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 * BreakVolumeCallBacksLater(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>
90 #ifdef HAVE_SYS_FILE_H
94 #include <afs/afs_assert.h>
96 #include <afs/nfs.h> /* yuck. This is an abomination. */
99 #include <afs/afscbint.h>
100 #include <afs/afsutil.h>
102 #include <afs/ihandle.h>
103 #include <afs/vnode.h>
104 #include <afs/volume.h>
105 #include "viced_prototypes.h"
108 #include <afs/ptclient.h> /* need definition of prlist for host.h */
110 #include "callback.h"
111 #ifdef AFS_DEMAND_ATTACH_FS
112 #include "../tviced/serialize_state.h"
113 #endif /* AFS_DEMAND_ATTACH_FS */
116 extern afsUUID FS_HostUUID;
117 extern int hostCount;
119 #ifndef INTERPRET_DUMP
120 static int ShowProblems = 1;
123 struct cbcounters cbstuff;
125 static struct FileEntry * FE = NULL; /* don't use FE[0] */
126 static struct CallBack * CB = NULL; /* don't use CB[0] */
128 static struct CallBack * CBfree = NULL;
129 static struct FileEntry * FEfree = NULL;
132 /* Time to live for call backs depends upon number of users of the file.
133 * TimeOuts is indexed by this number/8 (using TimeOut macro). Times
134 * in this table are for the workstation; server timeouts, add
137 static int TimeOuts[] = {
138 /* Note: don't make the first entry larger than 4 hours (see above) */
139 4 * 60 * 60, /* 0-7 users */
140 1 * 60 * 60, /* 8-15 users */
141 30 * 60, /* 16-23 users */
142 15 * 60, /* 24-31 users */
143 15 * 60, /* 32-39 users */
144 10 * 60, /* 40-47 users */
145 10 * 60, /* 48-55 users */
146 10 * 60, /* 56-63 users */
147 }; /* Anything more: MinTimeOut */
149 /* minimum time given for a call back */
150 #ifndef INTERPRET_DUMP
151 static int MinTimeOut = (7 * 60);
154 /* Heads of CB queues; a timeout index is 1+index into this array */
155 static afs_uint32 timeout[CB_NUM_TIMEOUT_QUEUES];
157 static afs_int32 tfirst; /* cbtime of oldest unexpired call back time queue */
160 /* 16 byte object get/free routines */
165 /* Prototypes for static routines */
166 static struct FileEntry *FindFE(AFSFid * fid);
168 #ifndef INTERPRET_DUMP
169 static struct CallBack *iGetCB(int *nused);
170 static int iFreeCB(struct CallBack *cb, int *nused);
171 static struct FileEntry *iGetFE(int *nused);
172 static int iFreeFE(struct FileEntry *fe, int *nused);
173 static int TAdd(struct CallBack *cb, afs_uint32 * thead);
174 static int TDel(struct CallBack *cb);
175 static int HAdd(struct CallBack *cb, struct host *host);
176 static int HDel(struct CallBack *cb);
177 static int CDel(struct CallBack *cb, int deletefe);
178 static int CDelPtr(struct FileEntry *fe, afs_uint32 * cbp,
180 static afs_uint32 *FindCBPtr(struct FileEntry *fe, struct host *host);
181 static int FDel(struct FileEntry *fe);
182 static int AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead,
183 int type, int locked);
184 static void MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
185 struct AFSCBFids *afidp, struct host *xhost);
186 static int MultiBreakVolumeCallBack_r(struct host *host, int isheld,
187 struct VCBParams *parms, int deletefe);
188 static int MultiBreakVolumeLaterCallBack(struct host *host, int isheld,
190 static int GetSomeSpace_r(struct host *hostp, int locked);
191 static int ClearHostCallbacks_r(struct host *hp, int locked);
192 static int DumpCallBackState_r(void);
195 #define GetCB() ((struct CallBack *)iGetCB(&cbstuff.nCBs))
196 #define GetFE() ((struct FileEntry *)iGetFE(&cbstuff.nFEs))
197 #define FreeCB(cb) iFreeCB((struct CallBack *)cb, &cbstuff.nCBs)
198 #define FreeFE(fe) iFreeFE((struct FileEntry *)fe, &cbstuff.nFEs)
201 /* Other protos - move out sometime */
202 void PrintCB(struct CallBack *cb, afs_uint32 now);
204 static afs_uint32 HashTable[FEHASH_SIZE]; /* File entry hash table */
206 static struct FileEntry *
211 struct FileEntry *fe;
213 hash = FEHash(fid->Volume, fid->Unique);
214 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
216 if (fe->volid == fid->Volume && fe->unique == fid->Unique
217 && fe->vnode == fid->Vnode && (fe->status & FE_LATER) != FE_LATER)
223 #ifndef INTERPRET_DUMP
225 static struct CallBack *
228 struct CallBack *ret;
230 if ((ret = CBfree)) {
231 CBfree = (struct CallBack *)(((struct object *)ret)->next);
238 iFreeCB(struct CallBack *cb, int *nused)
240 ((struct object *)cb)->next = (struct object *)CBfree;
246 static struct FileEntry *
249 struct FileEntry *ret;
251 if ((ret = FEfree)) {
252 FEfree = (struct FileEntry *)(((struct object *)ret)->next);
259 iFreeFE(struct FileEntry *fe, int *nused)
261 ((struct object *)fe)->next = (struct object *)FEfree;
267 /* Add cb to end of specified timeout list */
269 TAdd(struct CallBack *cb, afs_uint32 * thead)
272 (*thead) = cb->tnext = cb->tprev = cbtoi(cb);
274 struct CallBack *thp = itocb(*thead);
276 cb->tprev = thp->tprev;
280 thp->tprev = (itocb(thp->tprev)->tnext = cbtoi(cb));
282 thp->tprev = cbtoi(cb);
285 cb->thead = ttoi(thead);
289 /* Delete call back entry from timeout list */
291 TDel(struct CallBack *cb)
293 afs_uint32 *thead = itot(cb->thead);
295 if (*thead == cbtoi(cb))
296 *thead = (*thead == cb->tnext ? 0 : cb->tnext);
297 if (itocb(cb->tprev))
298 itocb(cb->tprev)->tnext = cb->tnext;
299 if (itocb(cb->tnext))
300 itocb(cb->tnext)->tprev = cb->tprev;
304 /* Add cb to end of specified host list */
306 HAdd(struct CallBack *cb, struct host *host)
308 cb->hhead = h_htoi(host);
310 host->cblist = cb->hnext = cb->hprev = cbtoi(cb);
312 struct CallBack *fcb = itocb(host->cblist);
314 cb->hprev = fcb->hprev;
315 cb->hnext = cbtoi(fcb);
316 fcb->hprev = (itocb(fcb->hprev)->hnext = cbtoi(cb));
321 /* Delete call back entry from host list */
323 HDel(struct CallBack *cb)
325 afs_uint32 *hhead = &h_itoh(cb->hhead)->cblist;
327 if (*hhead == cbtoi(cb))
328 *hhead = (*hhead == cb->hnext ? 0 : cb->hnext);
329 itocb(cb->hprev)->hnext = cb->hnext;
330 itocb(cb->hnext)->hprev = cb->hprev;
334 /* Delete call back entry from fid's chain of cb's */
335 /* N.B. This one also deletes the CB, and also possibly parent FE, so
336 * make sure that it is not on any other list before calling this
339 CDel(struct CallBack *cb, int deletefe)
342 struct FileEntry *fe = itofe(cb->fhead);
346 for (safety = 0, cbp = &fe->firstcb; *cbp && *cbp != cbi;
347 cbp = &itocb(*cbp)->cnext, safety++) {
348 if (safety > cbstuff.nblks + 10) {
349 osi_Panic("CDel: Internal Error -- shutting down: wanted %d from %d, now at %d\n",
350 cbi, fe->firstcb, *cbp);
352 ("CDel: Internal Error -- shutting down: wanted %d from %d, now at %d\n",
353 cbi, fe->firstcb, *cbp));
354 DumpCallBackState_r();
355 ShutDownAndCore(PANIC);
358 CDelPtr(fe, cbp, deletefe);
362 /* Same as CDel, but pointer to parent pointer to CB entry is passed,
363 * as well as file entry */
364 /* N.B. This one also deletes the CB, and also possibly parent FE, so
365 * make sure that it is not on any other list before calling this
367 static int Ccdelpt = 0, CcdelB = 0;
370 CDelPtr(struct FileEntry *fe, afs_uint32 * cbp,
383 if ((--fe->ncbs == 0) && deletefe)
389 FindCBPtr(struct FileEntry *fe, struct host *host)
391 afs_uint32 hostindex = h_htoi(host);
396 for (safety = 0, cbp = &fe->firstcb; *cbp; cbp = &cb->cnext, safety++) {
397 if (safety > cbstuff.nblks) {
398 ViceLog(0, ("FindCBPtr: Internal Error -- shutting down.\n"));
399 DumpCallBackState_r();
400 ShutDownAndCore(PANIC);
403 if (cb->hhead == hostindex)
409 /* Delete file entry from hash table */
411 FDel(struct FileEntry *fe)
414 afs_uint32 *p = &HashTable[FEHash(fe->volid, fe->unique)];
416 while (*p && *p != fei)
417 p = &itofe(*p)->fnext;
424 /* initialize the callback package */
426 InitCallBack(int nblks)
429 tfirst = CBtime(FT_ApproxTime());
430 /* N.B. The "-1", below, is because
431 * FE[0] and CB[0] are not used--and not allocated */
432 FE = ((struct FileEntry *)(calloc(nblks, sizeof(struct FileEntry))));
434 ViceLog(0, ("Failed malloc in InitCallBack\n"));
435 osi_Panic("Failed malloc in InitCallBack\n");
437 FE--; /* FE[0] is supposed to point to junk */
438 cbstuff.nFEs = nblks;
440 FreeFE(&FE[cbstuff.nFEs]); /* This is correct */
441 CB = ((struct CallBack *)(calloc(nblks, sizeof(struct CallBack))));
443 ViceLog(0, ("Failed malloc in InitCallBack\n"));
444 osi_Panic("Failed malloc in InitCallBack\n");
446 CB--; /* CB[0] is supposed to point to junk */
447 cbstuff.nCBs = nblks;
449 FreeCB(&CB[cbstuff.nCBs]); /* This is correct */
450 cbstuff.nblks = nblks;
451 cbstuff.nbreakers = 0;
457 XCallBackBulk_r(struct host * ahost, struct AFSFid * fids, afs_int32 nfids)
459 struct AFSCallBack tcbs[AFSCBMAX];
465 struct rx_connection *cb_conn = NULL;
468 rx_SetConnDeadTime(ahost->callback_rxcon, 4);
469 rx_SetConnHardDeadTime(ahost->callback_rxcon, AFS_HARDDEADTIME);
476 for (i = 0; i < nfids && i < AFSCBMAX; i++) {
477 tcbs[i].CallBackVersion = CALLBACK_VERSION;
478 tcbs[i].ExpirationTime = 0;
479 tcbs[i].CallBackType = CB_DROPPED;
481 tf.AFSCBFids_len = i;
482 tf.AFSCBFids_val = &(fids[j]);
486 tc.AFSCBs_val = tcbs;
488 cb_conn = ahost->callback_rxcon;
489 rx_GetConnection(cb_conn);
491 code |= RXAFSCB_CallBack(cb_conn, &tf, &tc);
492 rx_PutConnection(cb_conn);
500 /* the locked flag tells us if the host entry has already been locked
501 * by our parent. I don't think anybody actually calls us with the
502 * host locked, but here's how to make that work: GetSomeSpace has to
503 * change so that it doesn't attempt to lock any hosts < "host". That
504 * means that it might be unable to free any objects, so it has to
505 * return an exit status. If it fails, then AddCallBack1 might fail,
506 * as well. If so, the host->ResetDone should probably be set to 0,
507 * and we probably don't want to return a callback promise to the
508 * cache manager, either. */
510 AddCallBack1(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
518 if (!(host->hostFlags & HOSTDELETED))
519 retVal = AddCallBack1_r(host, fid, thead, type, 1);
529 AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
532 struct FileEntry *fe;
533 struct CallBack *cb = 0, *lastcb = 0;
534 struct FileEntry *newfe = 0;
535 afs_uint32 time_out = 0;
536 afs_uint32 *Thead = thead;
537 struct CallBack *newcb = 0;
540 cbstuff.AddCallBacks++;
544 /* allocate these guys first, since we can't call the allocator with
545 * the host structure locked -- or we might deadlock. However, we have
546 * to avoid races with FindFE... */
547 while (!(newcb = GetCB())) {
548 GetSomeSpace_r(host, locked);
550 while (!(newfe = GetFE())) { /* Get it now, so we don't have to call */
551 /* GetSomeSpace with the host locked, later. This might turn out to */
552 /* have been unneccessary, but that's actually kind of unlikely, since */
553 /* most files are not shared. */
554 GetSomeSpace_r(host, locked);
558 h_Lock_r(host); /* this can yield, so do it before we get any */
560 if (host->hostFlags & HOSTDELETED) {
568 if (type == CB_NORMAL) {
570 TimeCeiling(FT_ApproxTime() + TimeOut(fe ? fe->ncbs : 0) +
572 Thead = THead(CBtime(time_out));
573 } else if (type == CB_VOLUME) {
574 time_out = TimeCeiling((60 * 120 + FT_ApproxTime()) + ServerBias);
575 Thead = THead(CBtime(time_out));
576 } else if (type == CB_BULK) {
577 /* bulk status can get so many callbacks all at once, and most of them
578 * are probably not for things that will be used for long.
581 TimeCeiling(FT_ApproxTime() + ServerBias +
582 TimeOut(22 + (fe ? fe->ncbs : 0)));
583 Thead = THead(CBtime(time_out));
594 fe->volid = fid->Volume;
595 fe->vnode = fid->Vnode;
596 fe->unique = fid->Unique;
599 hash = FEHash(fid->Volume, fid->Unique);
600 fe->fnext = HashTable[hash];
601 HashTable[hash] = fetoi(fe);
603 for (safety = 0, lastcb = cb = itocb(fe->firstcb); cb;
604 lastcb = cb, cb = itocb(cb->cnext), safety++) {
605 if (safety > cbstuff.nblks) {
606 ViceLog(0, ("AddCallBack1: Internal Error -- shutting down.\n"));
607 DumpCallBackState_r();
608 ShutDownAndCore(PANIC);
610 if (cb->hhead == h_htoi(host))
613 if (cb) { /* Already have call back: move to new timeout list */
614 /* don't change delayed callbacks back to normal ones */
615 if (cb->status != CB_DELAYED)
617 /* Only move if new timeout is longer */
618 if (TNorm(ttoi(Thead)) > TNorm(cb->thead)) {
622 if (newfe == NULL) { /* we are using the new FE */
623 fe->firstcb = cbtoi(cb);
625 cb->fhead = fetoi(fe);
630 *(lastcb ? &lastcb->cnext : &fe->firstcb) = cbtoi(cb);
633 cb->fhead = fetoi(fe);
639 /* now free any still-unused callback or host entries */
645 if (!locked) /* freecb and freefe might(?) yield */
648 if (type == CB_NORMAL || type == CB_VOLUME || type == CB_BULK)
649 return time_out - ServerBias; /* Expires sooner at workstation */
655 CompareCBA(const void *e1, const void *e2)
657 const struct cbstruct *cba1 = (const struct cbstruct *)e1;
658 const struct cbstruct *cba2 = (const struct cbstruct *)e2;
659 return ((cba1->hp)->index - (cba2->hp)->index);
662 /* Take an array full of hosts, all held. Break callbacks to them, and
663 * release the holds once you're done, except don't release xhost. xhost
664 * may be NULL. Currently only works for a single Fid in afidp array.
665 * If you want to make this work with multiple fids, you need to fix
666 * the error handling. One approach would be to force a reset if a
667 * multi-fid call fails, or you could add delayed callbacks for each
668 * fid. You probably also need to sort and remove duplicate hosts.
669 * When this is called from the BreakVolumeCallBacks path, it does NOT
670 * force a reset if the RPC fails, it just marks the host down and tries
671 * to create a delayed callback. */
672 /* N.B. be sure that code works when ncbas == 0 */
673 /* N.B. requires all the cba[*].hp pointers to be valid... */
674 /* This routine does not hold a lock on the host for the duration of
675 * the BreakCallBack RPC, which is a significant deviation from tradition.
676 * It _does_ get a lock on the host before setting VenusDown = 1,
677 * which is sufficient only if VenusDown = 0 only happens when the
678 * lock is held over the RPC and the subsequent VenusDown == 0
679 * wherever that is done. */
681 MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
682 struct AFSCBFids *afidp, struct host *xhost)
685 struct rx_connection *conns[MAX_CB_HOSTS];
686 static struct AFSCBs tc = { 0, 0 };
687 int multi_to_cba_map[MAX_CB_HOSTS];
689 osi_Assert(ncbas <= MAX_CB_HOSTS);
691 /* sort cba list to avoid makecall issues */
692 qsort(cba, ncbas, sizeof(struct cbstruct), CompareCBA);
694 /* set up conns for multi-call */
695 for (i = 0, j = 0; i < ncbas; i++) {
696 struct host *thishost = cba[i].hp;
697 if (!thishost || (thishost->hostFlags & HOSTDELETED)) {
700 rx_GetConnection(thishost->callback_rxcon);
701 multi_to_cba_map[j] = i;
702 conns[j++] = thishost->callback_rxcon;
705 rx_SetConnDeadTime(thishost->callback_rxcon, 4);
706 rx_SetConnHardDeadTime(thishost->callback_rxcon, AFS_HARDDEADTIME);
710 if (j) { /* who knows what multi would do with 0 conns? */
714 multi_RXAFSCB_CallBack(afidp, &tc);
720 i = multi_to_cba_map[multi_i];
726 ("BCB: INTERNAL ERROR: hp=%p, cba=%p, thead=%u\n",
730 ** try breaking callbacks on alternate interface addresses
732 if (MultiBreakCallBackAlternateAddress(hp, afidp)) {
735 ("BCB: Failed on file %u.%u.%u, "
736 "Host %p (%s:%d) is down\n",
737 afidp->AFSCBFids_val->Volume,
738 afidp->AFSCBFids_val->Vnode,
739 afidp->AFSCBFids_val->Unique,
741 afs_inet_ntoa_r(hp->host, hoststr),
747 if (!(hp->hostFlags & HOSTDELETED)) {
748 hp->hostFlags |= VENUSDOWN;
750 * We always go into AddCallBack1_r with the host locked
752 AddCallBack1_r(hp, afidp->AFSCBFids_val, itot(idx),
766 for (i = 0; i < ncbas; i++) {
769 if (hp && xhost != hp) {
774 /* H_UNLOCK around this so h_FreeConnection does not deadlock.
775 h_FreeConnection should *never* be called on a callback connection,
776 but on 10/27/04 a deadlock occurred where it was, when we know why,
777 this should be reverted. -- shadow */
779 for (i = 0; i < j; i++) {
780 rx_PutConnection(conns[i]);
788 * Break all call backs for fid, except for the specified host (unless flag
789 * is true, in which case all get a callback message. Assumption: the specified
790 * host is h_Held, by the caller; the others aren't.
791 * Specified host may be bogus, that's ok. This used to check to see if the
792 * host was down in two places, once right after the host was h_held, and
793 * again after it was locked. That race condition is incredibly rare and
794 * relatively harmless even when it does occur, so we don't check for it now.
796 /* if flag is true, send a break callback msg to "host", too */
798 BreakCallBack(struct host *xhost, AFSFid * fid, int flag)
800 struct FileEntry *fe;
801 struct CallBack *cb, *nextcb;
802 struct cbstruct cbaDef[MAX_CB_HOSTS], *cba = cbaDef;
803 unsigned int ncbas, cbaAlloc = MAX_CB_HOSTS;
809 ("BCB: BreakCallBack(Host %p all but %s:%d, (%u,%u,%u))\n",
810 xhost, afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
811 fid->Volume, fid->Vnode, fid->Unique));
814 cbstuff.BreakCallBacks++;
819 hostindex = h_htoi(xhost);
820 cb = itocb(fe->firstcb);
821 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
822 /* the most common case is what follows the || */
825 tf.AFSCBFids_len = 1;
826 tf.AFSCBFids_val = fid;
828 for (ncbas = 0; cb ; cb = nextcb) {
829 nextcb = itocb(cb->cnext);
830 if ((cb->hhead != hostindex || flag)
831 && (cb->status == CB_BULK || cb->status == CB_NORMAL
832 || cb->status == CB_VOLUME)) {
833 struct host *thishost = h_itoh(cb->hhead);
835 ViceLog(0, ("BCB: BOGUS! cb->hhead is NULL!\n"));
836 } else if (thishost->hostFlags & VENUSDOWN) {
838 ("BCB: %p (%s:%d) is down; delaying break call back\n",
839 thishost, afs_inet_ntoa_r(thishost->host, hoststr),
840 ntohs(thishost->port)));
841 cb->status = CB_DELAYED;
843 if (!(thishost->hostFlags & HOSTDELETED)) {
845 if (ncbas == cbaAlloc) { /* Need more space */
846 int curLen = cbaAlloc*sizeof(cba[0]);
847 struct cbstruct *cbaOld = (cba == cbaDef) ? NULL : cba;
849 /* There are logical contraints elsewhere that the number of hosts
850 (i.e. h_HTSPERBLOCK*h_MAXHOSTTABLES) remains in the realm of a signed "int".
851 cbaAlloc is defined unsigned int hence doubling below cannot overflow
853 cbaAlloc = cbaAlloc<<1; /* double */
854 cba = realloc(cbaOld, cbaAlloc * sizeof(cba[0]));
856 if (cbaOld == NULL) { /* realloc wouldn't have copied from cbaDef */
857 memcpy(cba, cbaDef, curLen);
860 cba[ncbas].hp = thishost;
861 cba[ncbas].thead = cb->thead;
866 CDel(cb, 1); /* Usually first; so this delete
867 * is reasonably inexpensive */
873 struct cbstruct *cba2;
876 for (cba2 = cba, num = ncbas; ncbas > 0; cba2 += num, ncbas -= num) {
877 num = (ncbas > MAX_CB_HOSTS) ? MAX_CB_HOSTS : ncbas;
878 MultiBreakCallBack_r(cba2, num, &tf, xhost);
882 if (cba != cbaDef) free(cba);
889 /* Delete (do not break) single call back for fid */
891 DeleteCallBack(struct host *host, AFSFid * fid)
893 struct FileEntry *fe;
898 cbstuff.DeleteCallBacks++;
901 /* do not care if the host has been HOSTDELETED */
907 ("DCB: No call backs for fid (%u, %u, %u)\n", fid->Volume,
908 fid->Vnode, fid->Unique));
911 pcb = FindCBPtr(fe, host);
914 ("DCB: No call back for host %p (%s:%d), (%u, %u, %u)\n",
915 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
916 fid->Volume, fid->Vnode, fid->Unique));
930 * Delete (do not break) all call backs for fid. This call doesn't
931 * set all of the various host locks, but it shouldn't really matter
932 * since we're not adding callbacks, but deleting them. I'm not sure
933 * why it doesn't set the lock, however; perhaps it should.
936 DeleteFileCallBacks(AFSFid * fid)
938 struct FileEntry *fe;
944 cbstuff.DeleteFiles++;
949 ("DF: No fid (%u,%u,%u) to delete\n", fid->Volume, fid->Vnode,
953 for (n = 0, cbi = fe->firstcb; cbi; n++) {
966 /* Delete (do not break) all call backs for host. The host should be
969 DeleteAllCallBacks_r(struct host *host, int deletefe)
974 cbstuff.DeleteAllCallBacks++;
975 cbi = first = host->cblist;
977 ViceLog(8, ("DV: no call backs\n"));
985 } while (cbi != first);
991 * Break all delayed call backs for host. Returns 1 if all call backs
992 * successfully broken; 0 otherwise. Assumes host is h_Held and h_Locked.
993 * Must be called with VenusDown set for this host
996 BreakDelayedCallBacks(struct host *host)
1000 retVal = BreakDelayedCallBacks_r(host);
1006 BreakDelayedCallBacks_r(struct host *host)
1008 struct AFSFid fids[AFSCBMAX];
1009 int cbi, first, nfids;
1010 struct CallBack *cb;
1013 struct rx_connection *cb_conn;
1015 cbstuff.nbreakers++;
1016 if (!(host->hostFlags & RESETDONE) && !(host->hostFlags & HOSTDELETED)) {
1017 host->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1018 cb_conn = host->callback_rxcon;
1019 rx_GetConnection(cb_conn);
1020 if (host->interface) {
1023 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1026 code = RXAFSCB_InitCallBackState(cb_conn);
1028 rx_PutConnection(cb_conn);
1031 host->hostFlags |= ALTADDR; /* alternate addresses are valid */
1035 ("CB: Call back connect back failed (in break delayed) "
1036 "for Host %p (%s:%d)\n",
1037 host, afs_inet_ntoa_r(host->host, hoststr),
1038 ntohs(host->port)));
1040 host->hostFlags |= VENUSDOWN;
1043 ("InitCallBackState success on %p (%s:%d)\n",
1044 host, afs_inet_ntoa_r(host->host, hoststr),
1045 ntohs(host->port)));
1046 /* reset was done successfully */
1047 host->hostFlags |= RESETDONE;
1048 host->hostFlags &= ~VENUSDOWN;
1051 while (!(host->hostFlags & HOSTDELETED)) {
1053 host->hostFlags &= ~VENUSDOWN; /* presume up */
1054 cbi = first = host->cblist;
1058 first = host->cblist;
1061 if (cb->status == CB_DELAYED) {
1062 struct FileEntry *fe = itofe(cb->fhead);
1063 fids[nfids].Volume = fe->volid;
1064 fids[nfids].Vnode = fe->vnode;
1065 fids[nfids].Unique = fe->unique;
1071 } while (cbi && cbi != first && nfids < AFSCBMAX);
1077 if (XCallBackBulk_r(host, fids, nfids)) {
1078 /* Failed, again: put them back, probably with old
1083 ("CB: XCallBackBulk failed, Host %p (%s:%d); "
1084 "callback list follows:\n",
1085 host, afs_inet_ntoa_r(host->host, hoststr),
1086 ntohs(host->port)));
1088 for (i = 0; i < nfids; i++) {
1091 ("CB: Host %p (%s:%d), file %u.%u.%u "
1092 "(part of bulk callback)\n",
1093 host, afs_inet_ntoa_r(host->host, hoststr),
1094 ntohs(host->port), fids[i].Volume,
1095 fids[i].Vnode, fids[i].Unique));
1098 * AddCallBack1_r(host, &fids[i], itot(thead[i]), CB_DELAYED, 1);
1099 * * but it turns out to cause too many tricky locking problems.
1100 * * now, if break delayed fails, screw it. */
1102 host->hostFlags |= VENUSDOWN; /* Failed */
1103 ClearHostCallbacks_r(host, 1 /* locked */ );
1107 if (nfids < AFSCBMAX)
1111 cbstuff.nbreakers--;
1112 /* If we succeeded it's always ok to unset HFE_LATER */
1113 if (!(host->hostFlags & VENUSDOWN))
1114 host->hostFlags &= ~HFE_LATER;
1115 return (host->hostFlags & VENUSDOWN);
1119 ** isheld is 0 if the host is held in h_Enumerate
1120 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1123 MultiBreakVolumeCallBack_r(struct host *host, int isheld,
1124 struct VCBParams *parms, int deletefe)
1128 if (host->hostFlags & HOSTDELETED)
1129 return 0; /* host is deleted, release hold */
1131 if (!(host->hostFlags & HCBREAK))
1132 return 0; /* host is not flagged to notify */
1134 if (host->hostFlags & VENUSDOWN) {
1136 /* Do not care if the host is now HOSTDELETED */
1139 ("BVCB: volume callback for Host %p (%s:%d) failed\n",
1140 host, afs_inet_ntoa_r(host->host, hoststr),
1141 ntohs(host->port)));
1143 DeleteAllCallBacks_r(host, deletefe); /* Delete all callback state
1144 * rather than attempting to
1145 * selectively remember to
1146 * delete the volume callbacks
1148 host->hostFlags &= ~(RESETDONE|HCBREAK); /* Do InitCallBackState when host returns */
1150 return 0; /* parent will release hold */
1152 osi_Assert(parms->ncbas <= MAX_CB_HOSTS);
1154 /* Do not call MultiBreakCallBack on the current host structure
1155 ** because it would prematurely release the hold on the host
1157 if (parms->ncbas == MAX_CB_HOSTS) {
1158 struct AFSCBFids tf;
1160 tf.AFSCBFids_len = 1;
1161 tf.AFSCBFids_val = parms->fid;
1163 /* this releases all the hosts */
1164 MultiBreakCallBack_r(parms->cba, parms->ncbas, &tf, 0 /* xhost */ );
1168 parms->cba[parms->ncbas].hp = host;
1169 parms->cba[(parms->ncbas)++].thead = parms->thead;
1170 host->hostFlags &= ~HCBREAK;
1171 return 1; /* parent shouldn't release hold, more work to do */
1175 ** isheld is 0 if the host is held in h_Enumerate
1176 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1179 MultiBreakVolumeLaterCallBack(struct host *host, int isheld, void *rock)
1181 struct VCBParams *parms = (struct VCBParams *)rock;
1184 retval = MultiBreakVolumeCallBack_r(host, isheld, parms, 0);
1190 * Break all call backs on a single volume. Don't call this with any
1191 * hosts h_held. Note that this routine clears the callbacks before
1192 * actually breaking them, and that the vnode isn't locked during this
1193 * operation, so that people might see temporary callback loss while
1194 * this function is executing. It is just a temporary state, however,
1195 * since the callback will be broken later by this same function.
1197 * Now uses multi-RX for CallBack RPC in a different thread,
1198 * only marking them here.
1200 #ifdef AFS_PTHREAD_ENV
1201 extern pthread_cond_t fsync_cond;
1203 extern char fsync_wait[];
1207 BreakVolumeCallBacksLater(afs_uint32 volume)
1211 struct FileEntry *fe;
1212 struct CallBack *cb;
1216 ViceLog(25, ("Setting later on volume %u\n", volume));
1218 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1219 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1220 if (fe->volid == volume) {
1221 struct CallBack *cbnext;
1222 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1223 host = h_itoh(cb->hhead);
1224 host->hostFlags |= HFE_LATER;
1225 cb->status = CB_DELAYED;
1226 cbnext = itocb(cb->cnext);
1229 fe->status |= FE_LATER;
1238 /* didn't find any callbacks, so return right away. */
1242 ViceLog(25, ("Fsync thread wakeup\n"));
1243 #ifdef AFS_PTHREAD_ENV
1245 CV_BROADCAST(&fsync_cond);
1248 LWP_NoYieldSignal(fsync_wait);
1254 BreakLaterCallBacks(void)
1259 struct CallBack *cb;
1260 struct FileEntry *fe = NULL;
1261 struct FileEntry *myfe = NULL;
1263 struct VCBParams henumParms;
1264 unsigned short tthead = 0; /* zero is illegal value */
1268 ViceLog(25, ("Looking for FileEntries to unchain\n"));
1271 /* Pick the first volume we see to clean up */
1272 fid.Volume = fid.Vnode = fid.Unique = 0;
1274 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1275 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1276 if (fe && (fe->status & FE_LATER)
1277 && (fid.Volume == 0 || fid.Volume == fe->volid)) {
1278 /* Ugly, but used to avoid left side casting */
1279 struct object *tmpfe;
1281 ("Unchaining for %u:%u:%u\n", fe->vnode, fe->unique,
1283 fid.Volume = fe->volid;
1285 fe->status &= ~FE_LATER; /* not strictly needed */
1286 /* Works since volid is deeper than the largest pointer */
1287 tmpfe = (struct object *)fe;
1288 tmpfe->next = (struct object *)myfe;
1301 /* loop over FEs from myfe and free/break */
1303 for (fe = myfe; fe;) {
1304 struct CallBack *cbnext;
1305 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1306 cbnext = itocb(cb->cnext);
1307 host = h_itoh(cb->hhead);
1308 if (cb->status == CB_DELAYED) {
1309 if (!(host->hostFlags & HOSTDELETED)) {
1310 /* mark this host for notification */
1311 host->hostFlags |= HCBREAK;
1312 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1318 CDel(cb, 0); /* Don't let CDel clean up the fe */
1319 /* leave flag for MultiBreakVolumeCallBack to clear */
1322 ("Found host %p (%s:%d) non-DELAYED cb for %u:%u:%u\n",
1323 host, afs_inet_ntoa_r(host->host, hoststr),
1324 ntohs(host->port), fe->vnode, fe->unique, fe->volid));
1328 fe = (struct FileEntry *)((struct object *)fe)->next;
1333 ViceLog(125, ("Breaking volume %u\n", fid.Volume));
1334 henumParms.ncbas = 0;
1335 henumParms.fid = &fid;
1336 henumParms.thead = tthead;
1338 h_Enumerate(MultiBreakVolumeLaterCallBack, (char *)&henumParms);
1340 if (henumParms.ncbas) { /* do left-overs */
1341 struct AFSCBFids tf;
1342 tf.AFSCBFids_len = 1;
1343 tf.AFSCBFids_val = &fid;
1345 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf, 0);
1346 henumParms.ncbas = 0;
1351 /* Arrange to be called again */
1356 * Delete all timed-out call back entries (to be called periodically by file
1360 CleanupTimedOutCallBacks(void)
1363 CleanupTimedOutCallBacks_r();
1369 CleanupTimedOutCallBacks_r(void)
1371 afs_uint32 now = CBtime(FT_ApproxTime());
1373 struct CallBack *cb;
1377 while (tfirst <= now) {
1379 cbi = *(thead = THead(tfirst));
1385 ("CCB: deleting timed out call back %x (%s:%d), (%u,%u,%u)\n",
1386 h_itoh(cb->hhead)->host,
1387 afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
1388 h_itoh(cb->hhead)->port, itofe(cb->fhead)->volid,
1389 itofe(cb->fhead)->vnode, itofe(cb->fhead)->unique));
1393 if (ntimedout > cbstuff.nblks) {
1394 ViceLog(0, ("CCB: Internal Error -- shutting down...\n"));
1395 DumpCallBackState_r();
1396 ShutDownAndCore(PANIC);
1398 } while (cbi != *thead);
1403 cbstuff.CBsTimedOut += ntimedout;
1404 ViceLog(7, ("CCB: deleted %d timed out callbacks\n", ntimedout));
1405 return (ntimedout > 0);
1409 * parameters to pass to lih*_r from h_Enumerate_r when trying to find a host
1410 * from which to clear callbacks.
1414 * Points to the least interesting host found; try to clear callbacks on
1415 * this host after h_Enumerate_r(lih*_r)'ing.
1420 * The last host we got from lih*_r, but we couldn't clear its callbacks
1421 * for some reason. Choose the next-best host after this one (with the
1422 * current lih*_r, this means to only select hosts that have an ActiveCall
1423 * newer than lastlih).
1425 struct host *lastlih;
1428 /* Value of host->refCount that allows us to reliably infer that
1429 * host may be held by some other thread */
1430 #define OTHER_MUSTHOLD_LIH 2
1432 /* This version does not allow 'host' to be selected unless its ActiveCall
1433 * is newer than 'params->lastlih' which is the host with the oldest
1434 * ActiveCall from the last pass (if it is provided). We filter out any hosts
1435 * that are are held by other threads.
1437 * There is a small problem here, but it may not be easily fixable. Say we
1438 * select some host A, and give it back to GetSomeSpace_r. GSS_r for some
1439 * reason cannot clear the callbacks on A, and so calls us again with
1440 * lastlih = A. Suppose there is another host B that has the same ActiveCall
1441 * time as A. We will now skip over host B, since
1442 * 'hostB->ActiveCall > hostA->ActiveCall' is not true. This could result in
1443 * us prematurely going to the GSS_r 2nd or 3rd pass, and making us a little
1444 * inefficient. This should be pretty rare, though, except perhaps in cases
1445 * with very small numbers of hosts.
1447 * Also filter out any hosts with HOSTDELETED set. h_Enumerate_r should in
1448 * theory not give these to us anyway, but be paranoid.
1451 lih0_r(struct host *host, int flags, void *rock)
1453 struct lih_params *params = (struct lih_params *)rock;
1455 /* OTHER_MUSTHOLD_LIH is because the h_Enum loop holds us once */
1457 && (!(host->hostFlags & HOSTDELETED))
1458 && (host->refCount < OTHER_MUSTHOLD_LIH)
1459 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1460 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1463 h_Release_r(params->lih); /* release prev host */
1472 /* same as lih0_r, except we do not prevent held hosts from being selected. */
1474 lih1_r(struct host *host, int flags, void *rock)
1476 struct lih_params *params = (struct lih_params *)rock;
1479 && (!(host->hostFlags & HOSTDELETED))
1480 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1481 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1484 h_Release_r(params->lih); /* release prev host */
1493 /* This could be upgraded to get more space each time */
1494 /* first pass: sequentially find the oldest host which isn't held by
1495 anyone for which we can clear callbacks;
1497 /* second pass: sequentially find the oldest host regardless of
1498 whether or not the host is held; skipping 'hostp' */
1499 /* third pass: attempt to clear callbacks from 'hostp' */
1500 /* always called with hostp unlocked */
1502 /* Note: hostlist is ordered most recently created host first and
1503 * its order has no relationship to the most recently used. */
1504 extern struct host *hostList;
1506 GetSomeSpace_r(struct host *hostp, int locked)
1509 struct lih_params params;
1512 cbstuff.GotSomeSpaces++;
1514 ("GSS: First looking for timed out call backs via CleanupCallBacks\n"));
1515 if (CleanupTimedOutCallBacks_r()) {
1521 params.lastlih = NULL;
1526 h_Enumerate_r(i == 0 ? lih0_r : lih1_r, hostList, ¶ms);
1529 if (params.lastlih) {
1530 h_Release_r(params.lastlih);
1531 params.lastlih = NULL;
1535 /* note that 'hp' was held by lih*_r; we will need to release it */
1537 if ((hp != hostp) && !ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
1542 params.lastlih = hp;
1543 /* params.lastlih will be released on the next iteration, after
1548 * Next time try getting callbacks from any host even if
1549 * it's held, since the only other option is starvation for
1550 * the file server (i.e. until the callback timeout arrives).
1553 params.lastlih = NULL;
1556 ("GSS: Try harder for longest inactive host cnt= %d\n",
1561 /* Could not obtain space from other hosts, clear hostp's callback state */
1566 ClearHostCallbacks_r(hostp, 1 /*already locked */ );
1573 /* locked - set if caller has already locked the host */
1575 ClearHostCallbacks_r(struct host *hp, int locked)
1579 struct rx_connection *cb_conn = NULL;
1582 ("GSS: Delete longest inactive host %p (%s:%d)\n",
1583 hp, afs_inet_ntoa_r(hp->host, hoststr), ntohs(hp->port)));
1585 if ((hp->hostFlags & HOSTDELETED)) {
1586 /* hp could go away after reacquiring H_LOCK in h_NBLock_r, so we can't
1587 * really use it; its callbacks will get cleared anyway when
1588 * h_TossStuff_r gets its hands on it */
1594 /** Try a non-blocking lock. If the lock is already held return
1595 * after releasing hold on hp
1598 if (h_NBLock_r(hp)) {
1603 if (hp->Console & 2) {
1605 * If the special console field is set it means that a thread
1606 * is waiting in AddCallBack1 after it set pointers to the
1607 * file entry and/or callback entry. Because of the bogus
1608 * usage of h_hold it won't prevent from another thread, this
1609 * one, to remove all the callbacks so just to be safe we keep
1610 * a reference. NOTE, on the last phase we'll free the calling
1611 * host's callbacks but that's ok...
1615 DeleteAllCallBacks_r(hp, 1);
1616 if (hp->hostFlags & VENUSDOWN) {
1617 hp->hostFlags &= ~RESETDONE; /* remember that we must do a reset */
1618 } else if (!(hp->hostFlags & HOSTDELETED)) {
1619 /* host is up, try a call */
1620 hp->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1621 cb_conn = hp->callback_rxcon;
1622 rx_GetConnection(hp->callback_rxcon);
1623 if (hp->interface) {
1626 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1629 code = RXAFSCB_InitCallBackState(cb_conn);
1631 rx_PutConnection(cb_conn);
1634 hp->hostFlags |= ALTADDR; /* alternate addresses are valid */
1636 /* failed, mark host down and need reset */
1637 hp->hostFlags |= VENUSDOWN;
1638 hp->hostFlags &= ~RESETDONE;
1640 /* reset succeeded, we're done */
1641 hp->hostFlags |= RESETDONE;
1650 #endif /* INTERPRET_DUMP */
1654 PrintCallBackStats(void)
1657 "%d add CB, %d break CB, %d del CB, %d del FE, %d CB's timed out, %d space reclaim, %d del host\n",
1658 cbstuff.AddCallBacks, cbstuff.BreakCallBacks,
1659 cbstuff.DeleteCallBacks, cbstuff.DeleteFiles, cbstuff.CBsTimedOut,
1660 cbstuff.GotSomeSpaces, cbstuff.DeleteAllCallBacks);
1661 fprintf(stderr, "%d CBs, %d FEs, (%d of total of %d 16-byte blocks)\n",
1662 cbstuff.nCBs, cbstuff.nFEs, cbstuff.nCBs + cbstuff.nFEs,
1664 fprintf(stderr, "%d GSS1, %d GSS2, %d GSS3, %d GSS4, %d GSS5 (internal counters)\n",
1665 cbstuff.GSS1, cbstuff.GSS2, cbstuff.GSS3, cbstuff.GSS4, cbstuff.GSS5);
1670 #define MAGIC 0x12345678 /* To check byte ordering of dump when it is read in */
1671 #define MAGICV2 0x12345679 /* To check byte ordering & version of dump when it is read in */
1674 #ifndef INTERPRET_DUMP
1676 #ifdef AFS_DEMAND_ATTACH_FS
1679 * callback state serialization
1681 static int cb_stateSaveTimeouts(struct fs_dump_state * state);
1682 static int cb_stateSaveFEHash(struct fs_dump_state * state);
1683 static int cb_stateSaveFEs(struct fs_dump_state * state);
1684 static int cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe);
1685 static int cb_stateRestoreTimeouts(struct fs_dump_state * state);
1686 static int cb_stateRestoreFEHash(struct fs_dump_state * state);
1687 static int cb_stateRestoreFEs(struct fs_dump_state * state);
1688 static int cb_stateRestoreFE(struct fs_dump_state * state);
1689 static int cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
1690 struct iovec * iov, int niovecs);
1692 static int cb_stateVerifyFEHash(struct fs_dump_state * state);
1693 static int cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe);
1694 static int cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe);
1695 static int cb_stateVerifyTimeoutQueues(struct fs_dump_state * state);
1697 static int cb_stateFEToDiskEntry(struct FileEntry *, struct FEDiskEntry *);
1698 static int cb_stateDiskEntryToFE(struct fs_dump_state * state,
1699 struct FEDiskEntry *, struct FileEntry *);
1701 static int cb_stateCBToDiskEntry(struct CallBack *, struct CBDiskEntry *);
1702 static int cb_stateDiskEntryToCB(struct fs_dump_state * state,
1703 struct CBDiskEntry *, struct CallBack *);
1705 static int cb_stateFillHeader(struct callback_state_header * hdr);
1706 static int cb_stateCheckHeader(struct callback_state_header * hdr);
1708 static int cb_stateAllocMap(struct fs_dump_state * state);
1711 cb_stateSave(struct fs_dump_state * state)
1715 AssignInt64(state->eof_offset, &state->hdr->cb_offset);
1717 /* invalidate callback state header */
1718 memset(state->cb_hdr, 0, sizeof(struct callback_state_header));
1719 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1720 sizeof(struct callback_state_header))) {
1725 fs_stateIncEOF(state, sizeof(struct callback_state_header));
1727 /* dump timeout state */
1728 if (cb_stateSaveTimeouts(state)) {
1733 /* dump fe hashtable state */
1734 if (cb_stateSaveFEHash(state)) {
1739 /* dump callback state */
1740 if (cb_stateSaveFEs(state)) {
1745 /* write the callback state header to disk */
1746 cb_stateFillHeader(state->cb_hdr);
1747 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1748 sizeof(struct callback_state_header))) {
1758 cb_stateRestore(struct fs_dump_state * state)
1762 if (fs_stateReadHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1763 sizeof(struct callback_state_header))) {
1768 if (cb_stateCheckHeader(state->cb_hdr)) {
1773 if (cb_stateAllocMap(state)) {
1778 if (cb_stateRestoreTimeouts(state)) {
1783 if (cb_stateRestoreFEHash(state)) {
1788 /* restore FEs and CBs from disk */
1789 if (cb_stateRestoreFEs(state)) {
1794 /* restore the timeout queue heads */
1795 tfirst = state->cb_hdr->tfirst;
1802 cb_stateRestoreIndices(struct fs_dump_state * state)
1805 struct FileEntry * fe;
1806 struct CallBack * cb;
1808 /* restore indices in the FileEntry structures */
1809 for (i = 1; i < state->fe_map.len; i++) {
1810 if (state->fe_map.entries[i].new_idx) {
1811 fe = itofe(state->fe_map.entries[i].new_idx);
1813 /* restore the fe->fnext entry */
1814 if (fe_OldToNew(state, fe->fnext, &fe->fnext)) {
1819 /* restore the fe->firstcb entry */
1820 if (cb_OldToNew(state, fe->firstcb, &fe->firstcb)) {
1827 /* restore indices in the CallBack structures */
1828 for (i = 1; i < state->cb_map.len; i++) {
1829 if (state->cb_map.entries[i].new_idx) {
1830 cb = itocb(state->cb_map.entries[i].new_idx);
1832 /* restore the cb->cnext entry */
1833 if (cb_OldToNew(state, cb->cnext, &cb->cnext)) {
1838 /* restore the cb->fhead entry */
1839 if (fe_OldToNew(state, cb->fhead, &cb->fhead)) {
1844 /* restore the cb->hhead entry */
1845 if (h_OldToNew(state, cb->hhead, &cb->hhead)) {
1850 /* restore the cb->tprev entry */
1851 if (cb_OldToNew(state, cb->tprev, &cb->tprev)) {
1856 /* restore the cb->tnext entry */
1857 if (cb_OldToNew(state, cb->tnext, &cb->tnext)) {
1862 /* restore the cb->hprev entry */
1863 if (cb_OldToNew(state, cb->hprev, &cb->hprev)) {
1868 /* restore the cb->hnext entry */
1869 if (cb_OldToNew(state, cb->hnext, &cb->hnext)) {
1876 /* restore the timeout queue head indices */
1877 for (i = 0; i < state->cb_timeout_hdr->records; i++) {
1878 if (cb_OldToNew(state, timeout[i], &timeout[i])) {
1884 /* restore the FE hash table queue heads */
1885 for (i = 0; i < state->cb_fehash_hdr->records; i++) {
1886 if (fe_OldToNew(state, HashTable[i], &HashTable[i])) {
1897 cb_stateVerify(struct fs_dump_state * state)
1901 if (cb_stateVerifyFEHash(state)) {
1905 if (cb_stateVerifyTimeoutQueues(state)) {
1913 cb_stateVerifyFEHash(struct fs_dump_state * state)
1916 struct FileEntry * fe;
1917 afs_uint32 fei, chain_len;
1919 for (i = 0; i < FEHASH_SIZE; i++) {
1921 for (fei = HashTable[i], fe = itofe(fei);
1923 fei = fe->fnext, fe = itofe(fei)) {
1924 if (fei > cbstuff.nblks) {
1925 ViceLog(0, ("cb_stateVerifyFEHash: error: index out of range (fei=%d)\n", fei));
1929 if (cb_stateVerifyFE(state, fe)) {
1932 if (chain_len > FS_STATE_FE_MAX_HASH_CHAIN_LEN) {
1933 ViceLog(0, ("cb_stateVerifyFEHash: error: hash chain %d length exceeds %d; assuming there's a loop\n",
1934 i, FS_STATE_FE_MAX_HASH_CHAIN_LEN));
1946 cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe)
1950 if ((fe->firstcb && !fe->ncbs) ||
1951 (!fe->firstcb && fe->ncbs)) {
1952 ViceLog(0, ("cb_stateVerifyFE: error: fe->firstcb does not agree with fe->ncbs (fei=%lu, fe->firstcb=%lu, fe->ncbs=%lu)\n",
1953 afs_printable_uint32_lu(fetoi(fe)),
1954 afs_printable_uint32_lu(fe->firstcb),
1955 afs_printable_uint32_lu(fe->ncbs)));
1958 if (cb_stateVerifyFCBList(state, fe)) {
1959 ViceLog(0, ("cb_stateVerifyFE: error: FCBList failed verification (fei=%lu)\n",
1960 afs_printable_uint32_lu(fetoi(fe))));
1968 cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe)
1971 afs_uint32 cbi, fei, chain_len = 0;
1972 struct CallBack * cb;
1976 for (cbi = fe->firstcb, cb = itocb(cbi);
1978 cbi = cb->cnext, cb = itocb(cbi)) {
1979 if (cbi > cbstuff.nblks) {
1980 ViceLog(0, ("cb_stateVerifyFCBList: error: list index out of range (cbi=%d, ncbs=%d)\n",
1981 cbi, cbstuff.nblks));
1985 if (cb->fhead != fei) {
1986 ViceLog(0, ("cb_stateVerifyFCBList: error: cb->fhead != fei (fei=%d, cb->fhead=%d)\n",
1990 if (chain_len > FS_STATE_FCB_MAX_LIST_LEN) {
1991 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (fei=%d); assuming there's a loop\n",
1992 FS_STATE_FCB_MAX_LIST_LEN, fei));
1999 if (fe->ncbs != chain_len) {
2000 ViceLog(0, ("cb_stateVerifyFCBList: error: list length mismatch (len=%d, fe->ncbs=%d)\n",
2001 chain_len, fe->ncbs));
2010 cb_stateVerifyHCBList(struct fs_dump_state * state, struct host * host)
2013 afs_uint32 hi, chain_len, cbi;
2014 struct CallBack *cb, *ncb;
2019 for (cbi = host->cblist, cb = itocb(cbi);
2021 cbi = cb->hnext, cb = ncb) {
2022 if (chain_len && (host->cblist == cbi)) {
2023 /* we've wrapped around the circular list, and everything looks ok */
2026 if (cb->hhead != hi) {
2027 ViceLog(0, ("cb_stateVerifyHCBList: error: incorrect cb->hhead (cbi=%d, h->index=%d, cb->hhead=%d)\n",
2028 cbi, hi, cb->hhead));
2031 if (!cb->hprev || !cb->hnext) {
2032 ViceLog(0, ("cb_stateVerifyHCBList: error: null index in circular list (cbi=%d, h->index=%d)\n",
2037 if ((cb->hprev > cbstuff.nblks) ||
2038 (cb->hnext > cbstuff.nblks)) {
2039 ViceLog(0, ("cb_stateVerifyHCBList: error: list index out of range (cbi=%d, h->index=%d, cb->hprev=%d, cb->hnext=%d, nCBs=%d)\n",
2040 cbi, hi, cb->hprev, cb->hnext, cbstuff.nblks));
2044 ncb = itocb(cb->hnext);
2045 if (cbi != ncb->hprev) {
2046 ViceLog(0, ("cb_stateVerifyHCBList: error: corrupt linked list (cbi=%d, h->index=%d)\n",
2051 if (chain_len > FS_STATE_HCB_MAX_LIST_LEN) {
2052 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (h->index=%d); assuming there's a loop\n",
2053 FS_STATE_HCB_MAX_LIST_LEN, hi));
2065 cb_stateVerifyTimeoutQueues(struct fs_dump_state * state)
2068 afs_uint32 cbi, chain_len;
2069 struct CallBack *cb, *ncb;
2071 for (i = 0; i < CB_NUM_TIMEOUT_QUEUES; i++) {
2073 for (cbi = timeout[i], cb = itocb(cbi);
2075 cbi = cb->tnext, cb = ncb) {
2076 if (chain_len && (cbi == timeout[i])) {
2077 /* we've wrapped around the circular list, and everything looks ok */
2080 if (cbi > cbstuff.nblks) {
2081 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: list index out of range (cbi=%d, tindex=%d)\n",
2086 if (itot(cb->thead) != &timeout[i]) {
2087 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: cb->thead points to wrong timeout queue (tindex=%d, cbi=%d, cb->thead=%d)\n",
2088 i, cbi, cb->thead));
2091 if (!cb->tprev || !cb->tnext) {
2092 ViceLog(0, ("cb_stateVerifyTimeoutQueues: null index in circular list (cbi=%d, tindex=%d)\n",
2097 if ((cb->tprev > cbstuff.nblks) ||
2098 (cb->tnext > cbstuff.nblks)) {
2099 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list index out of range (cbi=%d, tindex=%d, cb->tprev=%d, cb->tnext=%d, nCBs=%d)\n",
2100 cbi, i, cb->tprev, cb->tnext, cbstuff.nblks));
2104 ncb = itocb(cb->tnext);
2105 if (cbi != ncb->tprev) {
2106 ViceLog(0, ("cb_stateVerifyTimeoutQueues: corrupt linked list (cbi=%d, tindex=%d)\n",
2111 if (chain_len > FS_STATE_TCB_MAX_LIST_LEN) {
2112 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list length exceeds %d (tindex=%d); assuming there's a loop\n",
2113 FS_STATE_TCB_MAX_LIST_LEN, i));
2125 cb_stateSaveTimeouts(struct fs_dump_state * state)
2128 struct iovec iov[2];
2130 AssignInt64(state->eof_offset, &state->cb_hdr->timeout_offset);
2132 memset(state->cb_timeout_hdr, 0, sizeof(struct callback_state_fehash_header));
2133 state->cb_timeout_hdr->magic = CALLBACK_STATE_TIMEOUT_MAGIC;
2134 state->cb_timeout_hdr->records = CB_NUM_TIMEOUT_QUEUES;
2135 state->cb_timeout_hdr->len = sizeof(struct callback_state_timeout_header) +
2136 (state->cb_timeout_hdr->records * sizeof(afs_uint32));
2138 iov[0].iov_base = (char *)state->cb_timeout_hdr;
2139 iov[0].iov_len = sizeof(struct callback_state_timeout_header);
2140 iov[1].iov_base = (char *)timeout;
2141 iov[1].iov_len = sizeof(timeout);
2143 if (fs_stateSeek(state, &state->cb_hdr->timeout_offset)) {
2148 if (fs_stateWriteV(state, iov, 2)) {
2153 fs_stateIncEOF(state, state->cb_timeout_hdr->len);
2160 cb_stateRestoreTimeouts(struct fs_dump_state * state)
2164 if (fs_stateReadHeader(state, &state->cb_hdr->timeout_offset,
2165 state->cb_timeout_hdr,
2166 sizeof(struct callback_state_timeout_header))) {
2171 if (state->cb_timeout_hdr->magic != CALLBACK_STATE_TIMEOUT_MAGIC) {
2175 if (state->cb_timeout_hdr->records != CB_NUM_TIMEOUT_QUEUES) {
2180 len = state->cb_timeout_hdr->records * sizeof(afs_uint32);
2182 if (state->cb_timeout_hdr->len !=
2183 (sizeof(struct callback_state_timeout_header) + len)) {
2188 if (fs_stateRead(state, timeout, len)) {
2198 cb_stateSaveFEHash(struct fs_dump_state * state)
2201 struct iovec iov[2];
2203 AssignInt64(state->eof_offset, &state->cb_hdr->fehash_offset);
2205 memset(state->cb_fehash_hdr, 0, sizeof(struct callback_state_fehash_header));
2206 state->cb_fehash_hdr->magic = CALLBACK_STATE_FEHASH_MAGIC;
2207 state->cb_fehash_hdr->records = FEHASH_SIZE;
2208 state->cb_fehash_hdr->len = sizeof(struct callback_state_fehash_header) +
2209 (state->cb_fehash_hdr->records * sizeof(afs_uint32));
2211 iov[0].iov_base = (char *)state->cb_fehash_hdr;
2212 iov[0].iov_len = sizeof(struct callback_state_fehash_header);
2213 iov[1].iov_base = (char *)HashTable;
2214 iov[1].iov_len = sizeof(HashTable);
2216 if (fs_stateSeek(state, &state->cb_hdr->fehash_offset)) {
2221 if (fs_stateWriteV(state, iov, 2)) {
2226 fs_stateIncEOF(state, state->cb_fehash_hdr->len);
2233 cb_stateRestoreFEHash(struct fs_dump_state * state)
2237 if (fs_stateReadHeader(state, &state->cb_hdr->fehash_offset,
2238 state->cb_fehash_hdr,
2239 sizeof(struct callback_state_fehash_header))) {
2244 if (state->cb_fehash_hdr->magic != CALLBACK_STATE_FEHASH_MAGIC) {
2248 if (state->cb_fehash_hdr->records != FEHASH_SIZE) {
2253 len = state->cb_fehash_hdr->records * sizeof(afs_uint32);
2255 if (state->cb_fehash_hdr->len !=
2256 (sizeof(struct callback_state_fehash_header) + len)) {
2261 if (fs_stateRead(state, HashTable, len)) {
2271 cb_stateSaveFEs(struct fs_dump_state * state)
2275 struct FileEntry *fe;
2277 AssignInt64(state->eof_offset, &state->cb_hdr->fe_offset);
2279 for (hash = 0; hash < FEHASH_SIZE ; hash++) {
2280 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
2282 if (cb_stateSaveFE(state, fe)) {
2294 cb_stateRestoreFEs(struct fs_dump_state * state)
2296 int count, nFEs, ret = 0;
2298 nFEs = state->cb_hdr->nFEs;
2300 for (count = 0; count < nFEs; count++) {
2301 if (cb_stateRestoreFE(state)) {
2312 cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe)
2314 int ret = 0, iovcnt, cbi, written = 0;
2316 struct callback_state_entry_header hdr;
2317 struct FEDiskEntry fedsk;
2318 struct CBDiskEntry cbdsk[16];
2319 struct iovec iov[16];
2320 struct CallBack *cb;
2323 if (fei > state->cb_hdr->fe_max) {
2324 state->cb_hdr->fe_max = fei;
2327 memset(&hdr, 0, sizeof(struct callback_state_entry_header));
2329 if (cb_stateFEToDiskEntry(fe, &fedsk)) {
2334 iov[0].iov_base = (char *)&hdr;
2335 iov[0].iov_len = sizeof(hdr);
2336 iov[1].iov_base = (char *)&fedsk;
2337 iov[1].iov_len = sizeof(struct FEDiskEntry);
2340 for (cbi = fe->firstcb, cb = itocb(cbi);
2342 cbi = cb->cnext, cb = itocb(cbi), hdr.nCBs++) {
2343 if (cbi > state->cb_hdr->cb_max) {
2344 state->cb_hdr->cb_max = cbi;
2346 if (cb_stateCBToDiskEntry(cb, &cbdsk[iovcnt])) {
2350 cbdsk[iovcnt].index = cbi;
2351 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2352 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2354 if ((iovcnt == 16) || (!cb->cnext)) {
2355 if (fs_stateWriteV(state, iov, iovcnt)) {
2364 hdr.magic = CALLBACK_STATE_ENTRY_MAGIC;
2365 hdr.len = sizeof(hdr) + sizeof(struct FEDiskEntry) +
2366 (hdr.nCBs * sizeof(struct CBDiskEntry));
2369 if (fs_stateWriteV(state, iov, iovcnt)) {
2374 if (fs_stateWriteHeader(state, &state->eof_offset, &hdr, sizeof(hdr))) {
2380 fs_stateIncEOF(state, hdr.len);
2383 if (fs_stateSeek(state, &state->eof_offset)) {
2389 state->cb_hdr->nFEs++;
2390 state->cb_hdr->nCBs += hdr.nCBs;
2397 cb_stateRestoreFE(struct fs_dump_state * state)
2399 int ret = 0, iovcnt, nCBs;
2400 struct callback_state_entry_header hdr;
2401 struct FEDiskEntry fedsk;
2402 struct CBDiskEntry cbdsk[16];
2403 struct iovec iov[16];
2404 struct FileEntry * fe;
2406 iov[0].iov_base = (char *)&hdr;
2407 iov[0].iov_len = sizeof(hdr);
2408 iov[1].iov_base = (char *)&fedsk;
2409 iov[1].iov_len = sizeof(fedsk);
2412 if (fs_stateReadV(state, iov, iovcnt)) {
2417 if (hdr.magic != CALLBACK_STATE_ENTRY_MAGIC) {
2424 ViceLog(0, ("cb_stateRestoreFE: ran out of free FileEntry structures\n"));
2429 if (cb_stateDiskEntryToFE(state, &fedsk, fe)) {
2435 for (iovcnt = 0, nCBs = 0;
2438 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2439 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2441 if ((iovcnt == 16) || (nCBs == hdr.nCBs - 1)) {
2442 if (fs_stateReadV(state, iov, iovcnt)) {
2446 if (cb_stateRestoreCBs(state, fe, iov, iovcnt)) {
2460 cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
2461 struct iovec * iov, int niovecs)
2464 struct CallBack * cb;
2465 struct CBDiskEntry * cbdsk;
2467 for (idx = 0; idx < niovecs; idx++) {
2468 cbdsk = (struct CBDiskEntry *) iov[idx].iov_base;
2469 if ((cb = GetCB()) == NULL) {
2470 ViceLog(0, ("cb_stateRestoreCBs: ran out of free CallBack structures\n"));
2474 if (cb_stateDiskEntryToCB(state, cbdsk, cb)) {
2475 ViceLog(0, ("cb_stateRestoreCBs: corrupt CallBack disk entry\n"));
2487 cb_stateFillHeader(struct callback_state_header * hdr)
2489 hdr->stamp.magic = CALLBACK_STATE_MAGIC;
2490 hdr->stamp.version = CALLBACK_STATE_VERSION;
2491 hdr->tfirst = tfirst;
2496 cb_stateCheckHeader(struct callback_state_header * hdr)
2500 if (hdr->stamp.magic != CALLBACK_STATE_MAGIC) {
2502 } else if (hdr->stamp.version != CALLBACK_STATE_VERSION) {
2504 } else if ((hdr->nFEs > cbstuff.nblks) || (hdr->nCBs > cbstuff.nblks)) {
2505 ViceLog(0, ("cb_stateCheckHeader: saved callback state larger than callback memory allocation\n"));
2511 /* disk entry conversion routines */
2513 cb_stateFEToDiskEntry(struct FileEntry * in, struct FEDiskEntry * out)
2515 memcpy(&out->fe, in, sizeof(struct FileEntry));
2516 out->index = fetoi(in);
2521 cb_stateDiskEntryToFE(struct fs_dump_state * state,
2522 struct FEDiskEntry * in, struct FileEntry * out)
2526 memcpy(out, &in->fe, sizeof(struct FileEntry));
2528 /* setup FE map entry */
2529 if (!in->index || (in->index >= state->fe_map.len)) {
2530 ViceLog(0, ("cb_stateDiskEntryToFE: index (%d) out of range",
2535 state->fe_map.entries[in->index].old_idx = in->index;
2536 state->fe_map.entries[in->index].new_idx = fetoi(out);
2543 cb_stateCBToDiskEntry(struct CallBack * in, struct CBDiskEntry * out)
2545 memcpy(&out->cb, in, sizeof(struct CallBack));
2546 out->index = cbtoi(in);
2551 cb_stateDiskEntryToCB(struct fs_dump_state * state,
2552 struct CBDiskEntry * in, struct CallBack * out)
2556 memcpy(out, &in->cb, sizeof(struct CallBack));
2558 /* setup CB map entry */
2559 if (!in->index || (in->index >= state->cb_map.len)) {
2560 ViceLog(0, ("cb_stateDiskEntryToCB: index (%d) out of range\n",
2565 state->cb_map.entries[in->index].old_idx = in->index;
2566 state->cb_map.entries[in->index].new_idx = cbtoi(out);
2572 /* index map routines */
2574 cb_stateAllocMap(struct fs_dump_state * state)
2576 state->fe_map.len = state->cb_hdr->fe_max + 1;
2577 state->cb_map.len = state->cb_hdr->cb_max + 1;
2578 state->fe_map.entries = (struct idx_map_entry_t *)
2579 calloc(state->fe_map.len, sizeof(struct idx_map_entry_t));
2580 state->cb_map.entries = (struct idx_map_entry_t *)
2581 calloc(state->cb_map.len, sizeof(struct idx_map_entry_t));
2582 return ((state->fe_map.entries != NULL) && (state->cb_map.entries != NULL)) ? 0 : 1;
2586 fe_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2590 /* FEs use a one-based indexing system, so old==0 implies no mapping */
2596 if (old >= state->fe_map.len) {
2597 ViceLog(0, ("fe_OldToNew: index %d is out of range\n", old));
2599 } else if (state->fe_map.entries[old].old_idx != old) { /* sanity check */
2600 ViceLog(0, ("fe_OldToNew: index %d points to an invalid FileEntry record\n", old));
2603 *new = state->fe_map.entries[old].new_idx;
2611 cb_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2615 /* CBs use a one-based indexing system, so old==0 implies no mapping */
2621 if (old >= state->cb_map.len) {
2622 ViceLog(0, ("cb_OldToNew: index %d is out of range\n", old));
2624 } else if (state->cb_map.entries[old].old_idx != old) { /* sanity check */
2625 ViceLog(0, ("cb_OldToNew: index %d points to an invalid CallBack record\n", old));
2628 *new = state->cb_map.entries[old].new_idx;
2634 #endif /* AFS_DEMAND_ATTACH_FS */
2637 DumpCallBackState_r(void)
2640 afs_uint32 magic = MAGICV2, now = (afs_int32) FT_ApproxTime(), freelisthead;
2642 oflag = O_WRONLY | O_CREAT | O_TRUNC;
2646 fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, oflag, 0666);
2649 ("Couldn't create callback dump file %s\n",
2650 AFSDIR_SERVER_CBKDUMP_FILEPATH));
2653 (void)write(fd, &magic, sizeof(magic));
2654 (void)write(fd, &now, sizeof(now));
2655 (void)write(fd, &cbstuff, sizeof(cbstuff));
2656 (void)write(fd, TimeOuts, sizeof(TimeOuts));
2657 (void)write(fd, timeout, sizeof(timeout));
2658 (void)write(fd, &tfirst, sizeof(tfirst));
2659 freelisthead = cbtoi((struct CallBack *)CBfree);
2660 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2661 freelisthead = fetoi((struct FileEntry *)FEfree);
2662 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2663 (void)write(fd, HashTable, sizeof(HashTable));
2664 (void)write(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2665 (void)write(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2672 DumpCallBackState(void) {
2676 rc = DumpCallBackState_r();
2682 #endif /* !INTERPRET_DUMP */
2684 #ifdef INTERPRET_DUMP
2686 /* This is only compiled in for the callback analyzer program */
2687 /* Returns the time of the dump */
2689 ReadDump(char *file, int timebits)
2692 afs_uint32 magic, freelisthead;
2700 fd = open(file, oflag);
2702 fprintf(stderr, "Couldn't read dump file %s\n", file);
2705 read(fd, &magic, sizeof(magic));
2706 if (magic == MAGICV2) {
2709 if (magic != MAGIC) {
2711 "Magic number of %s is invalid. You might be trying to\n",
2714 "run this program on a machine type with a different byte ordering.\n");
2718 if (timebits == 64) {
2719 read(fd, &now64, sizeof(afs_int64));
2720 now = (afs_int32) now64;
2722 read(fd, &now, sizeof(afs_int32));
2724 read(fd, &cbstuff, sizeof(cbstuff));
2725 read(fd, TimeOuts, sizeof(TimeOuts));
2726 read(fd, timeout, sizeof(timeout));
2727 read(fd, &tfirst, sizeof(tfirst));
2728 read(fd, &freelisthead, sizeof(freelisthead));
2729 CB = ((struct CallBack
2730 *)(calloc(cbstuff.nblks, sizeof(struct CallBack)))) - 1;
2731 FE = ((struct FileEntry
2732 *)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
2733 CBfree = (struct CallBack *)itocb(freelisthead);
2734 read(fd, &freelisthead, sizeof(freelisthead));
2735 FEfree = (struct FileEntry *)itofe(freelisthead);
2736 read(fd, HashTable, sizeof(HashTable));
2737 read(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2738 read(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2740 perror("Error reading dumpfile");
2747 #include "AFS_component_version_number.h"
2749 #include "AFS_component_version_number.c"
2752 static afs_uint32 *cbTrack;
2755 main(int argc, char **argv)
2757 int err = 0, cbi = 0, stats = 0, noptions = 0, all = 0, vol = 0, raw = 0;
2759 struct FileEntry *fe;
2760 struct CallBack *cb;
2764 memset(&fid, 0, sizeof(fid));
2767 while (argc && **argv == '-') {
2770 if (!strcmp(*argv, "-host")) {
2776 cbi = atoi(*++argv);
2777 } else if (!strcmp(*argv, "-fid")) {
2783 fid.Volume = atoi(*++argv);
2784 fid.Vnode = atoi(*++argv);
2785 fid.Unique = atoi(*++argv);
2786 } else if (!strcmp(*argv, "-time")) {
2787 fprintf(stderr, "-time not supported\n");
2789 } else if (!strcmp(*argv, "-stats")) {
2791 } else if (!strcmp(*argv, "-all")) {
2793 } else if (!strcmp(*argv, "-raw")) {
2795 } else if (!strcmp(*argv, "-timebits")) {
2801 timebits = atoi(*++argv);
2802 if ((timebits != 32)
2806 } else if (!strcmp(*argv, "-volume")) {
2812 vol = atoi(*++argv);
2817 if (err || argc != 1) {
2819 "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] [-timebits 32"
2821 "] callbackdumpfile\n");
2823 "[cbid is shown for each host in the hosts.dump file]\n");
2826 now = ReadDump(*argv, timebits);
2827 if (stats || noptions == 0) {
2828 time_t uxtfirst = UXtime(tfirst), tnow = now;
2829 printf("The time of the dump was %u %s", (unsigned int) now, ctime(&tnow));
2830 printf("The last time cleanup ran was %u %s", (unsigned int) uxtfirst,
2832 PrintCallBackStats();
2835 cbTrack = calloc(cbstuff.nblks, sizeof(cbTrack[0]));
2840 struct CallBack *cb;
2841 struct FileEntry *fe;
2843 for (hash = 0; hash < FEHASH_SIZE; hash++) {
2844 for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
2845 if (!vol || (fe->volid == vol)) {
2846 afs_uint32 fe_i = fetoi(fe);
2848 for (cb = itocb(fe->firstcb); cb; cb = itocb(cb->cnext)) {
2849 afs_uint32 cb_i = cbtoi(cb);
2851 if (cb_i > cbstuff.nblks) {
2852 printf("CB index out of range (%u > %d), stopped for this FE\n",
2853 cb_i, cbstuff.nblks);
2857 if (cbTrack[cb_i]) {
2858 printf("CB entry already claimed for FE[%u] (this is FE[%u]), stopped\n",
2859 cbTrack[cb_i], fe_i);
2862 cbTrack[cb_i] = fe_i;
2874 afs_uint32 cfirst = cbi;
2879 } while (cbi != cfirst);
2884 printf("No callback entries for %u.%u\n", fid.Volume, fid.Vnode);
2887 cb = itocb(fe->firstcb);
2890 cb = itocb(cb->cnext);
2895 for (i = 1; i < cbstuff.nblks; i++) {
2896 p = (afs_int32 *) & FE[i];
2897 printf("%d:%12x%12x%12x%12x\n", i, p[0], p[1], p[2], p[3]);
2906 PrintCB(struct CallBack *cb, afs_uint32 now)
2908 struct FileEntry *fe = itofe(cb->fhead);
2909 time_t expires = TIndexToTime(cb->thead);
2914 printf("vol=%u vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %lu secs at %s",
2915 fe->volid, fe->vnode, fe->ncbs, cb->hhead, cb->status, fe->status,
2916 expires - now, ctime(&expires));
2921 #if !defined(INTERPRET_DUMP)
2923 ** try breaking calbacks on afidp from host. Use multi_rx.
2924 ** return 0 on success, non-zero on failure
2927 MultiBreakCallBackAlternateAddress(struct host *host, struct AFSCBFids *afidp)
2931 retVal = MultiBreakCallBackAlternateAddress_r(host, afidp);
2937 MultiBreakCallBackAlternateAddress_r(struct host *host,
2938 struct AFSCBFids *afidp)
2941 struct rx_connection **conns;
2942 struct rx_connection *connSuccess = 0;
2943 struct AddrPort *interfaces;
2944 static struct rx_securityClass *sc = 0;
2945 static struct AFSCBs tc = { 0, 0 };
2948 /* nothing more can be done */
2949 if (!host->interface)
2950 return 1; /* failure */
2952 /* the only address is the primary interface */
2953 if (host->interface->numberOfInterfaces <= 1)
2954 return 1; /* failure */
2956 /* initialise a security object only once */
2958 sc = rxnull_NewClientSecurityObject();
2960 i = host->interface->numberOfInterfaces;
2961 interfaces = calloc(i, sizeof(struct AddrPort));
2962 conns = calloc(i, sizeof(struct rx_connection *));
2963 if (!interfaces || !conns) {
2965 ("Failed malloc in MultiBreakCallBackAlternateAddress_r\n"));
2966 osi_Panic("Failed malloc in MultiBreakCallBackAlternateAddress_r\n");
2969 /* initialize alternate rx connections */
2970 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
2971 /* this is the current primary address */
2972 if (host->host == host->interface->interface[i].addr &&
2973 host->port == host->interface->interface[i].port)
2976 interfaces[j] = host->interface->interface[i];
2978 rx_NewConnection(interfaces[j].addr,
2979 interfaces[j].port, 1, sc, 0);
2980 rx_SetConnDeadTime(conns[j], 2);
2981 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
2985 osi_Assert(j); /* at least one alternate address */
2987 ("Starting multibreakcall back on all addr for host %p (%s:%d)\n",
2988 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
2990 multi_Rx(conns, j) {
2991 multi_RXAFSCB_CallBack(afidp, &tc);
2995 if (host->callback_rxcon)
2996 rx_DestroyConnection(host->callback_rxcon);
2997 host->callback_rxcon = conns[multi_i];
2998 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
2999 host->host = interfaces[multi_i].addr;
3000 host->port = interfaces[multi_i].port;
3001 h_AddHostToAddrHashTable_r(host->host, host->port, host);
3002 connSuccess = conns[multi_i];
3003 rx_SetConnDeadTime(host->callback_rxcon, 50);
3004 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3006 ("multibreakcall success with addr %s:%d\n",
3007 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3008 ntohs(interfaces[multi_i].port)));
3015 /* Destroy all connections except the one on which we succeeded */
3016 for (i = 0; i < j; i++)
3017 if (conns[i] != connSuccess)
3018 rx_DestroyConnection(conns[i]);
3024 return 0; /* success */
3026 return 1; /* failure */
3031 ** try multi_RX probes to host.
3032 ** return 0 on success, non-0 on failure
3035 MultiProbeAlternateAddress_r(struct host *host)
3038 struct rx_connection **conns;
3039 struct rx_connection *connSuccess = 0;
3040 struct AddrPort *interfaces;
3041 static struct rx_securityClass *sc = 0;
3044 /* nothing more can be done */
3045 if (!host->interface)
3046 return 1; /* failure */
3048 /* the only address is the primary interface */
3049 if (host->interface->numberOfInterfaces <= 1)
3050 return 1; /* failure */
3052 /* initialise a security object only once */
3054 sc = rxnull_NewClientSecurityObject();
3056 i = host->interface->numberOfInterfaces;
3057 interfaces = calloc(i, sizeof(struct AddrPort));
3058 conns = calloc(i, sizeof(struct rx_connection *));
3059 if (!interfaces || !conns) {
3060 ViceLog(0, ("Failed malloc in MultiProbeAlternateAddress_r\n"));
3061 osi_Panic("Failed malloc in MultiProbeAlternateAddress_r\n");
3064 /* initialize alternate rx connections */
3065 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3066 /* this is the current primary address */
3067 if (host->host == host->interface->interface[i].addr &&
3068 host->port == host->interface->interface[i].port)
3071 interfaces[j] = host->interface->interface[i];
3073 rx_NewConnection(interfaces[j].addr,
3074 interfaces[j].port, 1, sc, 0);
3075 rx_SetConnDeadTime(conns[j], 2);
3076 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3080 osi_Assert(j); /* at least one alternate address */
3082 ("Starting multiprobe on all addr for host %p (%s:%d)\n",
3083 host, afs_inet_ntoa_r(host->host, hoststr),
3084 ntohs(host->port)));
3086 multi_Rx(conns, j) {
3087 multi_RXAFSCB_ProbeUuid(&host->interface->uuid);
3091 if (host->callback_rxcon)
3092 rx_DestroyConnection(host->callback_rxcon);
3093 host->callback_rxcon = conns[multi_i];
3094 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
3095 host->host = interfaces[multi_i].addr;
3096 host->port = interfaces[multi_i].port;
3097 h_AddHostToAddrHashTable_r(host->host, host->port, host);
3098 connSuccess = conns[multi_i];
3099 rx_SetConnDeadTime(host->callback_rxcon, 50);
3100 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3102 ("multiprobe success with addr %s:%d\n",
3103 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3104 ntohs(interfaces[multi_i].port)));
3109 ("multiprobe failure with addr %s:%d\n",
3110 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3111 ntohs(interfaces[multi_i].port)));
3113 /* This is less than desirable but its the best we can do.
3114 * The AFS Cache Manager will return either 0 for a Uuid
3115 * match and a 1 for a non-match. If the error is 1 we
3116 * therefore know that our mapping of IP address to Uuid
3117 * is wrong. We should attempt to find the correct
3118 * Uuid and fix the host tables.
3120 if (multi_error == 1) {
3121 /* remove the current alternate address from this host */
3123 removeInterfaceAddr_r(host, interfaces[multi_i].addr, interfaces[multi_i].port);
3127 #ifdef AFS_DEMAND_ATTACH_FS
3128 /* try to bail ASAP if the fileserver is shutting down */
3130 if (fs_state.mode == FS_MODE_SHUTDOWN) {
3139 /* Destroy all connections except the one on which we succeeded */
3140 for (i = 0; i < j; i++)
3141 if (conns[i] != connSuccess)
3142 rx_DestroyConnection(conns[i]);
3148 return 0; /* success */
3150 return 1; /* failure */
3153 #endif /* !defined(INTERPRET_DUMP) */