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 ViceLogThenPanic(0, ("CDel: Internal Error -- shutting down: "
350 "wanted %d from %d, now at %d\n",
351 cbi, fe->firstcb, *cbp));
352 DumpCallBackState_r();
353 ShutDownAndCore(PANIC);
356 CDelPtr(fe, cbp, deletefe);
360 /* Same as CDel, but pointer to parent pointer to CB entry is passed,
361 * as well as file entry */
362 /* N.B. This one also deletes the CB, and also possibly parent FE, so
363 * make sure that it is not on any other list before calling this
365 static int Ccdelpt = 0, CcdelB = 0;
368 CDelPtr(struct FileEntry *fe, afs_uint32 * cbp,
381 if ((--fe->ncbs == 0) && deletefe)
387 FindCBPtr(struct FileEntry *fe, struct host *host)
389 afs_uint32 hostindex = h_htoi(host);
394 for (safety = 0, cbp = &fe->firstcb; *cbp; cbp = &cb->cnext, safety++) {
395 if (safety > cbstuff.nblks) {
396 ViceLog(0, ("FindCBPtr: Internal Error -- shutting down.\n"));
397 DumpCallBackState_r();
398 ShutDownAndCore(PANIC);
401 if (cb->hhead == hostindex)
407 /* Delete file entry from hash table */
409 FDel(struct FileEntry *fe)
412 afs_uint32 *p = &HashTable[FEHash(fe->volid, fe->unique)];
414 while (*p && *p != fei)
415 p = &itofe(*p)->fnext;
422 /* initialize the callback package */
424 InitCallBack(int nblks)
427 tfirst = CBtime(FT_ApproxTime());
428 /* N.B. The "-1", below, is because
429 * FE[0] and CB[0] are not used--and not allocated */
430 FE = ((struct FileEntry *)(calloc(nblks, sizeof(struct FileEntry))));
432 ViceLogThenPanic(0, ("Failed malloc in InitCallBack\n"));
434 FE--; /* FE[0] is supposed to point to junk */
435 cbstuff.nFEs = nblks;
437 FreeFE(&FE[cbstuff.nFEs]); /* This is correct */
438 CB = ((struct CallBack *)(calloc(nblks, sizeof(struct CallBack))));
440 ViceLogThenPanic(0, ("Failed malloc in InitCallBack\n"));
442 CB--; /* CB[0] is supposed to point to junk */
443 cbstuff.nCBs = nblks;
445 FreeCB(&CB[cbstuff.nCBs]); /* This is correct */
446 cbstuff.nblks = nblks;
447 cbstuff.nbreakers = 0;
453 XCallBackBulk_r(struct host * ahost, struct AFSFid * fids, afs_int32 nfids)
455 struct AFSCallBack tcbs[AFSCBMAX];
461 struct rx_connection *cb_conn = NULL;
464 rx_SetConnDeadTime(ahost->callback_rxcon, 4);
465 rx_SetConnHardDeadTime(ahost->callback_rxcon, AFS_HARDDEADTIME);
472 for (i = 0; i < nfids && i < AFSCBMAX; i++) {
473 tcbs[i].CallBackVersion = CALLBACK_VERSION;
474 tcbs[i].ExpirationTime = 0;
475 tcbs[i].CallBackType = CB_DROPPED;
477 tf.AFSCBFids_len = i;
478 tf.AFSCBFids_val = &(fids[j]);
482 tc.AFSCBs_val = tcbs;
484 cb_conn = ahost->callback_rxcon;
485 rx_GetConnection(cb_conn);
487 code |= RXAFSCB_CallBack(cb_conn, &tf, &tc);
488 rx_PutConnection(cb_conn);
496 /* the locked flag tells us if the host entry has already been locked
497 * by our parent. I don't think anybody actually calls us with the
498 * host locked, but here's how to make that work: GetSomeSpace has to
499 * change so that it doesn't attempt to lock any hosts < "host". That
500 * means that it might be unable to free any objects, so it has to
501 * return an exit status. If it fails, then AddCallBack1 might fail,
502 * as well. If so, the host->ResetDone should probably be set to 0,
503 * and we probably don't want to return a callback promise to the
504 * cache manager, either. */
506 AddCallBack1(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
514 if (!(host->hostFlags & HOSTDELETED))
515 retVal = AddCallBack1_r(host, fid, thead, type, 1);
525 AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
528 struct FileEntry *fe;
529 struct CallBack *cb = 0, *lastcb = 0;
530 struct FileEntry *newfe = 0;
531 afs_uint32 time_out = 0;
532 afs_uint32 *Thead = thead;
533 struct CallBack *newcb = 0;
536 cbstuff.AddCallBacks++;
540 /* allocate these guys first, since we can't call the allocator with
541 * the host structure locked -- or we might deadlock. However, we have
542 * to avoid races with FindFE... */
543 while (!(newcb = GetCB())) {
544 GetSomeSpace_r(host, locked);
546 while (!(newfe = GetFE())) { /* Get it now, so we don't have to call */
547 /* GetSomeSpace with the host locked, later. This might turn out to */
548 /* have been unneccessary, but that's actually kind of unlikely, since */
549 /* most files are not shared. */
550 GetSomeSpace_r(host, locked);
554 h_Lock_r(host); /* this can yield, so do it before we get any */
556 if (host->hostFlags & HOSTDELETED) {
564 if (type == CB_NORMAL) {
566 TimeCeiling(FT_ApproxTime() + TimeOut(fe ? fe->ncbs : 0) +
568 Thead = THead(CBtime(time_out));
569 } else if (type == CB_VOLUME) {
570 time_out = TimeCeiling((60 * 120 + FT_ApproxTime()) + ServerBias);
571 Thead = THead(CBtime(time_out));
572 } else if (type == CB_BULK) {
573 /* bulk status can get so many callbacks all at once, and most of them
574 * are probably not for things that will be used for long.
577 TimeCeiling(FT_ApproxTime() + ServerBias +
578 TimeOut(22 + (fe ? fe->ncbs : 0)));
579 Thead = THead(CBtime(time_out));
590 fe->volid = fid->Volume;
591 fe->vnode = fid->Vnode;
592 fe->unique = fid->Unique;
595 hash = FEHash(fid->Volume, fid->Unique);
596 fe->fnext = HashTable[hash];
597 HashTable[hash] = fetoi(fe);
599 for (safety = 0, lastcb = cb = itocb(fe->firstcb); cb;
600 lastcb = cb, cb = itocb(cb->cnext), safety++) {
601 if (safety > cbstuff.nblks) {
602 ViceLog(0, ("AddCallBack1: Internal Error -- shutting down.\n"));
603 DumpCallBackState_r();
604 ShutDownAndCore(PANIC);
606 if (cb->hhead == h_htoi(host))
609 if (cb) { /* Already have call back: move to new timeout list */
610 /* don't change delayed callbacks back to normal ones */
611 if (cb->status != CB_DELAYED)
613 /* Only move if new timeout is longer */
614 if (TNorm(ttoi(Thead)) > TNorm(cb->thead)) {
618 if (newfe == NULL) { /* we are using the new FE */
619 fe->firstcb = cbtoi(cb);
621 cb->fhead = fetoi(fe);
626 *(lastcb ? &lastcb->cnext : &fe->firstcb) = cbtoi(cb);
629 cb->fhead = fetoi(fe);
635 /* now free any still-unused callback or host entries */
641 if (!locked) /* freecb and freefe might(?) yield */
644 if (type == CB_NORMAL || type == CB_VOLUME || type == CB_BULK)
645 return time_out - ServerBias; /* Expires sooner at workstation */
651 CompareCBA(const void *e1, const void *e2)
653 const struct cbstruct *cba1 = (const struct cbstruct *)e1;
654 const struct cbstruct *cba2 = (const struct cbstruct *)e2;
655 return ((cba1->hp)->index - (cba2->hp)->index);
658 /* Take an array full of hosts, all held. Break callbacks to them, and
659 * release the holds once you're done, except don't release xhost. xhost
660 * may be NULL. Currently only works for a single Fid in afidp array.
661 * If you want to make this work with multiple fids, you need to fix
662 * the error handling. One approach would be to force a reset if a
663 * multi-fid call fails, or you could add delayed callbacks for each
664 * fid. You probably also need to sort and remove duplicate hosts.
665 * When this is called from the BreakVolumeCallBacks path, it does NOT
666 * force a reset if the RPC fails, it just marks the host down and tries
667 * to create a delayed callback. */
668 /* N.B. be sure that code works when ncbas == 0 */
669 /* N.B. requires all the cba[*].hp pointers to be valid... */
670 /* This routine does not hold a lock on the host for the duration of
671 * the BreakCallBack RPC, which is a significant deviation from tradition.
672 * It _does_ get a lock on the host before setting VenusDown = 1,
673 * which is sufficient only if VenusDown = 0 only happens when the
674 * lock is held over the RPC and the subsequent VenusDown == 0
675 * wherever that is done. */
677 MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
678 struct AFSCBFids *afidp, struct host *xhost)
681 struct rx_connection *conns[MAX_CB_HOSTS];
682 static struct AFSCBs tc = { 0, 0 };
683 int multi_to_cba_map[MAX_CB_HOSTS];
685 osi_Assert(ncbas <= MAX_CB_HOSTS);
687 /* sort cba list to avoid makecall issues */
688 qsort(cba, ncbas, sizeof(struct cbstruct), CompareCBA);
690 /* set up conns for multi-call */
691 for (i = 0, j = 0; i < ncbas; i++) {
692 struct host *thishost = cba[i].hp;
693 if (!thishost || (thishost->hostFlags & HOSTDELETED)) {
696 rx_GetConnection(thishost->callback_rxcon);
697 multi_to_cba_map[j] = i;
698 conns[j++] = thishost->callback_rxcon;
701 rx_SetConnDeadTime(thishost->callback_rxcon, 4);
702 rx_SetConnHardDeadTime(thishost->callback_rxcon, AFS_HARDDEADTIME);
706 if (j) { /* who knows what multi would do with 0 conns? */
710 multi_RXAFSCB_CallBack(afidp, &tc);
716 i = multi_to_cba_map[multi_i];
722 ("BCB: INTERNAL ERROR: hp=%p, cba=%p, thead=%u\n",
726 ** try breaking callbacks on alternate interface addresses
728 if (MultiBreakCallBackAlternateAddress(hp, afidp)) {
731 ("BCB: Failed on file %u.%u.%u, "
732 "Host %p (%s:%d) is down\n",
733 afidp->AFSCBFids_val->Volume,
734 afidp->AFSCBFids_val->Vnode,
735 afidp->AFSCBFids_val->Unique,
737 afs_inet_ntoa_r(hp->host, hoststr),
743 if (!(hp->hostFlags & HOSTDELETED)) {
744 hp->hostFlags |= VENUSDOWN;
746 * We always go into AddCallBack1_r with the host locked
748 AddCallBack1_r(hp, afidp->AFSCBFids_val, itot(idx),
762 for (i = 0; i < ncbas; i++) {
765 if (hp && xhost != hp) {
770 /* H_UNLOCK around this so h_FreeConnection does not deadlock.
771 h_FreeConnection should *never* be called on a callback connection,
772 but on 10/27/04 a deadlock occurred where it was, when we know why,
773 this should be reverted. -- shadow */
775 for (i = 0; i < j; i++) {
776 rx_PutConnection(conns[i]);
784 * Break all call backs for fid, except for the specified host (unless flag
785 * is true, in which case all get a callback message. Assumption: the specified
786 * host is h_Held, by the caller; the others aren't.
787 * Specified host may be bogus, that's ok. This used to check to see if the
788 * host was down in two places, once right after the host was h_held, and
789 * again after it was locked. That race condition is incredibly rare and
790 * relatively harmless even when it does occur, so we don't check for it now.
792 /* if flag is true, send a break callback msg to "host", too */
794 BreakCallBack(struct host *xhost, AFSFid * fid, int flag)
796 struct FileEntry *fe;
797 struct CallBack *cb, *nextcb;
798 struct cbstruct cbaDef[MAX_CB_HOSTS], *cba = cbaDef;
799 unsigned int ncbas, cbaAlloc = MAX_CB_HOSTS;
805 ("BCB: BreakCallBack(Host %p all but %s:%d, (%u,%u,%u))\n",
806 xhost, afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
807 fid->Volume, fid->Vnode, fid->Unique));
810 cbstuff.BreakCallBacks++;
815 hostindex = h_htoi(xhost);
816 cb = itocb(fe->firstcb);
817 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
818 /* the most common case is what follows the || */
821 tf.AFSCBFids_len = 1;
822 tf.AFSCBFids_val = fid;
824 for (ncbas = 0; cb ; cb = nextcb) {
825 nextcb = itocb(cb->cnext);
826 if ((cb->hhead != hostindex || flag)
827 && (cb->status == CB_BULK || cb->status == CB_NORMAL
828 || cb->status == CB_VOLUME)) {
829 struct host *thishost = h_itoh(cb->hhead);
831 ViceLog(0, ("BCB: BOGUS! cb->hhead is NULL!\n"));
832 } else if (thishost->hostFlags & VENUSDOWN) {
834 ("BCB: %p (%s:%d) is down; delaying break call back\n",
835 thishost, afs_inet_ntoa_r(thishost->host, hoststr),
836 ntohs(thishost->port)));
837 cb->status = CB_DELAYED;
839 if (!(thishost->hostFlags & HOSTDELETED)) {
841 if (ncbas == cbaAlloc) { /* Need more space */
842 int curLen = cbaAlloc*sizeof(cba[0]);
843 struct cbstruct *cbaOld = (cba == cbaDef) ? NULL : cba;
845 /* There are logical contraints elsewhere that the number of hosts
846 (i.e. h_HTSPERBLOCK*h_MAXHOSTTABLES) remains in the realm of a signed "int".
847 cbaAlloc is defined unsigned int hence doubling below cannot overflow
849 cbaAlloc = cbaAlloc<<1; /* double */
850 cba = realloc(cbaOld, cbaAlloc * sizeof(cba[0]));
852 if (cbaOld == NULL) { /* realloc wouldn't have copied from cbaDef */
853 memcpy(cba, cbaDef, curLen);
856 cba[ncbas].hp = thishost;
857 cba[ncbas].thead = cb->thead;
862 CDel(cb, 1); /* Usually first; so this delete
863 * is reasonably inexpensive */
869 struct cbstruct *cba2;
872 for (cba2 = cba, num = ncbas; ncbas > 0; cba2 += num, ncbas -= num) {
873 num = (ncbas > MAX_CB_HOSTS) ? MAX_CB_HOSTS : ncbas;
874 MultiBreakCallBack_r(cba2, num, &tf, xhost);
878 if (cba != cbaDef) free(cba);
885 /* Delete (do not break) single call back for fid */
887 DeleteCallBack(struct host *host, AFSFid * fid)
889 struct FileEntry *fe;
894 cbstuff.DeleteCallBacks++;
897 /* do not care if the host has been HOSTDELETED */
903 ("DCB: No call backs for fid (%u, %u, %u)\n", fid->Volume,
904 fid->Vnode, fid->Unique));
907 pcb = FindCBPtr(fe, host);
910 ("DCB: No call back for host %p (%s:%d), (%u, %u, %u)\n",
911 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
912 fid->Volume, fid->Vnode, fid->Unique));
926 * Delete (do not break) all call backs for fid. This call doesn't
927 * set all of the various host locks, but it shouldn't really matter
928 * since we're not adding callbacks, but deleting them. I'm not sure
929 * why it doesn't set the lock, however; perhaps it should.
932 DeleteFileCallBacks(AFSFid * fid)
934 struct FileEntry *fe;
940 cbstuff.DeleteFiles++;
945 ("DF: No fid (%u,%u,%u) to delete\n", fid->Volume, fid->Vnode,
949 for (n = 0, cbi = fe->firstcb; cbi; n++) {
962 /* Delete (do not break) all call backs for host. The host should be
965 DeleteAllCallBacks_r(struct host *host, int deletefe)
970 cbstuff.DeleteAllCallBacks++;
971 cbi = first = host->cblist;
973 ViceLog(8, ("DV: no call backs\n"));
981 } while (cbi != first);
987 * Break all delayed call backs for host. Returns 1 if all call backs
988 * successfully broken; 0 otherwise. Assumes host is h_Held and h_Locked.
989 * Must be called with VenusDown set for this host
992 BreakDelayedCallBacks(struct host *host)
996 retVal = BreakDelayedCallBacks_r(host);
1002 BreakDelayedCallBacks_r(struct host *host)
1004 struct AFSFid fids[AFSCBMAX];
1005 int cbi, first, nfids;
1006 struct CallBack *cb;
1009 struct rx_connection *cb_conn;
1011 cbstuff.nbreakers++;
1012 if (!(host->hostFlags & RESETDONE) && !(host->hostFlags & HOSTDELETED)) {
1013 host->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1014 cb_conn = host->callback_rxcon;
1015 rx_GetConnection(cb_conn);
1016 if (host->interface) {
1019 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1022 code = RXAFSCB_InitCallBackState(cb_conn);
1024 rx_PutConnection(cb_conn);
1027 host->hostFlags |= ALTADDR; /* alternate addresses are valid */
1031 ("CB: Call back connect back failed (in break delayed) "
1032 "for Host %p (%s:%d)\n",
1033 host, afs_inet_ntoa_r(host->host, hoststr),
1034 ntohs(host->port)));
1036 host->hostFlags |= VENUSDOWN;
1039 ("InitCallBackState success on %p (%s:%d)\n",
1040 host, afs_inet_ntoa_r(host->host, hoststr),
1041 ntohs(host->port)));
1042 /* reset was done successfully */
1043 host->hostFlags |= RESETDONE;
1044 host->hostFlags &= ~VENUSDOWN;
1047 while (!(host->hostFlags & HOSTDELETED)) {
1049 host->hostFlags &= ~VENUSDOWN; /* presume up */
1050 cbi = first = host->cblist;
1054 first = host->cblist;
1057 if (cb->status == CB_DELAYED) {
1058 struct FileEntry *fe = itofe(cb->fhead);
1059 fids[nfids].Volume = fe->volid;
1060 fids[nfids].Vnode = fe->vnode;
1061 fids[nfids].Unique = fe->unique;
1067 } while (cbi && cbi != first && nfids < AFSCBMAX);
1073 if (XCallBackBulk_r(host, fids, nfids)) {
1074 /* Failed, again: put them back, probably with old
1079 ("CB: XCallBackBulk failed, Host %p (%s:%d); "
1080 "callback list follows:\n",
1081 host, afs_inet_ntoa_r(host->host, hoststr),
1082 ntohs(host->port)));
1084 for (i = 0; i < nfids; i++) {
1087 ("CB: Host %p (%s:%d), file %u.%u.%u "
1088 "(part of bulk callback)\n",
1089 host, afs_inet_ntoa_r(host->host, hoststr),
1090 ntohs(host->port), fids[i].Volume,
1091 fids[i].Vnode, fids[i].Unique));
1094 * AddCallBack1_r(host, &fids[i], itot(thead[i]), CB_DELAYED, 1);
1095 * * but it turns out to cause too many tricky locking problems.
1096 * * now, if break delayed fails, screw it. */
1098 host->hostFlags |= VENUSDOWN; /* Failed */
1099 ClearHostCallbacks_r(host, 1 /* locked */ );
1103 if (nfids < AFSCBMAX)
1107 cbstuff.nbreakers--;
1108 /* If we succeeded it's always ok to unset HFE_LATER */
1109 if (!(host->hostFlags & VENUSDOWN))
1110 host->hostFlags &= ~HFE_LATER;
1111 return (host->hostFlags & VENUSDOWN);
1115 ** isheld is 0 if the host is held in h_Enumerate
1116 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1119 MultiBreakVolumeCallBack_r(struct host *host, int isheld,
1120 struct VCBParams *parms, int deletefe)
1124 if (host->hostFlags & HOSTDELETED)
1125 return 0; /* host is deleted, release hold */
1127 if (!(host->hostFlags & HCBREAK))
1128 return 0; /* host is not flagged to notify */
1130 if (host->hostFlags & VENUSDOWN) {
1132 /* Do not care if the host is now HOSTDELETED */
1135 ("BVCB: volume callback for Host %p (%s:%d) failed\n",
1136 host, afs_inet_ntoa_r(host->host, hoststr),
1137 ntohs(host->port)));
1139 DeleteAllCallBacks_r(host, deletefe); /* Delete all callback state
1140 * rather than attempting to
1141 * selectively remember to
1142 * delete the volume callbacks
1144 host->hostFlags &= ~(RESETDONE|HCBREAK); /* Do InitCallBackState when host returns */
1146 return 0; /* parent will release hold */
1148 osi_Assert(parms->ncbas <= MAX_CB_HOSTS);
1150 /* Do not call MultiBreakCallBack on the current host structure
1151 ** because it would prematurely release the hold on the host
1153 if (parms->ncbas == MAX_CB_HOSTS) {
1154 struct AFSCBFids tf;
1156 tf.AFSCBFids_len = 1;
1157 tf.AFSCBFids_val = parms->fid;
1159 /* this releases all the hosts */
1160 MultiBreakCallBack_r(parms->cba, parms->ncbas, &tf, 0 /* xhost */ );
1164 parms->cba[parms->ncbas].hp = host;
1165 parms->cba[(parms->ncbas)++].thead = parms->thead;
1166 host->hostFlags &= ~HCBREAK;
1167 return 1; /* parent shouldn't release hold, more work to do */
1171 ** isheld is 0 if the host is held in h_Enumerate
1172 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1175 MultiBreakVolumeLaterCallBack(struct host *host, int isheld, void *rock)
1177 struct VCBParams *parms = (struct VCBParams *)rock;
1180 retval = MultiBreakVolumeCallBack_r(host, isheld, parms, 0);
1186 * Break all call backs on a single volume. Don't call this with any
1187 * hosts h_held. Note that this routine clears the callbacks before
1188 * actually breaking them, and that the vnode isn't locked during this
1189 * operation, so that people might see temporary callback loss while
1190 * this function is executing. It is just a temporary state, however,
1191 * since the callback will be broken later by this same function.
1193 * Now uses multi-RX for CallBack RPC in a different thread,
1194 * only marking them here.
1196 #ifdef AFS_PTHREAD_ENV
1197 extern pthread_cond_t fsync_cond;
1199 extern char fsync_wait[];
1203 BreakVolumeCallBacksLater(afs_uint32 volume)
1207 struct FileEntry *fe;
1208 struct CallBack *cb;
1212 ViceLog(25, ("Setting later on volume %u\n", volume));
1214 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1215 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1216 if (fe->volid == volume) {
1217 struct CallBack *cbnext;
1218 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1219 host = h_itoh(cb->hhead);
1220 host->hostFlags |= HFE_LATER;
1221 cb->status = CB_DELAYED;
1222 cbnext = itocb(cb->cnext);
1225 fe->status |= FE_LATER;
1234 /* didn't find any callbacks, so return right away. */
1238 ViceLog(25, ("Fsync thread wakeup\n"));
1239 #ifdef AFS_PTHREAD_ENV
1241 CV_BROADCAST(&fsync_cond);
1244 LWP_NoYieldSignal(fsync_wait);
1250 BreakLaterCallBacks(void)
1255 struct CallBack *cb;
1256 struct FileEntry *fe = NULL;
1257 struct FileEntry *myfe = NULL;
1259 struct VCBParams henumParms;
1260 unsigned short tthead = 0; /* zero is illegal value */
1264 ViceLog(25, ("Looking for FileEntries to unchain\n"));
1267 /* Pick the first volume we see to clean up */
1268 fid.Volume = fid.Vnode = fid.Unique = 0;
1270 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1271 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1272 if (fe && (fe->status & FE_LATER)
1273 && (fid.Volume == 0 || fid.Volume == fe->volid)) {
1274 /* Ugly, but used to avoid left side casting */
1275 struct object *tmpfe;
1277 ("Unchaining for %u:%u:%u\n", fe->vnode, fe->unique,
1279 fid.Volume = fe->volid;
1281 fe->status &= ~FE_LATER; /* not strictly needed */
1282 /* Works since volid is deeper than the largest pointer */
1283 tmpfe = (struct object *)fe;
1284 tmpfe->next = (struct object *)myfe;
1297 /* loop over FEs from myfe and free/break */
1299 for (fe = myfe; fe;) {
1300 struct CallBack *cbnext;
1301 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1302 cbnext = itocb(cb->cnext);
1303 host = h_itoh(cb->hhead);
1304 if (cb->status == CB_DELAYED) {
1305 if (!(host->hostFlags & HOSTDELETED)) {
1306 /* mark this host for notification */
1307 host->hostFlags |= HCBREAK;
1308 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1314 CDel(cb, 0); /* Don't let CDel clean up the fe */
1315 /* leave flag for MultiBreakVolumeCallBack to clear */
1318 ("Found host %p (%s:%d) non-DELAYED cb for %u:%u:%u\n",
1319 host, afs_inet_ntoa_r(host->host, hoststr),
1320 ntohs(host->port), fe->vnode, fe->unique, fe->volid));
1324 fe = (struct FileEntry *)((struct object *)fe)->next;
1329 ViceLog(125, ("Breaking volume %u\n", fid.Volume));
1330 henumParms.ncbas = 0;
1331 henumParms.fid = &fid;
1332 henumParms.thead = tthead;
1334 h_Enumerate(MultiBreakVolumeLaterCallBack, (char *)&henumParms);
1336 if (henumParms.ncbas) { /* do left-overs */
1337 struct AFSCBFids tf;
1338 tf.AFSCBFids_len = 1;
1339 tf.AFSCBFids_val = &fid;
1341 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf, 0);
1342 henumParms.ncbas = 0;
1347 /* Arrange to be called again */
1352 * Delete all timed-out call back entries (to be called periodically by file
1356 CleanupTimedOutCallBacks(void)
1359 CleanupTimedOutCallBacks_r();
1365 CleanupTimedOutCallBacks_r(void)
1367 afs_uint32 now = CBtime(FT_ApproxTime());
1369 struct CallBack *cb;
1373 while (tfirst <= now) {
1375 cbi = *(thead = THead(tfirst));
1381 ("CCB: deleting timed out call back %x (%s:%d), (%u,%u,%u)\n",
1382 h_itoh(cb->hhead)->host,
1383 afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
1384 h_itoh(cb->hhead)->port, itofe(cb->fhead)->volid,
1385 itofe(cb->fhead)->vnode, itofe(cb->fhead)->unique));
1389 if (ntimedout > cbstuff.nblks) {
1390 ViceLog(0, ("CCB: Internal Error -- shutting down...\n"));
1391 DumpCallBackState_r();
1392 ShutDownAndCore(PANIC);
1394 } while (cbi != *thead);
1399 cbstuff.CBsTimedOut += ntimedout;
1400 ViceLog(7, ("CCB: deleted %d timed out callbacks\n", ntimedout));
1401 return (ntimedout > 0);
1405 * parameters to pass to lih*_r from h_Enumerate_r when trying to find a host
1406 * from which to clear callbacks.
1410 * Points to the least interesting host found; try to clear callbacks on
1411 * this host after h_Enumerate_r(lih*_r)'ing.
1416 * The last host we got from lih*_r, but we couldn't clear its callbacks
1417 * for some reason. Choose the next-best host after this one (with the
1418 * current lih*_r, this means to only select hosts that have an ActiveCall
1419 * newer than lastlih).
1421 struct host *lastlih;
1424 /* Value of host->refCount that allows us to reliably infer that
1425 * host may be held by some other thread */
1426 #define OTHER_MUSTHOLD_LIH 2
1428 /* This version does not allow 'host' to be selected unless its ActiveCall
1429 * is newer than 'params->lastlih' which is the host with the oldest
1430 * ActiveCall from the last pass (if it is provided). We filter out any hosts
1431 * that are are held by other threads.
1433 * There is a small problem here, but it may not be easily fixable. Say we
1434 * select some host A, and give it back to GetSomeSpace_r. GSS_r for some
1435 * reason cannot clear the callbacks on A, and so calls us again with
1436 * lastlih = A. Suppose there is another host B that has the same ActiveCall
1437 * time as A. We will now skip over host B, since
1438 * 'hostB->ActiveCall > hostA->ActiveCall' is not true. This could result in
1439 * us prematurely going to the GSS_r 2nd or 3rd pass, and making us a little
1440 * inefficient. This should be pretty rare, though, except perhaps in cases
1441 * with very small numbers of hosts.
1443 * Also filter out any hosts with HOSTDELETED set. h_Enumerate_r should in
1444 * theory not give these to us anyway, but be paranoid.
1447 lih0_r(struct host *host, int flags, void *rock)
1449 struct lih_params *params = (struct lih_params *)rock;
1451 /* OTHER_MUSTHOLD_LIH is because the h_Enum loop holds us once */
1453 && (!(host->hostFlags & HOSTDELETED))
1454 && (host->refCount < OTHER_MUSTHOLD_LIH)
1455 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1456 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1459 h_Release_r(params->lih); /* release prev host */
1468 /* same as lih0_r, except we do not prevent held hosts from being selected. */
1470 lih1_r(struct host *host, int flags, void *rock)
1472 struct lih_params *params = (struct lih_params *)rock;
1475 && (!(host->hostFlags & HOSTDELETED))
1476 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1477 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1480 h_Release_r(params->lih); /* release prev host */
1489 /* This could be upgraded to get more space each time */
1490 /* first pass: sequentially find the oldest host which isn't held by
1491 anyone for which we can clear callbacks;
1493 /* second pass: sequentially find the oldest host regardless of
1494 whether or not the host is held; skipping 'hostp' */
1495 /* third pass: attempt to clear callbacks from 'hostp' */
1496 /* always called with hostp unlocked */
1498 /* Note: hostlist is ordered most recently created host first and
1499 * its order has no relationship to the most recently used. */
1500 extern struct host *hostList;
1502 GetSomeSpace_r(struct host *hostp, int locked)
1505 struct lih_params params;
1508 cbstuff.GotSomeSpaces++;
1510 ("GSS: First looking for timed out call backs via CleanupCallBacks\n"));
1511 if (CleanupTimedOutCallBacks_r()) {
1517 params.lastlih = NULL;
1522 h_Enumerate_r(i == 0 ? lih0_r : lih1_r, hostList, ¶ms);
1525 if (params.lastlih) {
1526 h_Release_r(params.lastlih);
1527 params.lastlih = NULL;
1531 /* note that 'hp' was held by lih*_r; we will need to release it */
1533 if ((hp != hostp) && !ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
1538 params.lastlih = hp;
1539 /* params.lastlih will be released on the next iteration, after
1544 * Next time try getting callbacks from any host even if
1545 * it's held, since the only other option is starvation for
1546 * the file server (i.e. until the callback timeout arrives).
1549 params.lastlih = NULL;
1552 ("GSS: Try harder for longest inactive host cnt= %d\n",
1557 /* Could not obtain space from other hosts, clear hostp's callback state */
1562 ClearHostCallbacks_r(hostp, 1 /*already locked */ );
1569 /* locked - set if caller has already locked the host */
1571 ClearHostCallbacks_r(struct host *hp, int locked)
1575 struct rx_connection *cb_conn = NULL;
1578 ("GSS: Delete longest inactive host %p (%s:%d)\n",
1579 hp, afs_inet_ntoa_r(hp->host, hoststr), ntohs(hp->port)));
1581 if ((hp->hostFlags & HOSTDELETED)) {
1582 /* hp could go away after reacquiring H_LOCK in h_NBLock_r, so we can't
1583 * really use it; its callbacks will get cleared anyway when
1584 * h_TossStuff_r gets its hands on it */
1590 /** Try a non-blocking lock. If the lock is already held return
1591 * after releasing hold on hp
1594 if (h_NBLock_r(hp)) {
1599 if (hp->Console & 2) {
1601 * If the special console field is set it means that a thread
1602 * is waiting in AddCallBack1 after it set pointers to the
1603 * file entry and/or callback entry. Because of the bogus
1604 * usage of h_hold it won't prevent from another thread, this
1605 * one, to remove all the callbacks so just to be safe we keep
1606 * a reference. NOTE, on the last phase we'll free the calling
1607 * host's callbacks but that's ok...
1611 DeleteAllCallBacks_r(hp, 1);
1612 if (hp->hostFlags & VENUSDOWN) {
1613 hp->hostFlags &= ~RESETDONE; /* remember that we must do a reset */
1614 } else if (!(hp->hostFlags & HOSTDELETED)) {
1615 /* host is up, try a call */
1616 hp->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1617 cb_conn = hp->callback_rxcon;
1618 rx_GetConnection(hp->callback_rxcon);
1619 if (hp->interface) {
1622 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1625 code = RXAFSCB_InitCallBackState(cb_conn);
1627 rx_PutConnection(cb_conn);
1630 hp->hostFlags |= ALTADDR; /* alternate addresses are valid */
1632 /* failed, mark host down and need reset */
1633 hp->hostFlags |= VENUSDOWN;
1634 hp->hostFlags &= ~RESETDONE;
1636 /* reset succeeded, we're done */
1637 hp->hostFlags |= RESETDONE;
1646 #endif /* INTERPRET_DUMP */
1650 PrintCallBackStats(void)
1653 "%d add CB, %d break CB, %d del CB, %d del FE, %d CB's timed out, %d space reclaim, %d del host\n",
1654 cbstuff.AddCallBacks, cbstuff.BreakCallBacks,
1655 cbstuff.DeleteCallBacks, cbstuff.DeleteFiles, cbstuff.CBsTimedOut,
1656 cbstuff.GotSomeSpaces, cbstuff.DeleteAllCallBacks);
1657 fprintf(stderr, "%d CBs, %d FEs, (%d of total of %d 16-byte blocks)\n",
1658 cbstuff.nCBs, cbstuff.nFEs, cbstuff.nCBs + cbstuff.nFEs,
1660 fprintf(stderr, "%d GSS1, %d GSS2, %d GSS3, %d GSS4, %d GSS5 (internal counters)\n",
1661 cbstuff.GSS1, cbstuff.GSS2, cbstuff.GSS3, cbstuff.GSS4, cbstuff.GSS5);
1666 #define MAGIC 0x12345678 /* To check byte ordering of dump when it is read in */
1667 #define MAGICV2 0x12345679 /* To check byte ordering & version of dump when it is read in */
1670 #ifndef INTERPRET_DUMP
1672 #ifdef AFS_DEMAND_ATTACH_FS
1675 * callback state serialization
1677 static int cb_stateSaveTimeouts(struct fs_dump_state * state);
1678 static int cb_stateSaveFEHash(struct fs_dump_state * state);
1679 static int cb_stateSaveFEs(struct fs_dump_state * state);
1680 static int cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe);
1681 static int cb_stateRestoreTimeouts(struct fs_dump_state * state);
1682 static int cb_stateRestoreFEHash(struct fs_dump_state * state);
1683 static int cb_stateRestoreFEs(struct fs_dump_state * state);
1684 static int cb_stateRestoreFE(struct fs_dump_state * state);
1685 static int cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
1686 struct iovec * iov, int niovecs);
1688 static int cb_stateVerifyFEHash(struct fs_dump_state * state);
1689 static int cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe);
1690 static int cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe);
1691 static int cb_stateVerifyTimeoutQueues(struct fs_dump_state * state);
1693 static int cb_stateFEToDiskEntry(struct FileEntry *, struct FEDiskEntry *);
1694 static int cb_stateDiskEntryToFE(struct fs_dump_state * state,
1695 struct FEDiskEntry *, struct FileEntry *);
1697 static int cb_stateCBToDiskEntry(struct CallBack *, struct CBDiskEntry *);
1698 static int cb_stateDiskEntryToCB(struct fs_dump_state * state,
1699 struct CBDiskEntry *, struct CallBack *);
1701 static int cb_stateFillHeader(struct callback_state_header * hdr);
1702 static int cb_stateCheckHeader(struct callback_state_header * hdr);
1704 static int cb_stateAllocMap(struct fs_dump_state * state);
1707 cb_stateSave(struct fs_dump_state * state)
1711 AssignInt64(state->eof_offset, &state->hdr->cb_offset);
1713 /* invalidate callback state header */
1714 memset(state->cb_hdr, 0, sizeof(struct callback_state_header));
1715 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1716 sizeof(struct callback_state_header))) {
1721 fs_stateIncEOF(state, sizeof(struct callback_state_header));
1723 /* dump timeout state */
1724 if (cb_stateSaveTimeouts(state)) {
1729 /* dump fe hashtable state */
1730 if (cb_stateSaveFEHash(state)) {
1735 /* dump callback state */
1736 if (cb_stateSaveFEs(state)) {
1741 /* write the callback state header to disk */
1742 cb_stateFillHeader(state->cb_hdr);
1743 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1744 sizeof(struct callback_state_header))) {
1754 cb_stateRestore(struct fs_dump_state * state)
1758 if (fs_stateReadHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1759 sizeof(struct callback_state_header))) {
1764 if (cb_stateCheckHeader(state->cb_hdr)) {
1769 if (cb_stateAllocMap(state)) {
1774 if (cb_stateRestoreTimeouts(state)) {
1779 if (cb_stateRestoreFEHash(state)) {
1784 /* restore FEs and CBs from disk */
1785 if (cb_stateRestoreFEs(state)) {
1790 /* restore the timeout queue heads */
1791 tfirst = state->cb_hdr->tfirst;
1798 cb_stateRestoreIndices(struct fs_dump_state * state)
1801 struct FileEntry * fe;
1802 struct CallBack * cb;
1804 /* restore indices in the FileEntry structures */
1805 for (i = 1; i < state->fe_map.len; i++) {
1806 if (state->fe_map.entries[i].new_idx) {
1807 fe = itofe(state->fe_map.entries[i].new_idx);
1809 /* restore the fe->fnext entry */
1810 if (fe_OldToNew(state, fe->fnext, &fe->fnext)) {
1815 /* restore the fe->firstcb entry */
1816 if (cb_OldToNew(state, fe->firstcb, &fe->firstcb)) {
1823 /* restore indices in the CallBack structures */
1824 for (i = 1; i < state->cb_map.len; i++) {
1825 if (state->cb_map.entries[i].new_idx) {
1826 cb = itocb(state->cb_map.entries[i].new_idx);
1828 /* restore the cb->cnext entry */
1829 if (cb_OldToNew(state, cb->cnext, &cb->cnext)) {
1834 /* restore the cb->fhead entry */
1835 if (fe_OldToNew(state, cb->fhead, &cb->fhead)) {
1840 /* restore the cb->hhead entry */
1841 if (h_OldToNew(state, cb->hhead, &cb->hhead)) {
1846 /* restore the cb->tprev entry */
1847 if (cb_OldToNew(state, cb->tprev, &cb->tprev)) {
1852 /* restore the cb->tnext entry */
1853 if (cb_OldToNew(state, cb->tnext, &cb->tnext)) {
1858 /* restore the cb->hprev entry */
1859 if (cb_OldToNew(state, cb->hprev, &cb->hprev)) {
1864 /* restore the cb->hnext entry */
1865 if (cb_OldToNew(state, cb->hnext, &cb->hnext)) {
1872 /* restore the timeout queue head indices */
1873 for (i = 0; i < state->cb_timeout_hdr->records; i++) {
1874 if (cb_OldToNew(state, timeout[i], &timeout[i])) {
1880 /* restore the FE hash table queue heads */
1881 for (i = 0; i < state->cb_fehash_hdr->records; i++) {
1882 if (fe_OldToNew(state, HashTable[i], &HashTable[i])) {
1893 cb_stateVerify(struct fs_dump_state * state)
1897 if (cb_stateVerifyFEHash(state)) {
1901 if (cb_stateVerifyTimeoutQueues(state)) {
1909 cb_stateVerifyFEHash(struct fs_dump_state * state)
1912 struct FileEntry * fe;
1913 afs_uint32 fei, chain_len;
1915 for (i = 0; i < FEHASH_SIZE; i++) {
1917 for (fei = HashTable[i], fe = itofe(fei);
1919 fei = fe->fnext, fe = itofe(fei)) {
1920 if (fei > cbstuff.nblks) {
1921 ViceLog(0, ("cb_stateVerifyFEHash: error: index out of range (fei=%d)\n", fei));
1925 if (cb_stateVerifyFE(state, fe)) {
1928 if (chain_len > FS_STATE_FE_MAX_HASH_CHAIN_LEN) {
1929 ViceLog(0, ("cb_stateVerifyFEHash: error: hash chain %d length exceeds %d; assuming there's a loop\n",
1930 i, FS_STATE_FE_MAX_HASH_CHAIN_LEN));
1942 cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe)
1946 if ((fe->firstcb && !fe->ncbs) ||
1947 (!fe->firstcb && fe->ncbs)) {
1948 ViceLog(0, ("cb_stateVerifyFE: error: fe->firstcb does not agree with fe->ncbs (fei=%lu, fe->firstcb=%lu, fe->ncbs=%lu)\n",
1949 afs_printable_uint32_lu(fetoi(fe)),
1950 afs_printable_uint32_lu(fe->firstcb),
1951 afs_printable_uint32_lu(fe->ncbs)));
1954 if (cb_stateVerifyFCBList(state, fe)) {
1955 ViceLog(0, ("cb_stateVerifyFE: error: FCBList failed verification (fei=%lu)\n",
1956 afs_printable_uint32_lu(fetoi(fe))));
1964 cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe)
1967 afs_uint32 cbi, fei, chain_len = 0;
1968 struct CallBack * cb;
1972 for (cbi = fe->firstcb, cb = itocb(cbi);
1974 cbi = cb->cnext, cb = itocb(cbi)) {
1975 if (cbi > cbstuff.nblks) {
1976 ViceLog(0, ("cb_stateVerifyFCBList: error: list index out of range (cbi=%d, ncbs=%d)\n",
1977 cbi, cbstuff.nblks));
1981 if (cb->fhead != fei) {
1982 ViceLog(0, ("cb_stateVerifyFCBList: error: cb->fhead != fei (fei=%d, cb->fhead=%d)\n",
1986 if (chain_len > FS_STATE_FCB_MAX_LIST_LEN) {
1987 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (fei=%d); assuming there's a loop\n",
1988 FS_STATE_FCB_MAX_LIST_LEN, fei));
1995 if (fe->ncbs != chain_len) {
1996 ViceLog(0, ("cb_stateVerifyFCBList: error: list length mismatch (len=%d, fe->ncbs=%d)\n",
1997 chain_len, fe->ncbs));
2006 cb_stateVerifyHCBList(struct fs_dump_state * state, struct host * host)
2009 afs_uint32 hi, chain_len, cbi;
2010 struct CallBack *cb, *ncb;
2015 for (cbi = host->cblist, cb = itocb(cbi);
2017 cbi = cb->hnext, cb = ncb) {
2018 if (chain_len && (host->cblist == cbi)) {
2019 /* we've wrapped around the circular list, and everything looks ok */
2022 if (cb->hhead != hi) {
2023 ViceLog(0, ("cb_stateVerifyHCBList: error: incorrect cb->hhead (cbi=%d, h->index=%d, cb->hhead=%d)\n",
2024 cbi, hi, cb->hhead));
2027 if (!cb->hprev || !cb->hnext) {
2028 ViceLog(0, ("cb_stateVerifyHCBList: error: null index in circular list (cbi=%d, h->index=%d)\n",
2033 if ((cb->hprev > cbstuff.nblks) ||
2034 (cb->hnext > cbstuff.nblks)) {
2035 ViceLog(0, ("cb_stateVerifyHCBList: error: list index out of range (cbi=%d, h->index=%d, cb->hprev=%d, cb->hnext=%d, nCBs=%d)\n",
2036 cbi, hi, cb->hprev, cb->hnext, cbstuff.nblks));
2040 ncb = itocb(cb->hnext);
2041 if (cbi != ncb->hprev) {
2042 ViceLog(0, ("cb_stateVerifyHCBList: error: corrupt linked list (cbi=%d, h->index=%d)\n",
2047 if (chain_len > FS_STATE_HCB_MAX_LIST_LEN) {
2048 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (h->index=%d); assuming there's a loop\n",
2049 FS_STATE_HCB_MAX_LIST_LEN, hi));
2061 cb_stateVerifyTimeoutQueues(struct fs_dump_state * state)
2064 afs_uint32 cbi, chain_len;
2065 struct CallBack *cb, *ncb;
2067 for (i = 0; i < CB_NUM_TIMEOUT_QUEUES; i++) {
2069 for (cbi = timeout[i], cb = itocb(cbi);
2071 cbi = cb->tnext, cb = ncb) {
2072 if (chain_len && (cbi == timeout[i])) {
2073 /* we've wrapped around the circular list, and everything looks ok */
2076 if (cbi > cbstuff.nblks) {
2077 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: list index out of range (cbi=%d, tindex=%d)\n",
2082 if (itot(cb->thead) != &timeout[i]) {
2083 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: cb->thead points to wrong timeout queue (tindex=%d, cbi=%d, cb->thead=%d)\n",
2084 i, cbi, cb->thead));
2087 if (!cb->tprev || !cb->tnext) {
2088 ViceLog(0, ("cb_stateVerifyTimeoutQueues: null index in circular list (cbi=%d, tindex=%d)\n",
2093 if ((cb->tprev > cbstuff.nblks) ||
2094 (cb->tnext > cbstuff.nblks)) {
2095 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list index out of range (cbi=%d, tindex=%d, cb->tprev=%d, cb->tnext=%d, nCBs=%d)\n",
2096 cbi, i, cb->tprev, cb->tnext, cbstuff.nblks));
2100 ncb = itocb(cb->tnext);
2101 if (cbi != ncb->tprev) {
2102 ViceLog(0, ("cb_stateVerifyTimeoutQueues: corrupt linked list (cbi=%d, tindex=%d)\n",
2107 if (chain_len > FS_STATE_TCB_MAX_LIST_LEN) {
2108 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list length exceeds %d (tindex=%d); assuming there's a loop\n",
2109 FS_STATE_TCB_MAX_LIST_LEN, i));
2121 cb_stateSaveTimeouts(struct fs_dump_state * state)
2124 struct iovec iov[2];
2126 AssignInt64(state->eof_offset, &state->cb_hdr->timeout_offset);
2128 memset(state->cb_timeout_hdr, 0, sizeof(struct callback_state_fehash_header));
2129 state->cb_timeout_hdr->magic = CALLBACK_STATE_TIMEOUT_MAGIC;
2130 state->cb_timeout_hdr->records = CB_NUM_TIMEOUT_QUEUES;
2131 state->cb_timeout_hdr->len = sizeof(struct callback_state_timeout_header) +
2132 (state->cb_timeout_hdr->records * sizeof(afs_uint32));
2134 iov[0].iov_base = (char *)state->cb_timeout_hdr;
2135 iov[0].iov_len = sizeof(struct callback_state_timeout_header);
2136 iov[1].iov_base = (char *)timeout;
2137 iov[1].iov_len = sizeof(timeout);
2139 if (fs_stateSeek(state, &state->cb_hdr->timeout_offset)) {
2144 if (fs_stateWriteV(state, iov, 2)) {
2149 fs_stateIncEOF(state, state->cb_timeout_hdr->len);
2156 cb_stateRestoreTimeouts(struct fs_dump_state * state)
2160 if (fs_stateReadHeader(state, &state->cb_hdr->timeout_offset,
2161 state->cb_timeout_hdr,
2162 sizeof(struct callback_state_timeout_header))) {
2167 if (state->cb_timeout_hdr->magic != CALLBACK_STATE_TIMEOUT_MAGIC) {
2171 if (state->cb_timeout_hdr->records != CB_NUM_TIMEOUT_QUEUES) {
2176 len = state->cb_timeout_hdr->records * sizeof(afs_uint32);
2178 if (state->cb_timeout_hdr->len !=
2179 (sizeof(struct callback_state_timeout_header) + len)) {
2184 if (fs_stateRead(state, timeout, len)) {
2194 cb_stateSaveFEHash(struct fs_dump_state * state)
2197 struct iovec iov[2];
2199 AssignInt64(state->eof_offset, &state->cb_hdr->fehash_offset);
2201 memset(state->cb_fehash_hdr, 0, sizeof(struct callback_state_fehash_header));
2202 state->cb_fehash_hdr->magic = CALLBACK_STATE_FEHASH_MAGIC;
2203 state->cb_fehash_hdr->records = FEHASH_SIZE;
2204 state->cb_fehash_hdr->len = sizeof(struct callback_state_fehash_header) +
2205 (state->cb_fehash_hdr->records * sizeof(afs_uint32));
2207 iov[0].iov_base = (char *)state->cb_fehash_hdr;
2208 iov[0].iov_len = sizeof(struct callback_state_fehash_header);
2209 iov[1].iov_base = (char *)HashTable;
2210 iov[1].iov_len = sizeof(HashTable);
2212 if (fs_stateSeek(state, &state->cb_hdr->fehash_offset)) {
2217 if (fs_stateWriteV(state, iov, 2)) {
2222 fs_stateIncEOF(state, state->cb_fehash_hdr->len);
2229 cb_stateRestoreFEHash(struct fs_dump_state * state)
2233 if (fs_stateReadHeader(state, &state->cb_hdr->fehash_offset,
2234 state->cb_fehash_hdr,
2235 sizeof(struct callback_state_fehash_header))) {
2240 if (state->cb_fehash_hdr->magic != CALLBACK_STATE_FEHASH_MAGIC) {
2244 if (state->cb_fehash_hdr->records != FEHASH_SIZE) {
2249 len = state->cb_fehash_hdr->records * sizeof(afs_uint32);
2251 if (state->cb_fehash_hdr->len !=
2252 (sizeof(struct callback_state_fehash_header) + len)) {
2257 if (fs_stateRead(state, HashTable, len)) {
2267 cb_stateSaveFEs(struct fs_dump_state * state)
2271 struct FileEntry *fe;
2273 AssignInt64(state->eof_offset, &state->cb_hdr->fe_offset);
2275 for (hash = 0; hash < FEHASH_SIZE ; hash++) {
2276 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
2278 if (cb_stateSaveFE(state, fe)) {
2290 cb_stateRestoreFEs(struct fs_dump_state * state)
2292 int count, nFEs, ret = 0;
2294 nFEs = state->cb_hdr->nFEs;
2296 for (count = 0; count < nFEs; count++) {
2297 if (cb_stateRestoreFE(state)) {
2308 cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe)
2310 int ret = 0, iovcnt, cbi, written = 0;
2312 struct callback_state_entry_header hdr;
2313 struct FEDiskEntry fedsk;
2314 struct CBDiskEntry cbdsk[16];
2315 struct iovec iov[16];
2316 struct CallBack *cb;
2319 if (fei > state->cb_hdr->fe_max) {
2320 state->cb_hdr->fe_max = fei;
2323 memset(&hdr, 0, sizeof(struct callback_state_entry_header));
2325 if (cb_stateFEToDiskEntry(fe, &fedsk)) {
2330 iov[0].iov_base = (char *)&hdr;
2331 iov[0].iov_len = sizeof(hdr);
2332 iov[1].iov_base = (char *)&fedsk;
2333 iov[1].iov_len = sizeof(struct FEDiskEntry);
2336 for (cbi = fe->firstcb, cb = itocb(cbi);
2338 cbi = cb->cnext, cb = itocb(cbi), hdr.nCBs++) {
2339 if (cbi > state->cb_hdr->cb_max) {
2340 state->cb_hdr->cb_max = cbi;
2342 if (cb_stateCBToDiskEntry(cb, &cbdsk[iovcnt])) {
2346 cbdsk[iovcnt].index = cbi;
2347 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2348 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2350 if ((iovcnt == 16) || (!cb->cnext)) {
2351 if (fs_stateWriteV(state, iov, iovcnt)) {
2360 hdr.magic = CALLBACK_STATE_ENTRY_MAGIC;
2361 hdr.len = sizeof(hdr) + sizeof(struct FEDiskEntry) +
2362 (hdr.nCBs * sizeof(struct CBDiskEntry));
2365 if (fs_stateWriteV(state, iov, iovcnt)) {
2370 if (fs_stateWriteHeader(state, &state->eof_offset, &hdr, sizeof(hdr))) {
2376 fs_stateIncEOF(state, hdr.len);
2379 if (fs_stateSeek(state, &state->eof_offset)) {
2385 state->cb_hdr->nFEs++;
2386 state->cb_hdr->nCBs += hdr.nCBs;
2393 cb_stateRestoreFE(struct fs_dump_state * state)
2395 int ret = 0, iovcnt, nCBs;
2396 struct callback_state_entry_header hdr;
2397 struct FEDiskEntry fedsk;
2398 struct CBDiskEntry cbdsk[16];
2399 struct iovec iov[16];
2400 struct FileEntry * fe;
2402 iov[0].iov_base = (char *)&hdr;
2403 iov[0].iov_len = sizeof(hdr);
2404 iov[1].iov_base = (char *)&fedsk;
2405 iov[1].iov_len = sizeof(fedsk);
2408 if (fs_stateReadV(state, iov, iovcnt)) {
2413 if (hdr.magic != CALLBACK_STATE_ENTRY_MAGIC) {
2420 ViceLog(0, ("cb_stateRestoreFE: ran out of free FileEntry structures\n"));
2425 if (cb_stateDiskEntryToFE(state, &fedsk, fe)) {
2431 for (iovcnt = 0, nCBs = 0;
2434 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2435 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2437 if ((iovcnt == 16) || (nCBs == hdr.nCBs - 1)) {
2438 if (fs_stateReadV(state, iov, iovcnt)) {
2442 if (cb_stateRestoreCBs(state, fe, iov, iovcnt)) {
2456 cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
2457 struct iovec * iov, int niovecs)
2460 struct CallBack * cb;
2461 struct CBDiskEntry * cbdsk;
2463 for (idx = 0; idx < niovecs; idx++) {
2464 cbdsk = (struct CBDiskEntry *) iov[idx].iov_base;
2465 if ((cb = GetCB()) == NULL) {
2466 ViceLog(0, ("cb_stateRestoreCBs: ran out of free CallBack structures\n"));
2470 if (cb_stateDiskEntryToCB(state, cbdsk, cb)) {
2471 ViceLog(0, ("cb_stateRestoreCBs: corrupt CallBack disk entry\n"));
2483 cb_stateFillHeader(struct callback_state_header * hdr)
2485 hdr->stamp.magic = CALLBACK_STATE_MAGIC;
2486 hdr->stamp.version = CALLBACK_STATE_VERSION;
2487 hdr->tfirst = tfirst;
2492 cb_stateCheckHeader(struct callback_state_header * hdr)
2496 if (hdr->stamp.magic != CALLBACK_STATE_MAGIC) {
2498 } else if (hdr->stamp.version != CALLBACK_STATE_VERSION) {
2500 } else if ((hdr->nFEs > cbstuff.nblks) || (hdr->nCBs > cbstuff.nblks)) {
2501 ViceLog(0, ("cb_stateCheckHeader: saved callback state larger than callback memory allocation\n"));
2507 /* disk entry conversion routines */
2509 cb_stateFEToDiskEntry(struct FileEntry * in, struct FEDiskEntry * out)
2511 memcpy(&out->fe, in, sizeof(struct FileEntry));
2512 out->index = fetoi(in);
2517 cb_stateDiskEntryToFE(struct fs_dump_state * state,
2518 struct FEDiskEntry * in, struct FileEntry * out)
2522 memcpy(out, &in->fe, sizeof(struct FileEntry));
2524 /* setup FE map entry */
2525 if (!in->index || (in->index >= state->fe_map.len)) {
2526 ViceLog(0, ("cb_stateDiskEntryToFE: index (%d) out of range",
2531 state->fe_map.entries[in->index].old_idx = in->index;
2532 state->fe_map.entries[in->index].new_idx = fetoi(out);
2539 cb_stateCBToDiskEntry(struct CallBack * in, struct CBDiskEntry * out)
2541 memcpy(&out->cb, in, sizeof(struct CallBack));
2542 out->index = cbtoi(in);
2547 cb_stateDiskEntryToCB(struct fs_dump_state * state,
2548 struct CBDiskEntry * in, struct CallBack * out)
2552 memcpy(out, &in->cb, sizeof(struct CallBack));
2554 /* setup CB map entry */
2555 if (!in->index || (in->index >= state->cb_map.len)) {
2556 ViceLog(0, ("cb_stateDiskEntryToCB: index (%d) out of range\n",
2561 state->cb_map.entries[in->index].old_idx = in->index;
2562 state->cb_map.entries[in->index].new_idx = cbtoi(out);
2568 /* index map routines */
2570 cb_stateAllocMap(struct fs_dump_state * state)
2572 state->fe_map.len = state->cb_hdr->fe_max + 1;
2573 state->cb_map.len = state->cb_hdr->cb_max + 1;
2574 state->fe_map.entries = (struct idx_map_entry_t *)
2575 calloc(state->fe_map.len, sizeof(struct idx_map_entry_t));
2576 state->cb_map.entries = (struct idx_map_entry_t *)
2577 calloc(state->cb_map.len, sizeof(struct idx_map_entry_t));
2578 return ((state->fe_map.entries != NULL) && (state->cb_map.entries != NULL)) ? 0 : 1;
2582 fe_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2586 /* FEs use a one-based indexing system, so old==0 implies no mapping */
2592 if (old >= state->fe_map.len) {
2593 ViceLog(0, ("fe_OldToNew: index %d is out of range\n", old));
2595 } else if (state->fe_map.entries[old].old_idx != old) { /* sanity check */
2596 ViceLog(0, ("fe_OldToNew: index %d points to an invalid FileEntry record\n", old));
2599 *new = state->fe_map.entries[old].new_idx;
2607 cb_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2611 /* CBs use a one-based indexing system, so old==0 implies no mapping */
2617 if (old >= state->cb_map.len) {
2618 ViceLog(0, ("cb_OldToNew: index %d is out of range\n", old));
2620 } else if (state->cb_map.entries[old].old_idx != old) { /* sanity check */
2621 ViceLog(0, ("cb_OldToNew: index %d points to an invalid CallBack record\n", old));
2624 *new = state->cb_map.entries[old].new_idx;
2630 #endif /* AFS_DEMAND_ATTACH_FS */
2633 DumpCallBackState_r(void)
2636 afs_uint32 magic = MAGICV2, now = (afs_int32) FT_ApproxTime(), freelisthead;
2638 oflag = O_WRONLY | O_CREAT | O_TRUNC;
2642 fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, oflag, 0666);
2645 ("Couldn't create callback dump file %s\n",
2646 AFSDIR_SERVER_CBKDUMP_FILEPATH));
2649 (void)write(fd, &magic, sizeof(magic));
2650 (void)write(fd, &now, sizeof(now));
2651 (void)write(fd, &cbstuff, sizeof(cbstuff));
2652 (void)write(fd, TimeOuts, sizeof(TimeOuts));
2653 (void)write(fd, timeout, sizeof(timeout));
2654 (void)write(fd, &tfirst, sizeof(tfirst));
2655 freelisthead = cbtoi((struct CallBack *)CBfree);
2656 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2657 freelisthead = fetoi((struct FileEntry *)FEfree);
2658 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2659 (void)write(fd, HashTable, sizeof(HashTable));
2660 (void)write(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2661 (void)write(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2668 DumpCallBackState(void) {
2672 rc = DumpCallBackState_r();
2678 #endif /* !INTERPRET_DUMP */
2680 #ifdef INTERPRET_DUMP
2682 /* This is only compiled in for the callback analyzer program */
2683 /* Returns the time of the dump */
2685 ReadDump(char *file, int timebits)
2688 afs_uint32 magic, freelisthead;
2696 fd = open(file, oflag);
2698 fprintf(stderr, "Couldn't read dump file %s\n", file);
2701 read(fd, &magic, sizeof(magic));
2702 if (magic == MAGICV2) {
2705 if (magic != MAGIC) {
2707 "Magic number of %s is invalid. You might be trying to\n",
2710 "run this program on a machine type with a different byte ordering.\n");
2714 if (timebits == 64) {
2715 read(fd, &now64, sizeof(afs_int64));
2716 now = (afs_int32) now64;
2718 read(fd, &now, sizeof(afs_int32));
2720 read(fd, &cbstuff, sizeof(cbstuff));
2721 read(fd, TimeOuts, sizeof(TimeOuts));
2722 read(fd, timeout, sizeof(timeout));
2723 read(fd, &tfirst, sizeof(tfirst));
2724 read(fd, &freelisthead, sizeof(freelisthead));
2725 CB = ((struct CallBack
2726 *)(calloc(cbstuff.nblks, sizeof(struct CallBack)))) - 1;
2727 FE = ((struct FileEntry
2728 *)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
2729 CBfree = (struct CallBack *)itocb(freelisthead);
2730 read(fd, &freelisthead, sizeof(freelisthead));
2731 FEfree = (struct FileEntry *)itofe(freelisthead);
2732 read(fd, HashTable, sizeof(HashTable));
2733 read(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2734 read(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2736 perror("Error reading dumpfile");
2743 #include "AFS_component_version_number.h"
2745 #include "AFS_component_version_number.c"
2748 static afs_uint32 *cbTrack;
2751 main(int argc, char **argv)
2753 int err = 0, cbi = 0, stats = 0, noptions = 0, all = 0, vol = 0, raw = 0;
2755 struct FileEntry *fe;
2756 struct CallBack *cb;
2760 memset(&fid, 0, sizeof(fid));
2763 while (argc && **argv == '-') {
2766 if (!strcmp(*argv, "-host")) {
2772 cbi = atoi(*++argv);
2773 } else if (!strcmp(*argv, "-fid")) {
2779 fid.Volume = atoi(*++argv);
2780 fid.Vnode = atoi(*++argv);
2781 fid.Unique = atoi(*++argv);
2782 } else if (!strcmp(*argv, "-time")) {
2783 fprintf(stderr, "-time not supported\n");
2785 } else if (!strcmp(*argv, "-stats")) {
2787 } else if (!strcmp(*argv, "-all")) {
2789 } else if (!strcmp(*argv, "-raw")) {
2791 } else if (!strcmp(*argv, "-timebits")) {
2797 timebits = atoi(*++argv);
2798 if ((timebits != 32)
2802 } else if (!strcmp(*argv, "-volume")) {
2808 vol = atoi(*++argv);
2813 if (err || argc != 1) {
2815 "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] [-timebits 32"
2817 "] callbackdumpfile\n");
2819 "[cbid is shown for each host in the hosts.dump file]\n");
2822 now = ReadDump(*argv, timebits);
2823 if (stats || noptions == 0) {
2824 time_t uxtfirst = UXtime(tfirst), tnow = now;
2825 printf("The time of the dump was %u %s", (unsigned int) now, ctime(&tnow));
2826 printf("The last time cleanup ran was %u %s", (unsigned int) uxtfirst,
2828 PrintCallBackStats();
2831 cbTrack = calloc(cbstuff.nblks, sizeof(cbTrack[0]));
2836 struct CallBack *cb;
2837 struct FileEntry *fe;
2839 for (hash = 0; hash < FEHASH_SIZE; hash++) {
2840 for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
2841 if (!vol || (fe->volid == vol)) {
2842 afs_uint32 fe_i = fetoi(fe);
2844 for (cb = itocb(fe->firstcb); cb; cb = itocb(cb->cnext)) {
2845 afs_uint32 cb_i = cbtoi(cb);
2847 if (cb_i > cbstuff.nblks) {
2848 printf("CB index out of range (%u > %d), stopped for this FE\n",
2849 cb_i, cbstuff.nblks);
2853 if (cbTrack[cb_i]) {
2854 printf("CB entry already claimed for FE[%u] (this is FE[%u]), stopped\n",
2855 cbTrack[cb_i], fe_i);
2858 cbTrack[cb_i] = fe_i;
2870 afs_uint32 cfirst = cbi;
2875 } while (cbi != cfirst);
2880 printf("No callback entries for %u.%u\n", fid.Volume, fid.Vnode);
2883 cb = itocb(fe->firstcb);
2886 cb = itocb(cb->cnext);
2891 for (i = 1; i < cbstuff.nblks; i++) {
2892 p = (afs_int32 *) & FE[i];
2893 printf("%d:%12x%12x%12x%12x\n", i, p[0], p[1], p[2], p[3]);
2902 PrintCB(struct CallBack *cb, afs_uint32 now)
2904 struct FileEntry *fe = itofe(cb->fhead);
2905 time_t expires = TIndexToTime(cb->thead);
2910 printf("vol=%u vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %lu secs at %s",
2911 fe->volid, fe->vnode, fe->ncbs, cb->hhead, cb->status, fe->status,
2912 expires - now, ctime(&expires));
2917 #if !defined(INTERPRET_DUMP)
2919 ** try breaking calbacks on afidp from host. Use multi_rx.
2920 ** return 0 on success, non-zero on failure
2923 MultiBreakCallBackAlternateAddress(struct host *host, struct AFSCBFids *afidp)
2927 retVal = MultiBreakCallBackAlternateAddress_r(host, afidp);
2933 MultiBreakCallBackAlternateAddress_r(struct host *host,
2934 struct AFSCBFids *afidp)
2937 struct rx_connection **conns;
2938 struct rx_connection *connSuccess = 0;
2939 struct AddrPort *interfaces;
2940 static struct rx_securityClass *sc = 0;
2941 static struct AFSCBs tc = { 0, 0 };
2944 /* nothing more can be done */
2945 if (!host->interface)
2946 return 1; /* failure */
2948 /* the only address is the primary interface */
2949 if (host->interface->numberOfInterfaces <= 1)
2950 return 1; /* failure */
2952 /* initialise a security object only once */
2954 sc = rxnull_NewClientSecurityObject();
2956 i = host->interface->numberOfInterfaces;
2957 interfaces = calloc(i, sizeof(struct AddrPort));
2958 conns = calloc(i, sizeof(struct rx_connection *));
2959 if (!interfaces || !conns) {
2960 ViceLogThenPanic(0, ("Failed malloc in "
2961 "MultiBreakCallBackAlternateAddress_r\n"));
2964 /* initialize alternate rx connections */
2965 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
2966 /* this is the current primary address */
2967 if (host->host == host->interface->interface[i].addr &&
2968 host->port == host->interface->interface[i].port)
2971 interfaces[j] = host->interface->interface[i];
2973 rx_NewConnection(interfaces[j].addr,
2974 interfaces[j].port, 1, sc, 0);
2975 rx_SetConnDeadTime(conns[j], 2);
2976 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
2980 osi_Assert(j); /* at least one alternate address */
2982 ("Starting multibreakcall back on all addr for host %p (%s:%d)\n",
2983 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
2985 multi_Rx(conns, j) {
2986 multi_RXAFSCB_CallBack(afidp, &tc);
2990 if (host->callback_rxcon)
2991 rx_DestroyConnection(host->callback_rxcon);
2992 host->callback_rxcon = conns[multi_i];
2993 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
2994 host->host = interfaces[multi_i].addr;
2995 host->port = interfaces[multi_i].port;
2996 h_AddHostToAddrHashTable_r(host->host, host->port, host);
2997 connSuccess = conns[multi_i];
2998 rx_SetConnDeadTime(host->callback_rxcon, 50);
2999 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3001 ("multibreakcall success with addr %s:%d\n",
3002 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3003 ntohs(interfaces[multi_i].port)));
3010 /* Destroy all connections except the one on which we succeeded */
3011 for (i = 0; i < j; i++)
3012 if (conns[i] != connSuccess)
3013 rx_DestroyConnection(conns[i]);
3019 return 0; /* success */
3021 return 1; /* failure */
3026 ** try multi_RX probes to host.
3027 ** return 0 on success, non-0 on failure
3030 MultiProbeAlternateAddress_r(struct host *host)
3033 struct rx_connection **conns;
3034 struct rx_connection *connSuccess = 0;
3035 struct AddrPort *interfaces;
3036 static struct rx_securityClass *sc = 0;
3039 /* nothing more can be done */
3040 if (!host->interface)
3041 return 1; /* failure */
3043 /* the only address is the primary interface */
3044 if (host->interface->numberOfInterfaces <= 1)
3045 return 1; /* failure */
3047 /* initialise a security object only once */
3049 sc = rxnull_NewClientSecurityObject();
3051 i = host->interface->numberOfInterfaces;
3052 interfaces = calloc(i, sizeof(struct AddrPort));
3053 conns = calloc(i, sizeof(struct rx_connection *));
3054 if (!interfaces || !conns) {
3055 ViceLogThenPanic(0, ("Failed malloc in "
3056 "MultiProbeAlternateAddress_r\n"));
3059 /* initialize alternate rx connections */
3060 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3061 /* this is the current primary address */
3062 if (host->host == host->interface->interface[i].addr &&
3063 host->port == host->interface->interface[i].port)
3066 interfaces[j] = host->interface->interface[i];
3068 rx_NewConnection(interfaces[j].addr,
3069 interfaces[j].port, 1, sc, 0);
3070 rx_SetConnDeadTime(conns[j], 2);
3071 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3075 osi_Assert(j); /* at least one alternate address */
3077 ("Starting multiprobe on all addr for host %p (%s:%d)\n",
3078 host, afs_inet_ntoa_r(host->host, hoststr),
3079 ntohs(host->port)));
3081 multi_Rx(conns, j) {
3082 multi_RXAFSCB_ProbeUuid(&host->interface->uuid);
3086 if (host->callback_rxcon)
3087 rx_DestroyConnection(host->callback_rxcon);
3088 host->callback_rxcon = conns[multi_i];
3089 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
3090 host->host = interfaces[multi_i].addr;
3091 host->port = interfaces[multi_i].port;
3092 h_AddHostToAddrHashTable_r(host->host, host->port, host);
3093 connSuccess = conns[multi_i];
3094 rx_SetConnDeadTime(host->callback_rxcon, 50);
3095 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3097 ("multiprobe success with addr %s:%d\n",
3098 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3099 ntohs(interfaces[multi_i].port)));
3104 ("multiprobe failure with addr %s:%d\n",
3105 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3106 ntohs(interfaces[multi_i].port)));
3108 /* This is less than desirable but its the best we can do.
3109 * The AFS Cache Manager will return either 0 for a Uuid
3110 * match and a 1 for a non-match. If the error is 1 we
3111 * therefore know that our mapping of IP address to Uuid
3112 * is wrong. We should attempt to find the correct
3113 * Uuid and fix the host tables.
3115 if (multi_error == 1) {
3116 /* remove the current alternate address from this host */
3118 removeInterfaceAddr_r(host, interfaces[multi_i].addr, interfaces[multi_i].port);
3122 #ifdef AFS_DEMAND_ATTACH_FS
3123 /* try to bail ASAP if the fileserver is shutting down */
3125 if (fs_state.mode == FS_MODE_SHUTDOWN) {
3134 /* Destroy all connections except the one on which we succeeded */
3135 for (i = 0; i < j; i++)
3136 if (conns[i] != connSuccess)
3137 rx_DestroyConnection(conns[i]);
3143 return 0; /* success */
3145 return 1; /* failure */
3148 #endif /* !defined(INTERPRET_DUMP) */