h_Enumerate
1149 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1152 MultiBreakVolumeCallBack(struct host *host, int isheld,
1153 struct VCBParams *parms)
1157 retval = MultiBreakVolumeCallBack_r(host, isheld, parms, 1);
1163 ** isheld is 0 if the host is held in h_Enumerate
1164 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1167 MultiBreakVolumeLaterCallBack(struct host *host, int isheld,
1168 struct VCBParams *parms)
1172 retval = MultiBreakVolumeCallBack_r(host, isheld, parms, 0);
1178 * Break all call backs on a single volume. Don't call this with any
1179 * hosts h_held. Note that this routine clears the callbacks before
1180 * actually breaking them, and that the vnode isn't locked during this
1181 * operation, so that people might see temporary callback loss while
1182 * this function is executing. It is just a temporary state, however,
1183 * since the callback will be broken later by this same function.
1185 * Now uses multi-RX for CallBack RPC. Note that the
1186 * multiBreakCallBacks routine does not force a reset if the RPC
1187 * fails, unlike the previous version of this routine, but does create
1188 * a delayed callback. Resets will be forced if the host is
1189 * determined to be down before the RPC is executed.
1192 BreakVolumeCallBacks(afs_uint32 volume)
1197 struct CallBack *cb;
1198 struct FileEntry *fe;
1200 struct VCBParams henumParms;
1201 afs_uint32 tthead = 0; /* zero is illegal value */
1204 fid.Volume = volume, fid.Vnode = fid.Unique = 0;
1205 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1206 for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
1207 if (fe->volid == volume) {
1208 register struct CallBack *cbnext;
1209 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1210 host = h_itoh(cb->hhead);
1212 cbnext = itocb(cb->cnext);
1213 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1219 /* leave hold for MultiBreakVolumeCallBack to clear */
1230 /* didn't find any callbacks, so return right away. */
1234 henumParms.ncbas = 0;
1235 henumParms.fid = &fid;
1236 henumParms.thead = tthead;
1238 h_Enumerate(MultiBreakVolumeCallBack, (char *)&henumParms);
1240 if (henumParms.ncbas) { /* do left-overs */
1241 struct AFSCBFids tf;
1242 tf.AFSCBFids_len = 1;
1243 tf.AFSCBFids_val = &fid;
1245 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf, 0);
1247 henumParms.ncbas = 0;
1253 #ifdef AFS_PTHREAD_ENV
1254 extern pthread_cond_t fsync_cond;
1256 extern char fsync_wait[];
1260 BreakVolumeCallBacksLater(afs_uint32 volume)
1264 struct FileEntry *fe;
1265 struct CallBack *cb;
1269 ViceLog(25, ("Setting later on volume %u\n", volume));
1271 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1272 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1273 if (fe->volid == volume) {
1274 register struct CallBack *cbnext;
1275 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1276 host = h_itoh(cb->hhead);
1277 host->hostFlags |= HFE_LATER;
1278 cb->status = CB_DELAYED;
1279 cbnext = itocb(cb->cnext);
1282 fe->status |= FE_LATER;
1291 /* didn't find any callbacks, so return right away. */
1295 ViceLog(25, ("Fsync thread wakeup\n"));
1296 #ifdef AFS_PTHREAD_ENV
1298 assert(pthread_cond_broadcast(&fsync_cond) == 0);
1301 LWP_NoYieldSignal(fsync_wait);
1307 BreakLaterCallBacks(void)
1312 struct CallBack *cb;
1313 struct FileEntry *fe = NULL;
1314 struct FileEntry *myfe = NULL;
1316 struct VCBParams henumParms;
1317 unsigned short tthead = 0; /* zero is illegal value */
1321 ViceLog(25, ("Looking for FileEntries to unchain\n"));
1324 /* Pick the first volume we see to clean up */
1325 fid.Volume = fid.Vnode = fid.Unique = 0;
1327 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1328 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1329 if (fe && (fe->status & FE_LATER)
1330 && (fid.Volume == 0 || fid.Volume == fe->volid)) {
1331 /* Ugly, but used to avoid left side casting */
1332 struct object *tmpfe;
1334 ("Unchaining for %u:%u:%u\n", fe->vnode, fe->unique,
1336 fid.Volume = fe->volid;
1338 /* Works since volid is deeper than the largest pointer */
1339 tmpfe = (struct object *)fe;
1340 tmpfe->next = (struct object *)myfe;
1353 /* loop over FEs from myfe and free/break */
1355 for (fe = myfe; fe;) {
1356 register struct CallBack *cbnext;
1357 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1358 cbnext = itocb(cb->cnext);
1359 host = h_itoh(cb->hhead);
1360 if (cb->status == CB_DELAYED) {
1362 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1367 CDel(cb, 0); /* Don't let CDel clean up the fe */
1368 /* leave hold for MultiBreakVolumeCallBack to clear */
1371 ("Found host %x (%s:%d) non-DELAYED cb for %u:%u:%u\n",
1372 host, afs_inet_ntoa_r(host->host, hoststr),
1373 ntohs(host->port), fe->vnode, fe->unique, fe->volid));
1377 fe = (struct FileEntry *)((struct object *)fe)->next;
1382 ViceLog(125, ("Breaking volume %u\n", fid.Volume));
1383 henumParms.ncbas = 0;
1384 henumParms.fid = &fid;
1385 henumParms.thead = tthead;
1387 h_Enumerate(MultiBreakVolumeLaterCallBack, (char *)&henumParms);
1389 if (henumParms.ncbas) { /* do left-overs */
1390 struct AFSCBFids tf;
1391 tf.AFSCBFids_len = 1;
1392 tf.AFSCBFids_val = &fid;
1394 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf, 0);
1395 henumParms.ncbas = 0;
1400 /* Arrange to be called again */
1405 * Delete all timed-out call back entries (to be called periodically by file
1409 CleanupTimedOutCallBacks(void)
1412 CleanupTimedOutCallBacks_r();
1418 CleanupTimedOutCallBacks_r(void)
1420 afs_uint32 now = CBtime(FT_ApproxTime());
1421 register afs_uint32 *thead;
1422 register struct CallBack *cb;
1423 register int ntimedout = 0;
1426 while (tfirst <= now) {
1428 cbi = *(thead = THead(tfirst));
1434 ("CCB: deleting timed out call back %x (%s:%d), (%u,%u,%u)\n",
1435 h_itoh(cb->hhead)->host,
1436 afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
1437 h_itoh(cb->hhead)->port, itofe(cb->fhead)->volid,
1438 itofe(cb->fhead)->vnode, itofe(cb->fhead)->unique));
1442 if (ntimedout > cbstuff.nblks) {
1443 ViceLog(0, ("CCB: Internal Error -- shutting down...\n"));
1444 DumpCallBackState();
1445 ShutDownAndCore(PANIC);
1447 } while (cbi != *thead);
1452 cbstuff.CBsTimedOut += ntimedout;
1453 ViceLog(7, ("CCB: deleted %d timed out callbacks\n", ntimedout));
1454 return (ntimedout > 0);
1457 static struct host *lih_host;
1458 static int lih_host_held;
1460 /* This version does not allow 'host' to be selected unless its ActiveCall
1461 * is newer than 'hostp' which is the host with the oldest ActiveCall from
1462 * the last pass (if it is provided). We filter out any hosts that are
1463 * are held by other threads.
1466 lih0_r(register struct host *host, register int held,
1467 register struct host *hostp)
1470 && (hostp && host != hostp)
1471 && (!held && !h_OtherHolds_r(host))
1472 && (!lih_host || host->ActiveCall < lih_host->ActiveCall)
1473 && (!hostp || host->ActiveCall > hostp->ActiveCall)) {
1474 if (lih_host != NULL && lih_host_held) {
1475 h_Release_r(lih_host);
1478 lih_host_held = !held;
1484 /* This version does not allow 'host' to be selected unless its ActiveCall
1485 * is newer than 'hostp' which is the host with the oldest ActiveCall from
1486 * the last pass (if it is provided). In this second varient, we do not
1487 * prevent held hosts from being selected.
1490 lih1_r(register struct host *host, register int held,
1491 register struct host *hostp)
1494 && (hostp && host != hostp)
1495 && (!lih_host || host->ActiveCall < lih_host->ActiveCall)
1496 && (!hostp || host->ActiveCall > hostp->ActiveCall)) {
1497 if (lih_host != NULL && lih_host_held) {
1498 h_Release_r(lih_host);
1501 lih_host_held = !held;
1507 /* This could be upgraded to get more space each time */
1508 /* first pass: sequentially find the oldest host which isn't held by
1509 anyone for which we can clear callbacks;
1511 /* second pass: sequentially find the oldest host regardless of
1512 whether or not the host is held; skipping 'hostp' */
1513 /* third pass: attempt to clear callbacks from 'hostp' */
1514 /* always called with hostp unlocked */
1516 /* Note: hostlist is ordered most recently created host first and
1517 * its order has no relationship to the most recently used. */
1518 extern struct host *hostList;
1520 GetSomeSpace_r(struct host *hostp, int locked)
1522 register struct host *hp, *hp1, *hp2;
1525 cbstuff.GotSomeSpaces++;
1527 ("GSS: First looking for timed out call backs via CleanupCallBacks\n"));
1528 if (CleanupTimedOutCallBacks_r()) {
1538 h_Enumerate_r(i == 0 ? lih0_r : lih1_r, hp2, (char *)hp1);
1541 /* set in lih_r! private copy before giving up H_LOCK */
1542 int lih_host_held2=lih_host_held;
1544 if ((hp != hostp) && !ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
1555 * Next time try getting callbacks from any host even if
1556 * it's deleted (that's actually great since we can freely
1557 * remove its callbacks) or it's held since the only other
1558 * option is starvation for the file server (i.e. until the
1559 * callback timeout arrives).
1566 ("GSS: Try harder for longest inactive host cnt= %d\n",
1571 /* Could not obtain space from other hosts, clear hostp's callback state */
1576 ClearHostCallbacks_r(hostp, 1 /*already locked */ );
1583 /* locked - set if caller has already locked the host */
1585 ClearHostCallbacks_r(struct host *hp, int locked)
1590 struct rx_connection *cb_conn = NULL;
1593 ("GSS: Delete longest inactive host %x (%s:%d)\n",
1594 hp, afs_inet_ntoa_r(hp->host, hoststr), ntohs(hp->port)));
1595 if (!(held = h_Held_r(hp)))
1598 /** Try a non-blocking lock. If the lock is already held return
1599 * after releasing hold on hp
1602 if (h_NBLock_r(hp)) {
1608 if (hp->Console & 2) {
1610 * If the special console field is set it means that a thread
1611 * is waiting in AddCallBack1 after it set pointers to the
1612 * file entry and/or callback entry. Because of the bogus
1613 * usage of h_hold it won't prevent from another thread, this
1614 * one, to remove all the callbacks so just to be safe we keep
1615 * a reference. NOTE, on the last phase we'll free the calling
1616 * host's callbacks but that's ok...
1620 DeleteAllCallBacks_r(hp, 1);
1621 if (hp->hostFlags & VENUSDOWN) {
1622 hp->hostFlags &= ~RESETDONE; /* remember that we must do a reset */
1624 /* host is up, try a call */
1625 hp->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1626 cb_conn = hp->callback_rxcon;
1627 rx_GetConnection(hp->callback_rxcon);
1628 if (hp->interface) {
1631 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1634 code = RXAFSCB_InitCallBackState(cb_conn);
1636 rx_PutConnection(cb_conn);
1639 hp->hostFlags |= ALTADDR; /* alternate addresses are valid */
1641 /* failed, mark host down and need reset */
1642 hp->hostFlags |= VENUSDOWN;
1643 hp->hostFlags &= ~RESETDONE;
1645 /* reset succeeded, we're done */
1646 hp->hostFlags |= RESETDONE;
1657 #endif /* INTERPRET_DUMP */
1661 PrintCallBackStats(void)
1664 "%d add CB, %d break CB, %d del CB, %d del FE, %d CB's timed out, %d space reclaim, %d del host\n",
1665 cbstuff.AddCallBacks, cbstuff.BreakCallBacks,
1666 cbstuff.DeleteCallBacks, cbstuff.DeleteFiles, cbstuff.CBsTimedOut,
1667 cbstuff.GotSomeSpaces, cbstuff.DeleteAllCallBacks);
1668 fprintf(stderr, "%d CBs, %d FEs, (%d of total of %d 16-byte blocks)\n",
1669 cbstuff.nCBs, cbstuff.nFEs, cbstuff.nCBs + cbstuff.nFEs,
1675 #define MAGIC 0x12345678 /* To check byte ordering of dump when it is read in */
1677 #ifndef INTERPRET_DUMP
1679 #ifdef AFS_DEMAND_ATTACH_FS
1682 * callback state serialization
1684 static int cb_stateSaveTimeouts(struct fs_dump_state * state);
1685 static int cb_stateSaveFEHash(struct fs_dump_state * state);
1686 static int cb_stateSaveFEs(struct fs_dump_state * state);
1687 static int cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe);
1688 static int cb_stateRestoreTimeouts(struct fs_dump_state * state);
1689 static int cb_stateRestoreFEHash(struct fs_dump_state * state);
1690 static int cb_stateRestoreFEs(struct fs_dump_state * state);
1691 static int cb_stateRestoreFE(struct fs_dump_state * state);
1692 static int cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
1693 struct iovec * iov, int niovecs);
1695 static int cb_stateVerifyFEHash(struct fs_dump_state * state);
1696 static int cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe);
1697 static int cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe);
1698 static int cb_stateVerifyTimeoutQueues(struct fs_dump_state * state);
1700 static int cb_stateFEToDiskEntry(struct FileEntry *, struct FEDiskEntry *);
1701 static int cb_stateDiskEntryToFE(struct fs_dump_state * state,
1702 struct FEDiskEntry *, struct FileEntry *);
1704 static int cb_stateCBToDiskEntry(struct CallBack *, struct CBDiskEntry *);
1705 static int cb_stateDiskEntryToCB(struct fs_dump_state * state,
1706 struct CBDiskEntry *, struct CallBack *);
1708 static int cb_stateFillHeader(struct callback_state_header * hdr);
1709 static int cb_stateCheckHeader(struct callback_state_header * hdr);
1711 static int cb_stateAllocMap(struct fs_dump_state * state);
1714 cb_stateSave(struct fs_dump_state * state)
1718 AssignInt64(state->eof_offset, &state->hdr->cb_offset);
1720 /* invalidate callback state header */
1721 memset(state->cb_hdr, 0, sizeof(struct callback_state_header));
1722 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1723 sizeof(struct callback_state_header))) {
1728 fs_stateIncEOF(state, sizeof(struct callback_state_header));
1730 /* dump timeout state */
1731 if (cb_stateSaveTimeouts(state)) {
1736 /* dump fe hashtable state */
1737 if (cb_stateSaveFEHash(state)) {
1742 /* dump callback state */
1743 if (cb_stateSaveFEs(state)) {
1748 /* write the callback state header to disk */
1749 cb_stateFillHeader(state->cb_hdr);
1750 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1751 sizeof(struct callback_state_header))) {
1761 cb_stateRestore(struct fs_dump_state * state)
1765 if (fs_stateReadHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1766 sizeof(struct callback_state_header))) {
1771 if (cb_stateCheckHeader(state->cb_hdr)) {
1776 if (cb_stateAllocMap(state)) {
1781 if (cb_stateRestoreTimeouts(state)) {
1786 if (cb_stateRestoreFEHash(state)) {
1791 /* restore FEs and CBs from disk */
1792 if (cb_stateRestoreFEs(state)) {
1797 /* restore the timeout queue heads */
1798 tfirst = state->cb_hdr->tfirst;
1805 cb_stateRestoreIndices(struct fs_dump_state * state)
1808 struct FileEntry * fe;
1809 struct CallBack * cb;
1811 /* restore indices in the FileEntry structures */
1812 for (i = 1; i < state->fe_map.len; i++) {
1813 if (state->fe_map.entries[i].new_idx) {
1814 fe = itofe(state->fe_map.entries[i].new_idx);
1816 /* restore the fe->fnext entry */
1817 if (fe_OldToNew(state, fe->fnext, &fe->fnext)) {
1822 /* restore the fe->firstcb entry */
1823 if (cb_OldToNew(state, fe->firstcb, &fe->firstcb)) {
1830 /* restore indices in the CallBack structures */
1831 for (i = 1; i < state->cb_map.len; i++) {
1832 if (state->cb_map.entries[i].new_idx) {
1833 cb = itocb(state->cb_map.entries[i].new_idx);
1835 /* restore the cb->cnext entry */
1836 if (cb_OldToNew(state, cb->cnext, &cb->cnext)) {
1841 /* restore the cb->fhead entry */
1842 if (fe_OldToNew(state, cb->fhead, &cb->fhead)) {
1847 /* restore the cb->hhead entry */
1848 if (h_OldToNew(state, cb->hhead, &cb->hhead)) {
1853 /* restore the cb->tprev entry */
1854 if (cb_OldToNew(state, cb->tprev, &cb->tprev)) {
1859 /* restore the cb->tnext entry */
1860 if (cb_OldToNew(state, cb->tnext, &cb->tnext)) {
1865 /* restore the cb->hprev entry */
1866 if (cb_OldToNew(state, cb->hprev, &cb->hprev)) {
1871 /* restore the cb->hnext entry */
1872 if (cb_OldToNew(state, cb->hnext, &cb->hnext)) {
1879 /* restore the timeout queue head indices */
1880 for (i = 0; i < state->cb_timeout_hdr->records; i++) {
1881 if (cb_OldToNew(state, timeout[i], &timeout[i])) {
1887 /* restore the FE hash table queue heads */
1888 for (i = 0; i < state->cb_fehash_hdr->records; i++) {
1889 if (fe_OldToNew(state, HashTable[i], &HashTable[i])) {
1900 cb_stateVerify(struct fs_dump_state * state)
1904 if (cb_stateVerifyFEHash(state)) {
1908 if (cb_stateVerifyTimeoutQueues(state)) {
1917 cb_stateVerifyFEHash(struct fs_dump_state * state)
1920 struct FileEntry * fe;
1921 afs_uint32 fei, chain_len;
1923 for (i = 0; i < FEHASH_SIZE; i++) {
1925 for (fei = HashTable[i], fe = itofe(fei);
1927 fei = fe->fnext, fe = itofe(fei)) {
1928 if (fei > cbstuff.nblks) {
1929 ViceLog(0, ("cb_stateVerifyFEHash: error: index out of range (fei=%d)\n", fei));
1933 if (cb_stateVerifyFE(state, fe)) {
1936 if (chain_len > FS_STATE_FE_MAX_HASH_CHAIN_LEN) {
1937 ViceLog(0, ("cb_stateVerifyFEHash: error: hash chain %d length exceeds %d; assuming there's a loop\n",
1938 i, FS_STATE_FE_MAX_HASH_CHAIN_LEN));
1951 cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe)
1955 if ((fe->firstcb && !fe->ncbs) ||
1956 (!fe->firstcb && fe->ncbs)) {
1957 ViceLog(0, ("cb_stateVerifyFE: error: fe->firstcb does not agree with fe->ncbs (fei=%d, fe->firstcb=%d, fe->ncbs=%d)\n",
1958 fetoi(fe), fe->firstcb, fe->ncbs));
1961 if (cb_stateVerifyFCBList(state, fe)) {
1962 ViceLog(0, ("cb_stateVerifyFE: error: FCBList failed verification (fei=%d)\n", fetoi(fe)));
1971 cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe)
1974 afs_uint32 cbi, fei, chain_len = 0;
1975 struct CallBack * cb;
1979 for (cbi = fe->firstcb, cb = itocb(cbi);
1981 cbi = cb->cnext, cb = itocb(cbi)) {
1982 if (cbi > cbstuff.nblks) {
1983 ViceLog(0, ("cb_stateVerifyFCBList: error: list index out of range (cbi=%d, ncbs=%d)\n",
1984 cbi, cbstuff.nblks));
1988 if (cb->fhead != fei) {
1989 ViceLog(0, ("cb_stateVerifyFCBList: error: cb->fhead != fei (fei=%d, cb->fhead=%d)\n",
1993 if (chain_len > FS_STATE_FCB_MAX_LIST_LEN) {
1994 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (fei=%d); assuming there's a loop\n",
1995 FS_STATE_FCB_MAX_LIST_LEN, fei));
2002 if (fe->ncbs != chain_len) {
2003 ViceLog(0, ("cb_stateVerifyFCBList: error: list length mismatch (len=%d, fe->ncbs=%d)\n",
2004 chain_len, fe->ncbs));
2013 cb_stateVerifyHCBList(struct fs_dump_state * state, struct host * host)
2016 afs_uint32 hi, chain_len, cbi;
2017 struct CallBack *cb, *ncb;
2022 for (cbi = host->cblist, cb = itocb(cbi);
2024 cbi = cb->hnext, cb = ncb) {
2025 if (chain_len && (host->cblist == cbi)) {
2026 /* we've wrapped around the circular list, and everything looks ok */
2029 if (cb->hhead != hi) {
2030 ViceLog(0, ("cb_stateVerifyHCBList: error: incorrect cb->hhead (cbi=%d, h->index=%d, cb->hhead=%d)\n",
2031 cbi, hi, cb->hhead));
2034 if (!cb->hprev || !cb->hnext) {
2035 ViceLog(0, ("cb_stateVerifyHCBList: error: null index in circular list (cbi=%d, h->index=%d)\n",
2040 if ((cb->hprev > cbstuff.nblks) ||
2041 (cb->hnext > cbstuff.nblks)) {
2042 ViceLog(0, ("cb_stateVerifyHCBList: error: list index out of range (cbi=%d, h->index=%d, cb->hprev=%d, cb->hnext=%d, nCBs=%d)\n",
2043 cbi, hi, cb->hprev, cb->hnext, cbstuff.nblks));
2047 ncb = itocb(cb->hnext);
2048 if (cbi != ncb->hprev) {
2049 ViceLog(0, ("cb_stateVerifyHCBList: error: corrupt linked list (cbi=%d, h->index=%d)\n",
2054 if (chain_len > FS_STATE_HCB_MAX_LIST_LEN) {
2055 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (h->index=%d); assuming there's a loop\n",
2056 FS_STATE_HCB_MAX_LIST_LEN, hi));
2068 cb_stateVerifyTimeoutQueues(struct fs_dump_state * state)
2071 afs_uint32 cbi, chain_len;
2072 struct CallBack *cb, *ncb;
2074 for (i = 0; i < CB_NUM_TIMEOUT_QUEUES; i++) {
2076 for (cbi = timeout[i], cb = itocb(cbi);
2078 cbi = cb->tnext, cb = ncb) {
2079 if (chain_len && (cbi == timeout[i])) {
2080 /* we've wrapped around the circular list, and everything looks ok */
2083 if (cbi > cbstuff.nblks) {
2084 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: list index out of range (cbi=%d, tindex=%d)\n",
2089 if (itot(cb->thead) != &timeout[i]) {
2090 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: cb->thead points to wrong timeout queue (tindex=%d, cbi=%d, cb->thead=%d)\n",
2091 i, cbi, cb->thead));
2094 if (!cb->tprev || !cb->tnext) {
2095 ViceLog(0, ("cb_stateVerifyTimeoutQueues: null index in circular list (cbi=%d, tindex=%d)\n",
2100 if ((cb->tprev > cbstuff.nblks) ||
2101 (cb->tnext > cbstuff.nblks)) {
2102 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list index out of range (cbi=%d, tindex=%d, cb->tprev=%d, cb->tnext=%d, nCBs=%d)\n",
2103 cbi, i, cb->tprev, cb->tnext, cbstuff.nblks));
2107 ncb = itocb(cb->tnext);
2108 if (cbi != ncb->tprev) {
2109 ViceLog(0, ("cb_stateVerifyTimeoutQueues: corrupt linked list (cbi=%d, tindex=%d)\n",
2114 if (chain_len > FS_STATE_TCB_MAX_LIST_LEN) {
2115 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list length exceeds %d (tindex=%d); assuming there's a loop\n",
2116 FS_STATE_TCB_MAX_LIST_LEN, i));
2129 cb_stateSaveTimeouts(struct fs_dump_state * state)
2132 struct iovec iov[2];
2134 AssignInt64(state->eof_offset, &state->cb_hdr->timeout_offset);
2136 memset(state->cb_timeout_hdr, 0, sizeof(struct callback_state_fehash_header));
2137 state->cb_timeout_hdr->magic = CALLBACK_STATE_TIMEOUT_MAGIC;
2138 state->cb_timeout_hdr->records = CB_NUM_TIMEOUT_QUEUES;
2139 state->cb_timeout_hdr->len = sizeof(struct callback_state_timeout_header) +
2140 (state->cb_timeout_hdr->records * sizeof(afs_uint32));
2142 iov[0].iov_base = (char *)state->cb_timeout_hdr;
2143 iov[0].iov_len = sizeof(struct callback_state_timeout_header);
2144 iov[1].iov_base = (char *)timeout;
2145 iov[1].iov_len = sizeof(timeout);
2147 if (fs_stateSeek(state, &state->cb_hdr->timeout_offset)) {
2152 if (fs_stateWriteV(state, iov, 2)) {
2157 fs_stateIncEOF(state, state->cb_timeout_hdr->len);
2164 cb_stateRestoreTimeouts(struct fs_dump_state * state)
2168 if (fs_stateReadHeader(state, &state->cb_hdr->timeout_offset,
2169 state->cb_timeout_hdr,
2170 sizeof(struct callback_state_timeout_header))) {
2175 if (state->cb_timeout_hdr->magic != CALLBACK_STATE_TIMEOUT_MAGIC) {
2179 if (state->cb_timeout_hdr->records != CB_NUM_TIMEOUT_QUEUES) {
2184 len = state->cb_timeout_hdr->records * sizeof(afs_uint32);
2186 if (state->cb_timeout_hdr->len !=
2187 (sizeof(struct callback_state_timeout_header) + len)) {
2192 if (fs_stateRead(state, timeout, len)) {
2202 cb_stateSaveFEHash(struct fs_dump_state * state)
2205 struct iovec iov[2];
2207 AssignInt64(state->eof_offset, &state->cb_hdr->fehash_offset);
2209 memset(state->cb_fehash_hdr, 0, sizeof(struct callback_state_fehash_header));
2210 state->cb_fehash_hdr->magic = CALLBACK_STATE_FEHASH_MAGIC;
2211 state->cb_fehash_hdr->records = FEHASH_SIZE;
2212 state->cb_fehash_hdr->len = sizeof(struct callback_state_fehash_header) +
2213 (state->cb_fehash_hdr->records * sizeof(afs_uint32));
2215 iov[0].iov_base = (char *)state->cb_fehash_hdr;
2216 iov[0].iov_len = sizeof(struct callback_state_fehash_header);
2217 iov[1].iov_base = (char *)HashTable;
2218 iov[1].iov_len = sizeof(HashTable);
2220 if (fs_stateSeek(state, &state->cb_hdr->fehash_offset)) {
2225 if (fs_stateWriteV(state, iov, 2)) {
2230 fs_stateIncEOF(state, state->cb_fehash_hdr->len);
2237 cb_stateRestoreFEHash(struct fs_dump_state * state)
2241 if (fs_stateReadHeader(state, &state->cb_hdr->fehash_offset,
2242 state->cb_fehash_hdr,
2243 sizeof(struct callback_state_fehash_header))) {
2248 if (state->cb_fehash_hdr->magic != CALLBACK_STATE_FEHASH_MAGIC) {
2252 if (state->cb_fehash_hdr->records != FEHASH_SIZE) {
2257 len = state->cb_fehash_hdr->records * sizeof(afs_uint32);
2259 if (state->cb_fehash_hdr->len !=
2260 (sizeof(struct callback_state_fehash_header) + len)) {
2265 if (fs_stateRead(state, HashTable, len)) {
2275 cb_stateSaveFEs(struct fs_dump_state * state)
2278 register int fei, hash;
2279 register struct FileEntry *fe;
2281 AssignInt64(state->eof_offset, &state->cb_hdr->fe_offset);
2283 for (hash = 0; hash < FEHASH_SIZE ; hash++) {
2284 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
2286 if (cb_stateSaveFE(state, fe)) {
2298 cb_stateRestoreFEs(struct fs_dump_state * state)
2300 int count, nFEs, ret = 0;
2302 nFEs = state->cb_hdr->nFEs;
2304 for (count = 0; count < nFEs; count++) {
2305 if (cb_stateRestoreFE(state)) {
2316 cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe)
2318 int ret = 0, iovcnt, cbi, written = 0;
2320 struct callback_state_entry_header hdr;
2321 struct FEDiskEntry fedsk;
2322 struct CBDiskEntry cbdsk[16];
2323 struct iovec iov[16];
2324 struct CallBack *cb;
2327 if (fei > state->cb_hdr->fe_max) {
2328 state->cb_hdr->fe_max = fei;
2331 memset(&hdr, 0, sizeof(struct callback_state_entry_header));
2333 if (cb_stateFEToDiskEntry(fe, &fedsk)) {
2338 iov[0].iov_base = (char *)&hdr;
2339 iov[0].iov_len = sizeof(hdr);
2340 iov[1].iov_base = (char *)&fedsk;
2341 iov[1].iov_len = sizeof(struct FEDiskEntry);
2344 for (cbi = fe->firstcb, cb = itocb(cbi);
2346 cbi = cb->cnext, cb = itocb(cbi), hdr.nCBs++) {
2347 if (cbi > state->cb_hdr->cb_max) {
2348 state->cb_hdr->cb_max = cbi;
2350 if (cb_stateCBToDiskEntry(cb, &cbdsk[iovcnt])) {
2354 cbdsk[iovcnt].index = cbi;
2355 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2356 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2358 if ((iovcnt == 16) || (!cb->cnext)) {
2359 if (fs_stateWriteV(state, iov, iovcnt)) {
2368 hdr.magic = CALLBACK_STATE_ENTRY_MAGIC;
2369 hdr.len = sizeof(hdr) + sizeof(struct FEDiskEntry) +
2370 (hdr.nCBs * sizeof(struct CBDiskEntry));
2373 if (fs_stateWriteV(state, iov, iovcnt)) {
2378 if (fs_stateWriteHeader(state, &state->eof_offset, &hdr, sizeof(hdr))) {
2384 fs_stateIncEOF(state, hdr.len);
2387 if (fs_stateSeek(state, &state->eof_offset)) {
2393 state->cb_hdr->nFEs++;
2394 state->cb_hdr->nCBs += hdr.nCBs;
2401 cb_stateRestoreFE(struct fs_dump_state * state)
2403 int ret = 0, iovcnt, nCBs;
2404 struct callback_state_entry_header hdr;
2405 struct FEDiskEntry fedsk;
2406 struct CBDiskEntry cbdsk[16];
2407 struct iovec iov[16];
2408 struct FileEntry * fe;
2409 struct CallBack * cb;
2411 iov[0].iov_base = (char *)&hdr;
2412 iov[0].iov_len = sizeof(hdr);
2413 iov[1].iov_base = (char *)&fedsk;
2414 iov[1].iov_len = sizeof(fedsk);
2417 if (fs_stateReadV(state, iov, iovcnt)) {
2422 if (hdr.magic != CALLBACK_STATE_ENTRY_MAGIC) {
2429 ViceLog(0, ("cb_stateRestoreFE: ran out of free FileEntry structures\n"));
2434 if (cb_stateDiskEntryToFE(state, &fedsk, fe)) {
2440 for (iovcnt = 0, nCBs = 0;
2443 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2444 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2446 if ((iovcnt == 16) || (nCBs == hdr.nCBs - 1)) {
2447 if (fs_stateReadV(state, iov, iovcnt)) {
2451 if (cb_stateRestoreCBs(state, fe, iov, iovcnt)) {
2465 cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
2466 struct iovec * iov, int niovecs)
2469 register struct CallBack * cb;
2470 struct CBDiskEntry * cbdsk;
2475 for (idx = 0; idx < niovecs; idx++) {
2476 cbdsk = (struct CBDiskEntry *) iov[idx].iov_base;
2477 if ((cb = GetCB()) == NULL) {
2478 ViceLog(0, ("cb_stateRestoreCBs: ran out of free CallBack structures\n"));
2482 if (cb_stateDiskEntryToCB(state, cbdsk, cb)) {
2483 ViceLog(0, ("cb_stateRestoreCBs: corrupt CallBack disk entry\n"));
2495 cb_stateFillHeader(struct callback_state_header * hdr)
2497 hdr->stamp.magic = CALLBACK_STATE_MAGIC;
2498 hdr->stamp.version = CALLBACK_STATE_VERSION;
2499 hdr->tfirst = tfirst;
2504 cb_stateCheckHeader(struct callback_state_header * hdr)
2508 if (hdr->stamp.magic != CALLBACK_STATE_MAGIC) {
2510 } else if (hdr->stamp.version != CALLBACK_STATE_VERSION) {
2512 } else if ((hdr->nFEs > cbstuff.nblks) || (hdr->nCBs > cbstuff.nblks)) {
2513 ViceLog(0, ("cb_stateCheckHeader: saved callback state larger than callback memory allocation\n"));
2519 /* disk entry conversion routines */
2521 cb_stateFEToDiskEntry(struct FileEntry * in, struct FEDiskEntry * out)
2523 memcpy(&out->fe, in, sizeof(struct FileEntry));
2524 out->index = fetoi(in);
2529 cb_stateDiskEntryToFE(struct fs_dump_state * state,
2530 struct FEDiskEntry * in, struct FileEntry * out)
2534 memcpy(out, &in->fe, sizeof(struct FileEntry));
2536 /* setup FE map entry */
2537 if (!in->index || (in->index >= state->fe_map.len)) {
2538 ViceLog(0, ("cb_stateDiskEntryToFE: index (%d) out of range",
2543 state->fe_map.entries[in->index].old_idx = in->index;
2544 state->fe_map.entries[in->index].new_idx = fetoi(out);
2551 cb_stateCBToDiskEntry(struct CallBack * in, struct CBDiskEntry * out)
2553 memcpy(&out->cb, in, sizeof(struct CallBack));
2554 out->index = cbtoi(in);
2559 cb_stateDiskEntryToCB(struct fs_dump_state * state,
2560 struct CBDiskEntry * in, struct CallBack * out)
2564 memcpy(out, &in->cb, sizeof(struct CallBack));
2566 /* setup CB map entry */
2567 if (!in->index || (in->index >= state->cb_map.len)) {
2568 ViceLog(0, ("cb_stateDiskEntryToCB: index (%d) out of range\n",
2573 state->cb_map.entries[in->index].old_idx = in->index;
2574 state->cb_map.entries[in->index].new_idx = cbtoi(out);
2580 /* index map routines */
2582 cb_stateAllocMap(struct fs_dump_state * state)
2584 state->fe_map.len = state->cb_hdr->fe_max + 1;
2585 state->cb_map.len = state->cb_hdr->cb_max + 1;
2586 state->fe_map.entries = (struct idx_map_entry_t *)
2587 calloc(state->fe_map.len, sizeof(struct idx_map_entry_t));
2588 state->cb_map.entries = (struct idx_map_entry_t *)
2589 calloc(state->cb_map.len, sizeof(struct idx_map_entry_t));
2590 return ((state->fe_map.entries != NULL) && (state->cb_map.entries != NULL)) ? 0 : 1;
2594 fe_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2598 /* FEs use a one-based indexing system, so old==0 implies no mapping */
2604 if (old >= state->fe_map.len) {
2605 ViceLog(0, ("fe_OldToNew: index %d is out of range\n", old));
2607 } else if (state->fe_map.entries[old].old_idx != old) { /* sanity check */
2608 ViceLog(0, ("fe_OldToNew: index %d points to an invalid FileEntry record\n", old));
2611 *new = state->fe_map.entries[old].new_idx;
2619 cb_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2623 /* CBs use a one-based indexing system, so old==0 implies no mapping */
2629 if (old >= state->cb_map.len) {
2630 ViceLog(0, ("cb_OldToNew: index %d is out of range\n", old));
2632 } else if (state->cb_map.entries[old].old_idx != old) { /* sanity check */
2633 ViceLog(0, ("cb_OldToNew: index %d points to an invalid CallBack record\n", old));
2636 *new = state->cb_map.entries[old].new_idx;
2642 #endif /* AFS_DEMAND_ATTACH_FS */
2645 DumpCallBackState(void)
2648 afs_uint32 magic = MAGIC, now = FT_ApproxTime(), freelisthead;
2650 fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, O_WRONLY | O_CREAT | O_TRUNC,
2654 ("Couldn't create callback dump file %s\n",
2655 AFSDIR_SERVER_CBKDUMP_FILEPATH));
2658 (void)write(fd, &magic, sizeof(magic));
2659 (void)write(fd, &now, sizeof(now));
2660 (void)write(fd, &cbstuff, sizeof(cbstuff));
2661 (void)write(fd, TimeOuts, sizeof(TimeOuts));
2662 (void)write(fd, timeout, sizeof(timeout));
2663 (void)write(fd, &tfirst, sizeof(tfirst));
2664 freelisthead = cbtoi((struct CallBack *)CBfree);
2665 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2666 freelisthead = fetoi((struct FileEntry *)FEfree);
2667 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2668 (void)write(fd, HashTable, sizeof(HashTable));
2669 (void)write(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2670 (void)write(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2676 #endif /* !INTERPRET_DUMP */
2678 #ifdef INTERPRET_DUMP
2680 /* This is only compiled in for the callback analyzer program */
2681 /* Returns the time of the dump */
2683 ReadDump(char *file)
2686 afs_uint32 magic, freelisthead;
2689 fd = open(file, O_RDONLY);
2691 fprintf(stderr, "Couldn't read dump file %s\n", file);
2694 read(fd, &magic, sizeof(magic));
2695 if (magic != MAGIC) {
2697 "Magic number of %s is invalid. You might be trying to\n",
2700 "run this program on a machine type with a different byte ordering.\n");
2703 read(fd, &now, sizeof(now));
2704 read(fd, &cbstuff, sizeof(cbstuff));
2705 read(fd, TimeOuts, sizeof(TimeOuts));
2706 read(fd, timeout, sizeof(timeout));
2707 read(fd, &tfirst, sizeof(tfirst));
2708 read(fd, &freelisthead, sizeof(freelisthead));
2709 CB = ((struct CallBack
2710 *)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
2711 FE = ((struct FileEntry
2712 *)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
2713 CBfree = (struct CallBack *)itocb(freelisthead);
2714 read(fd, &freelisthead, sizeof(freelisthead));
2715 FEfree = (struct FileEntry *)itofe(freelisthead);
2716 read(fd, HashTable, sizeof(HashTable));
2717 read(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2718 read(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2720 perror("Error reading dumpfile");
2727 #include "AFS_component_version_number.h"
2729 #include "AFS_component_version_number.c"
2733 main(int argc, char **argv)
2735 int err = 0, cbi = 0, stats = 0, noptions = 0, all = 0, vol = 0, raw = 0;
2737 register struct FileEntry *fe;
2738 register struct CallBack *cb;
2741 memset(&fid, 0, sizeof(fid));
2744 while (argc && **argv == '-') {
2747 if (!strcmp(*argv, "-host")) {
2753 cbi = atoi(*++argv);
2754 } else if (!strcmp(*argv, "-fid")) {
2760 fid.Volume = atoi(*++argv);
2761 fid.Vnode = atoi(*++argv);
2762 fid.Unique = atoi(*++argv);
2763 } else if (!strcmp(*argv, "-time")) {
2764 fprintf(stderr, "-time not supported\n");
2766 } else if (!strcmp(*argv, "-stats")) {
2768 } else if (!strcmp(*argv, "-all")) {
2770 } else if (!strcmp(*argv, "-raw")) {
2772 } else if (!strcmp(*argv, "-volume")) {
2778 vol = atoi(*++argv);
2783 if (err || argc != 1) {
2785 "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] callbackdumpfile\n");
2787 "[cbid is shown for each host in the hosts.dump file]\n");
2790 now = ReadDump(*argv);
2791 if (stats || noptions == 0) {
2792 time_t uxtfirst = UXtime(tfirst);
2793 printf("The time of the dump was %u %s", now, ctime(&now));
2794 printf("The last time cleanup ran was %u %s", uxtfirst,
2796 PrintCallBackStats();
2801 struct CallBack *cb;
2802 struct FileEntry *fe;
2804 for (hash = 0; hash < FEHASH_SIZE; hash++) {
2805 for (feip = &HashTable[hash]; fe = itofe(*feip);) {
2806 if (!vol || (fe->volid == vol)) {
2807 register struct CallBack *cbnext;
2808 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
2810 cbnext = itocb(cb->cnext);
2820 afs_uint32 cfirst = cbi;
2825 } while (cbi != cfirst);
2830 printf("No callback entries for %u.%u\n", fid.Volume, fid.Vnode);
2833 cb = itocb(fe->firstcb);
2836 cb = itocb(cb->cnext);
2841 for (i = 1; i < cbstuff.nblks; i++) {
2842 p = (afs_int32 *) & FE[i];
2843 printf("%d:%12x%12x%12x%12x\n", i, p[0], p[1], p[2], p[3]);
2850 PrintCB(register struct CallBack *cb, afs_uint32 now)
2852 struct FileEntry *fe = itofe(cb->fhead);
2853 time_t expires = TIndexToTime(cb->thead);
2858 printf("vol=%u vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %d secs at %s",
2859 fe->volid, fe->vnode, fe->ncbs, cb->hhead, cb->status, fe->status,
2860 expires - now, ctime(&expires));
2865 #if !defined(INTERPRET_DUMP)
2867 ** try breaking calbacks on afidp from host. Use multi_rx.
2868 ** return 0 on success, non-zero on failure
2871 MultiBreakCallBackAlternateAddress(struct host *host, struct AFSCBFids *afidp)
2875 retVal = MultiBreakCallBackAlternateAddress_r(host, afidp);
2881 MultiBreakCallBackAlternateAddress_r(struct host *host,
2882 struct AFSCBFids *afidp)
2885 struct rx_connection **conns;
2886 struct rx_connection *connSuccess = 0;
2887 struct AddrPort *interfaces;
2888 static struct rx_securityClass *sc = 0;
2889 static struct AFSCBs tc = { 0, 0 };
2892 /* nothing more can be done */
2893 if (!host->interface)
2894 return 1; /* failure */
2896 /* the only address is the primary interface */
2897 if (host->interface->numberOfInterfaces <= 1)
2898 return 1; /* failure */
2900 /* initialise a security object only once */
2902 sc = rxnull_NewClientSecurityObject();
2904 i = host->interface->numberOfInterfaces;
2905 interfaces = calloc(i, sizeof(struct AddrPort));
2906 conns = calloc(i, sizeof(struct rx_connection *));
2907 if (!interfaces || !conns) {
2909 ("Failed malloc in MultiBreakCallBackAlternateAddress_r\n"));
2913 /* initialize alternate rx connections */
2914 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
2915 /* this is the current primary address */
2916 if (host->host == host->interface->interface[i].addr &&
2917 host->port == host->interface->interface[i].port)
2920 interfaces[j] = host->interface->interface[i];
2922 rx_NewConnection(interfaces[j].addr,
2923 interfaces[j].port, 1, sc, 0);
2924 rx_SetConnDeadTime(conns[j], 2);
2925 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
2929 assert(j); /* at least one alternate address */
2931 ("Starting multibreakcall back on all addr for host %x (%s:%d)\n",
2932 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
2934 multi_Rx(conns, j) {
2935 multi_RXAFSCB_CallBack(afidp, &tc);
2939 if (host->callback_rxcon)
2940 rx_DestroyConnection(host->callback_rxcon);
2941 host->callback_rxcon = conns[multi_i];
2942 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
2943 host->host = interfaces[multi_i].addr;
2944 host->port = interfaces[multi_i].port;
2945 h_AddHostToAddrHashTable_r(host->host, host->port, host);
2946 connSuccess = conns[multi_i];
2947 rx_SetConnDeadTime(host->callback_rxcon, 50);
2948 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
2950 ("multibreakcall success with addr %s:%d\n",
2951 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
2952 ntohs(interfaces[multi_i].port)));
2959 /* Destroy all connections except the one on which we succeeded */
2960 for (i = 0; i < j; i++)
2961 if (conns[i] != connSuccess)
2962 rx_DestroyConnection(conns[i]);
2968 return 0; /* success */
2970 return 1; /* failure */
2975 ** try multi_RX probes to host.
2976 ** return 0 on success, non-0 on failure
2979 MultiProbeAlternateAddress_r(struct host *host)
2982 struct rx_connection **conns;
2983 struct rx_connection *connSuccess = 0;
2984 struct AddrPort *interfaces;
2985 static struct rx_securityClass *sc = 0;
2988 /* nothing more can be done */
2989 if (!host->interface)
2990 return 1; /* failure */
2992 /* the only address is the primary interface */
2993 if (host->interface->numberOfInterfaces <= 1)
2994 return 1; /* failure */
2996 /* initialise a security object only once */
2998 sc = rxnull_NewClientSecurityObject();
3000 i = host->interface->numberOfInterfaces;
3001 interfaces = calloc(i, sizeof(struct AddrPort));
3002 conns = calloc(i, sizeof(struct rx_connection *));
3003 if (!interfaces || !conns) {
3004 ViceLog(0, ("Failed malloc in MultiProbeAlternateAddress_r\n"));
3008 /* initialize alternate rx connections */
3009 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3010 /* this is the current primary address */
3011 if (host->host == host->interface->interface[i].addr &&
3012 host->port == host->interface->interface[i].port)
3015 interfaces[j] = host->interface->interface[i];
3017 rx_NewConnection(interfaces[j].addr,
3018 interfaces[j].port, 1, sc, 0);
3019 rx_SetConnDeadTime(conns[j], 2);
3020 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3024 assert(j); /* at least one alternate address */
3026 ("Starting multiprobe on all addr for host %x (%s:%d)\n",
3027 host, afs_inet_ntoa_r(host->host, hoststr),
3028 ntohs(host->port)));
3030 multi_Rx(conns, j) {
3031 multi_RXAFSCB_ProbeUuid(&host->interface->uuid);
3035 if (host->callback_rxcon)
3036 rx_DestroyConnection(host->callback_rxcon);
3037 host->callback_rxcon = conns[multi_i];
3038 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
3039 host->host = interfaces[multi_i].addr;
3040 host->port = interfaces[multi_i].port;
3041 h_AddHostToAddrHashTable_r(host->host, host->port, host);
3042 connSuccess = conns[multi_i];
3043 rx_SetConnDeadTime(host->callback_rxcon, 50);
3044 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3046 ("multiprobe success with addr %s:%d\n",
3047 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3048 ntohs(interfaces[multi_i].port)));
3053 ("multiprobe failure with addr %s:%d\n",
3054 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3055 ntohs(interfaces[multi_i].port)));
3057 /* This is less than desirable but its the best we can do.
3058 * The AFS Cache Manager will return either 0 for a Uuid
3059 * match and a 1 for a non-match. If the error is 1 we
3060 * therefore know that our mapping of IP address to Uuid
3061 * is wrong. We should attempt to find the correct
3062 * Uuid and fix the host tables.
3064 if (multi_error == 1) {
3065 /* remove the current alternate address from this host */
3067 removeInterfaceAddr_r(host, interfaces[multi_i].addr, interfaces[multi_i].port);
3071 #ifdef AFS_DEMAND_ATTACH_FS
3072 /* try to bail ASAP if the fileserver is shutting down */
3074 if (fs_state.mode == FS_MODE_SHUTDOWN) {
3083 /* Destroy all connections except the one on which we succeeded */
3084 for (i = 0; i < j; i++)
3085 if (conns[i] != connSuccess)
3086 rx_DestroyConnection(conns[i]);
3092 return 0; /* success */
3094 return 1; /* failure */
3097 #endif /* !defined(INTERPRET_DUMP) */