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
96 #include <afs/nfs.h> /* yuck. This is an abomination. */
98 #include <rx/rx_queue.h>
99 #include <afs/afscbint.h>
100 #include <afs/afsutil.h>
101 #include <afs/ihandle.h>
102 #include <afs/partition.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 "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);
186 static int MultiBreakVolumeCallBack_r(struct host *host,
187 struct VCBParams *parms, int deletefe);
188 static int MultiBreakVolumeLaterCallBack(struct host *host, void *rock);
189 static int GetSomeSpace_r(struct host *hostp, int locked);
190 static int ClearHostCallbacks_r(struct host *hp, int locked);
191 static int DumpCallBackState_r(void);
194 #define GetCB() ((struct CallBack *)iGetCB(&cbstuff.nCBs))
195 #define GetFE() ((struct FileEntry *)iGetFE(&cbstuff.nFEs))
196 #define FreeCB(cb) iFreeCB((struct CallBack *)cb, &cbstuff.nCBs)
197 #define FreeFE(fe) iFreeFE((struct FileEntry *)fe, &cbstuff.nFEs)
200 /* Other protos - move out sometime */
201 void PrintCB(struct CallBack *cb, afs_uint32 now);
203 static afs_uint32 HashTable[FEHASH_SIZE]; /* File entry hash table */
205 static struct FileEntry *
210 struct FileEntry *fe;
212 hash = FEHash(fid->Volume, fid->Unique);
213 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
215 if (fe->volid == fid->Volume && fe->unique == fid->Unique
216 && fe->vnode == fid->Vnode && (fe->status & FE_LATER) != FE_LATER)
222 #ifndef INTERPRET_DUMP
224 static struct CallBack *
227 struct CallBack *ret;
229 if ((ret = CBfree)) {
230 CBfree = (struct CallBack *)(((struct object *)ret)->next);
237 iFreeCB(struct CallBack *cb, int *nused)
239 ((struct object *)cb)->next = (struct object *)CBfree;
245 static struct FileEntry *
248 struct FileEntry *ret;
250 if ((ret = FEfree)) {
251 FEfree = (struct FileEntry *)(((struct object *)ret)->next);
258 iFreeFE(struct FileEntry *fe, int *nused)
260 ((struct object *)fe)->next = (struct object *)FEfree;
266 /* Add cb to end of specified timeout list */
268 TAdd(struct CallBack *cb, afs_uint32 * thead)
271 (*thead) = cb->tnext = cb->tprev = cbtoi(cb);
273 struct CallBack *thp = itocb(*thead);
275 cb->tprev = thp->tprev;
279 thp->tprev = (itocb(thp->tprev)->tnext = cbtoi(cb));
281 thp->tprev = cbtoi(cb);
284 cb->thead = ttoi(thead);
288 /* Delete call back entry from timeout list */
290 TDel(struct CallBack *cb)
292 afs_uint32 *thead = itot(cb->thead);
294 if (*thead == cbtoi(cb))
295 *thead = (*thead == cb->tnext ? 0 : cb->tnext);
296 if (itocb(cb->tprev))
297 itocb(cb->tprev)->tnext = cb->tnext;
298 if (itocb(cb->tnext))
299 itocb(cb->tnext)->tprev = cb->tprev;
303 /* Add cb to end of specified host list */
305 HAdd(struct CallBack *cb, struct host *host)
307 cb->hhead = h_htoi(host);
309 host->cblist = cb->hnext = cb->hprev = cbtoi(cb);
311 struct CallBack *fcb = itocb(host->cblist);
313 cb->hprev = fcb->hprev;
314 cb->hnext = cbtoi(fcb);
315 fcb->hprev = (itocb(fcb->hprev)->hnext = cbtoi(cb));
320 /* Delete call back entry from host list */
322 HDel(struct CallBack *cb)
324 afs_uint32 *hhead = &h_itoh(cb->hhead)->cblist;
326 if (*hhead == cbtoi(cb))
327 *hhead = (*hhead == cb->hnext ? 0 : cb->hnext);
328 itocb(cb->hprev)->hnext = cb->hnext;
329 itocb(cb->hnext)->hprev = cb->hprev;
333 /* Delete call back entry from fid's chain of cb's */
334 /* N.B. This one also deletes the CB, and also possibly parent FE, so
335 * make sure that it is not on any other list before calling this
338 CDel(struct CallBack *cb, int deletefe)
341 struct FileEntry *fe = itofe(cb->fhead);
345 for (safety = 0, cbp = &fe->firstcb; *cbp && *cbp != cbi;
346 cbp = &itocb(*cbp)->cnext, safety++) {
347 if (safety > cbstuff.nblks + 10) {
348 ViceLogThenPanic(0, ("CDel: Internal Error -- shutting down: "
349 "wanted %d from %d, now at %d\n",
350 cbi, fe->firstcb, *cbp));
351 DumpCallBackState_r();
352 ShutDownAndCore(PANIC);
355 CDelPtr(fe, cbp, deletefe);
359 /* Same as CDel, but pointer to parent pointer to CB entry is passed,
360 * as well as file entry */
361 /* N.B. This one also deletes the CB, and also possibly parent FE, so
362 * make sure that it is not on any other list before calling this
364 static int Ccdelpt = 0, CcdelB = 0;
367 CDelPtr(struct FileEntry *fe, afs_uint32 * cbp,
380 if ((--fe->ncbs == 0) && deletefe)
386 FindCBPtr(struct FileEntry *fe, struct host *host)
388 afs_uint32 hostindex = h_htoi(host);
393 for (safety = 0, cbp = &fe->firstcb; *cbp; cbp = &cb->cnext, safety++) {
394 if (safety > cbstuff.nblks) {
395 ViceLog(0, ("FindCBPtr: Internal Error -- shutting down.\n"));
396 DumpCallBackState_r();
397 ShutDownAndCore(PANIC);
400 if (cb->hhead == hostindex)
406 /* Delete file entry from hash table */
408 FDel(struct FileEntry *fe)
411 afs_uint32 *p = &HashTable[FEHash(fe->volid, fe->unique)];
413 while (*p && *p != fei)
414 p = &itofe(*p)->fnext;
421 /* initialize the callback package */
423 InitCallBack(int nblks)
425 opr_Assert(nblks > 0);
428 tfirst = CBtime(time(NULL));
429 /* N.B. The "-1", below, is because
430 * FE[0] and CB[0] are not used--and not allocated */
431 FE = calloc(nblks, sizeof(struct FileEntry));
433 ViceLogThenPanic(0, ("Failed malloc in InitCallBack\n"));
435 FE--; /* FE[0] is supposed to point to junk */
436 cbstuff.nFEs = nblks;
438 FreeFE(&FE[cbstuff.nFEs]); /* This is correct */
439 CB = calloc(nblks, sizeof(struct CallBack));
441 ViceLogThenPanic(0, ("Failed malloc in InitCallBack\n"));
443 CB--; /* CB[0] is supposed to point to junk */
444 cbstuff.nCBs = nblks;
446 FreeCB(&CB[cbstuff.nCBs]); /* This is correct */
447 cbstuff.nblks = nblks;
448 cbstuff.nbreakers = 0;
454 XCallBackBulk_r(struct host * ahost, struct AFSFid * fids, afs_int32 nfids)
456 struct AFSCallBack tcbs[AFSCBMAX];
462 struct rx_connection *cb_conn = NULL;
464 rx_SetConnDeadTime(ahost->callback_rxcon, 4);
465 rx_SetConnHardDeadTime(ahost->callback_rxcon, AFS_HARDDEADTIME);
471 for (i = 0; i < nfids && i < AFSCBMAX; i++) {
472 tcbs[i].CallBackVersion = CALLBACK_VERSION;
473 tcbs[i].ExpirationTime = 0;
474 tcbs[i].CallBackType = CB_DROPPED;
476 tf.AFSCBFids_len = i;
477 tf.AFSCBFids_val = &(fids[j]);
481 tc.AFSCBs_val = tcbs;
483 cb_conn = ahost->callback_rxcon;
484 rx_GetConnection(cb_conn);
486 code |= RXAFSCB_CallBack(cb_conn, &tf, &tc);
487 rx_PutConnection(cb_conn);
495 /* the locked flag tells us if the host entry has already been locked
496 * by our parent. I don't think anybody actually calls us with the
497 * host locked, but here's how to make that work: GetSomeSpace has to
498 * change so that it doesn't attempt to lock any hosts < "host". That
499 * means that it might be unable to free any objects, so it has to
500 * return an exit status. If it fails, then AddCallBack1 might fail,
501 * as well. If so, the host->ResetDone should probably be set to 0,
502 * and we probably don't want to return a callback promise to the
503 * cache manager, either. */
505 AddCallBack1(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
513 if (!(host->hostFlags & HOSTDELETED))
514 retVal = AddCallBack1_r(host, fid, thead, type, 1);
524 AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
527 struct FileEntry *fe;
528 struct CallBack *cb = 0, *lastcb = 0;
529 struct FileEntry *newfe = 0;
530 afs_uint32 time_out = 0;
531 afs_uint32 *Thead = thead;
532 struct CallBack *newcb = 0;
535 cbstuff.AddCallBacks++;
539 /* allocate these guys first, since we can't call the allocator with
540 * the host structure locked -- or we might deadlock. However, we have
541 * to avoid races with FindFE... */
542 while (!(newcb = GetCB())) {
543 GetSomeSpace_r(host, locked);
545 while (!(newfe = GetFE())) { /* Get it now, so we don't have to call */
546 /* GetSomeSpace with the host locked, later. This might turn out to */
547 /* have been unneccessary, but that's actually kind of unlikely, since */
548 /* most files are not shared. */
549 GetSomeSpace_r(host, locked);
553 h_Lock_r(host); /* this can yield, so do it before we get any */
555 if (host->hostFlags & HOSTDELETED) {
563 if (type == CB_NORMAL) {
565 TimeCeiling(time(NULL) + TimeOut(fe ? fe->ncbs : 0) +
567 Thead = THead(CBtime(time_out));
568 } else if (type == CB_VOLUME) {
569 time_out = TimeCeiling((60 * 120 + time(NULL)) + ServerBias);
570 Thead = THead(CBtime(time_out));
571 } else if (type == CB_BULK) {
572 /* bulk status can get so many callbacks all at once, and most of them
573 * are probably not for things that will be used for long.
576 TimeCeiling(time(NULL) + ServerBias +
577 TimeOut(22 + (fe ? fe->ncbs : 0)));
578 Thead = THead(CBtime(time_out));
589 fe->volid = fid->Volume;
590 fe->vnode = fid->Vnode;
591 fe->unique = fid->Unique;
594 hash = FEHash(fid->Volume, fid->Unique);
595 fe->fnext = HashTable[hash];
596 HashTable[hash] = fetoi(fe);
598 for (safety = 0, lastcb = cb = itocb(fe->firstcb); cb;
599 lastcb = cb, cb = itocb(cb->cnext), safety++) {
600 if (safety > cbstuff.nblks) {
601 ViceLog(0, ("AddCallBack1: Internal Error -- shutting down.\n"));
602 DumpCallBackState_r();
603 ShutDownAndCore(PANIC);
605 if (cb->hhead == h_htoi(host))
608 if (cb) { /* Already have call back: move to new timeout list */
609 /* don't change delayed callbacks back to normal ones */
610 if (cb->status != CB_DELAYED)
612 /* Only move if new timeout is longer */
613 if (TNorm(ttoi(Thead)) > TNorm(cb->thead)) {
617 if (newfe == NULL) { /* we are using the new FE */
618 fe->firstcb = cbtoi(cb);
620 cb->fhead = fetoi(fe);
625 *(lastcb ? &lastcb->cnext : &fe->firstcb) = cbtoi(cb);
628 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.
660 * 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)
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 opr_Assert(ncbas <= MAX_CB_HOSTS);
688 * When we issue a multi_Rx callback break, we must rx_NewCall a call for
689 * each host before we do anything. If there are no call channels
690 * available on the conn, we must wait for one of the existing calls to
691 * finish. If another thread is breaking callbacks at the same time, it is
692 * possible for us to be waiting on NewCall for one of their multi_Rx
693 * CallBack calls to finish, but they are waiting on NewCall for one of
694 * our calls to finish. So we deadlock.
696 * This can be thought of as similar to obtaining multiple locks at the
697 * same time. So if we establish an ordering, the possibility of deadlock
698 * goes away. Here we provide such an ordering, by sorting our CBAs
699 * according to CompareCBA.
701 qsort(cba, ncbas, sizeof(struct cbstruct), CompareCBA);
703 /* set up conns for multi-call */
704 for (i = 0, j = 0; i < ncbas; i++) {
705 struct host *thishost = cba[i].hp;
706 if (!thishost || (thishost->hostFlags & HOSTDELETED)) {
709 rx_GetConnection(thishost->callback_rxcon);
710 multi_to_cba_map[j] = i;
711 conns[j++] = thishost->callback_rxcon;
713 rx_SetConnDeadTime(thishost->callback_rxcon, 4);
714 rx_SetConnHardDeadTime(thishost->callback_rxcon, AFS_HARDDEADTIME);
717 if (j) { /* who knows what multi would do with 0 conns? */
721 multi_RXAFSCB_CallBack(afidp, &tc);
727 i = multi_to_cba_map[multi_i];
733 ("BCB: INTERNAL ERROR: hp=%p, cba=%p, thead=%u\n",
737 ** try breaking callbacks on alternate interface addresses
739 if (MultiBreakCallBackAlternateAddress(hp, afidp)) {
742 ("BCB: Failed on file %u.%u.%u, "
743 "Host %p (%s:%d) is down\n",
744 afidp->AFSCBFids_val->Volume,
745 afidp->AFSCBFids_val->Vnode,
746 afidp->AFSCBFids_val->Unique,
748 afs_inet_ntoa_r(hp->host, hoststr),
754 if (!(hp->hostFlags & HOSTDELETED)) {
755 hp->hostFlags |= VENUSDOWN;
757 * We always go into AddCallBack1_r with the host locked
759 AddCallBack1_r(hp, afidp->AFSCBFids_val, itot(idx),
773 for (i = 0; i < ncbas; i++) {
781 /* H_UNLOCK around this so h_FreeConnection does not deadlock.
782 h_FreeConnection should *never* be called on a callback connection,
783 but on 10/27/04 a deadlock occurred where it was, when we know why,
784 this should be reverted. -- shadow */
786 for (i = 0; i < j; i++) {
787 rx_PutConnection(conns[i]);
795 * Break all call backs for fid, except for the specified host (unless flag
796 * is true, in which case all get a callback message. Assumption: the specified
797 * host is h_Held, by the caller; the others aren't.
798 * Specified host may be bogus, that's ok. This used to check to see if the
799 * host was down in two places, once right after the host was h_held, and
800 * again after it was locked. That race condition is incredibly rare and
801 * relatively harmless even when it does occur, so we don't check for it now.
803 /* if flag is true, send a break callback msg to "host", too */
805 BreakCallBack(struct host *xhost, AFSFid * fid, int flag)
807 struct FileEntry *fe;
808 struct CallBack *cb, *nextcb;
809 struct cbstruct cba[MAX_CB_HOSTS];
817 ("BCB: BreakCallBack(Host %p all but %s:%d, (%u,%u,%u))\n",
818 xhost, afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
819 fid->Volume, fid->Vnode, fid->Unique));
822 ("BCB: BreakCallBack(No Host, (%u,%u,%u))\n",
823 fid->Volume, fid->Vnode, fid->Unique));
826 cbstuff.BreakCallBacks++;
831 hostindex = xhost ? h_htoi(xhost) : 0;
832 cb = itocb(fe->firstcb);
833 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
834 /* the most common case is what follows the || */
837 tf.AFSCBFids_len = 1;
838 tf.AFSCBFids_val = fid;
840 /* Set CBFLAG_BREAKING flag on all CBs we're looking at. We do this so we
841 * can loop through all relevant CBs while dropping H_LOCK, and not lose
842 * track of which CBs we want to look at. If we look at all CBs over and
843 * over again, we can loop indefinitely as new CBs are added. */
844 for (; cb; cb = nextcb) {
845 nextcb = itocb(cb->cnext);
847 if ((cb->hhead != hostindex || flag)
848 && (cb->status == CB_BULK || cb->status == CB_NORMAL
849 || cb->status == CB_VOLUME)) {
850 cb->flags |= CBFLAG_BREAKING;
854 cb = itocb(fe->firstcb);
857 /* loop through all CBs, only looking at ones with the CBFLAG_BREAKING
860 for (ncbas = 0; cb && ncbas < MAX_CB_HOSTS; cb = nextcb) {
861 nextcb = itocb(cb->cnext);
862 if ((cb->flags & CBFLAG_BREAKING)) {
863 struct host *thishost = h_itoh(cb->hhead);
864 cb->flags &= ~CBFLAG_BREAKING;
866 ViceLog(0, ("BCB: BOGUS! cb->hhead is NULL!\n"));
867 } else if (thishost->hostFlags & VENUSDOWN) {
869 ("BCB: %p (%s:%d) is down; delaying break call back\n",
870 thishost, afs_inet_ntoa_r(thishost->host, hoststr),
871 ntohs(thishost->port)));
872 cb->status = CB_DELAYED;
874 if (!(thishost->hostFlags & HOSTDELETED)) {
876 cba[ncbas].hp = thishost;
877 cba[ncbas].thead = cb->thead;
882 CDel(cb, 1); /* Usually first; so this delete
883 * is reasonably inexpensive */
889 MultiBreakCallBack_r(cba, ncbas, &tf);
891 /* we need to to all these initializations again because MultiBreakCallBack may block */
896 cb = itocb(fe->firstcb);
897 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
898 /* the most common case is what follows the || */
909 /* Delete (do not break) single call back for fid */
911 DeleteCallBack(struct host *host, AFSFid * fid)
913 struct FileEntry *fe;
918 cbstuff.DeleteCallBacks++;
921 /* do not care if the host has been HOSTDELETED */
927 ("DCB: No call backs for fid (%u, %u, %u)\n", fid->Volume,
928 fid->Vnode, fid->Unique));
931 pcb = FindCBPtr(fe, host);
934 ("DCB: No call back for host %p (%s:%d), (%u, %u, %u)\n",
935 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
936 fid->Volume, fid->Vnode, fid->Unique));
950 * Delete (do not break) all call backs for fid. This call doesn't
951 * set all of the various host locks, but it shouldn't really matter
952 * since we're not adding callbacks, but deleting them. I'm not sure
953 * why it doesn't set the lock, however; perhaps it should.
956 DeleteFileCallBacks(AFSFid * fid)
958 struct FileEntry *fe;
964 cbstuff.DeleteFiles++;
969 ("DF: No fid (%u,%u,%u) to delete\n", fid->Volume, fid->Vnode,
973 for (n = 0, cbi = fe->firstcb; cbi; n++) {
986 /* Delete (do not break) all call backs for host. The host should be
989 DeleteAllCallBacks_r(struct host *host, int deletefe)
994 cbstuff.DeleteAllCallBacks++;
995 cbi = first = host->cblist;
997 ViceLog(8, ("DV: no call backs\n"));
1005 } while (cbi != first);
1011 * Break all delayed call backs for host. Returns 1 if all call backs
1012 * successfully broken; 0 otherwise. Assumes host is h_Held and h_Locked.
1013 * Must be called with VenusDown set for this host
1016 BreakDelayedCallBacks(struct host *host)
1020 retVal = BreakDelayedCallBacks_r(host);
1026 BreakDelayedCallBacks_r(struct host *host)
1028 struct AFSFid fids[AFSCBMAX];
1029 int cbi, first, nfids;
1030 struct CallBack *cb;
1033 struct rx_connection *cb_conn;
1035 cbstuff.nbreakers++;
1036 if (!(host->hostFlags & RESETDONE) && !(host->hostFlags & HOSTDELETED)) {
1037 host->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1038 cb_conn = host->callback_rxcon;
1039 rx_GetConnection(cb_conn);
1040 if (host->interface) {
1043 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1046 code = RXAFSCB_InitCallBackState(cb_conn);
1048 rx_PutConnection(cb_conn);
1051 host->hostFlags |= ALTADDR; /* alternate addresses are valid */
1055 ("CB: Call back connect back failed (in break delayed) "
1056 "for Host %p (%s:%d)\n",
1057 host, afs_inet_ntoa_r(host->host, hoststr),
1058 ntohs(host->port)));
1060 host->hostFlags |= VENUSDOWN;
1063 ("InitCallBackState success on %p (%s:%d)\n",
1064 host, afs_inet_ntoa_r(host->host, hoststr),
1065 ntohs(host->port)));
1066 /* reset was done successfully */
1067 host->hostFlags |= RESETDONE;
1068 host->hostFlags &= ~VENUSDOWN;
1071 while (!(host->hostFlags & HOSTDELETED)) {
1073 host->hostFlags &= ~VENUSDOWN; /* presume up */
1074 cbi = first = host->cblist;
1078 first = host->cblist;
1081 if (cb->status == CB_DELAYED) {
1082 struct FileEntry *fe = itofe(cb->fhead);
1083 fids[nfids].Volume = fe->volid;
1084 fids[nfids].Vnode = fe->vnode;
1085 fids[nfids].Unique = fe->unique;
1091 } while (cbi && cbi != first && nfids < AFSCBMAX);
1097 if (XCallBackBulk_r(host, fids, nfids)) {
1098 /* Failed, again: put them back, probably with old
1103 ("CB: XCallBackBulk failed, Host %p (%s:%d); "
1104 "callback list follows:\n",
1105 host, afs_inet_ntoa_r(host->host, hoststr),
1106 ntohs(host->port)));
1108 for (i = 0; i < nfids; i++) {
1111 ("CB: Host %p (%s:%d), file %u.%u.%u "
1112 "(part of bulk callback)\n",
1113 host, afs_inet_ntoa_r(host->host, hoststr),
1114 ntohs(host->port), fids[i].Volume,
1115 fids[i].Vnode, fids[i].Unique));
1118 * AddCallBack1_r(host, &fids[i], itot(thead[i]), CB_DELAYED, 1);
1119 * * but it turns out to cause too many tricky locking problems.
1120 * * now, if break delayed fails, screw it. */
1122 host->hostFlags |= VENUSDOWN; /* Failed */
1123 ClearHostCallbacks_r(host, 1 /* locked */ );
1127 if (nfids < AFSCBMAX)
1131 cbstuff.nbreakers--;
1132 /* If we succeeded it's always ok to unset HFE_LATER */
1133 if (!(host->hostFlags & VENUSDOWN))
1134 host->hostFlags &= ~HFE_LATER;
1135 return (host->hostFlags & VENUSDOWN);
1139 MultiBreakVolumeCallBack_r(struct host *host,
1140 struct VCBParams *parms, int deletefe)
1144 if (host->hostFlags & HOSTDELETED)
1147 if (!(host->hostFlags & HCBREAK))
1148 return 0; /* host is not flagged to notify */
1150 if (host->hostFlags & VENUSDOWN) {
1152 /* Do not care if the host is now HOSTDELETED */
1155 ("BVCB: volume callback for Host %p (%s:%d) failed\n",
1156 host, afs_inet_ntoa_r(host->host, hoststr),
1157 ntohs(host->port)));
1159 DeleteAllCallBacks_r(host, deletefe); /* Delete all callback state
1160 * rather than attempting to
1161 * selectively remember to
1162 * delete the volume callbacks
1164 host->hostFlags &= ~(RESETDONE|HCBREAK); /* Do InitCallBackState when host returns */
1168 opr_Assert(parms->ncbas <= MAX_CB_HOSTS);
1170 /* Do not call MultiBreakCallBack on the current host structure
1171 ** because it would prematurely release the hold on the host
1173 if (parms->ncbas == MAX_CB_HOSTS) {
1174 struct AFSCBFids tf;
1176 tf.AFSCBFids_len = 1;
1177 tf.AFSCBFids_val = parms->fid;
1179 /* this releases all the hosts */
1180 MultiBreakCallBack_r(parms->cba, parms->ncbas, &tf);
1184 parms->cba[parms->ncbas].hp = host;
1185 parms->cba[(parms->ncbas)++].thead = parms->thead;
1186 host->hostFlags &= ~HCBREAK;
1188 /* we have more work to do on this host, so make sure we keep a reference
1196 MultiBreakVolumeLaterCallBack(struct host *host, void *rock)
1198 struct VCBParams *parms = (struct VCBParams *)rock;
1201 retval = MultiBreakVolumeCallBack_r(host, parms, 0);
1207 * Break all call backs on a single volume. Don't call this with any
1208 * hosts h_held. Note that this routine clears the callbacks before
1209 * actually breaking them, and that the vnode isn't locked during this
1210 * operation, so that people might see temporary callback loss while
1211 * this function is executing. It is just a temporary state, however,
1212 * since the callback will be broken later by this same function.
1214 * Now uses multi-RX for CallBack RPC in a different thread,
1215 * only marking them here.
1217 extern pthread_cond_t fsync_cond;
1220 BreakVolumeCallBacksLater(VolumeId volume)
1224 struct FileEntry *fe;
1225 struct CallBack *cb;
1229 ViceLog(25, ("Setting later on volume %" AFS_VOLID_FMT "\n",
1230 afs_printable_VolumeId_lu(volume)));
1232 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1233 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1234 if (fe->volid == volume) {
1235 struct CallBack *cbnext;
1236 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1237 host = h_itoh(cb->hhead);
1238 host->hostFlags |= HFE_LATER;
1239 cb->status = CB_DELAYED;
1240 cbnext = itocb(cb->cnext);
1243 fe->status |= FE_LATER;
1252 /* didn't find any callbacks, so return right away. */
1256 ViceLog(25, ("Fsync thread wakeup\n"));
1258 opr_cv_broadcast(&fsync_cond);
1264 BreakLaterCallBacks(void)
1269 struct CallBack *cb;
1270 struct FileEntry *fe = NULL;
1271 struct FileEntry *myfe = NULL;
1273 struct VCBParams henumParms;
1274 unsigned short tthead = 0; /* zero is illegal value */
1278 ViceLog(25, ("Looking for FileEntries to unchain\n"));
1281 /* Pick the first volume we see to clean up */
1282 fid.Volume = fid.Vnode = fid.Unique = 0;
1284 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1285 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1286 if (fe && (fe->status & FE_LATER)
1287 && (fid.Volume == 0 || fid.Volume == fe->volid)) {
1288 /* Ugly, but used to avoid left side casting */
1289 struct object *tmpfe;
1291 ("Unchaining for %u:%u:%" AFS_VOLID_FMT "\n", fe->vnode,
1292 fe->unique, afs_printable_VolumeId_lu(fe->volid)));
1293 fid.Volume = fe->volid;
1295 fe->status &= ~FE_LATER; /* not strictly needed */
1296 /* Works since volid is deeper than the largest pointer */
1297 tmpfe = (struct object *)fe;
1298 tmpfe->next = (struct object *)myfe;
1311 /* loop over FEs from myfe and free/break */
1313 for (fe = myfe; fe;) {
1314 struct CallBack *cbnext;
1315 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1316 cbnext = itocb(cb->cnext);
1317 host = h_itoh(cb->hhead);
1318 if (cb->status == CB_DELAYED) {
1319 if (!(host->hostFlags & HOSTDELETED)) {
1320 /* mark this host for notification */
1321 host->hostFlags |= HCBREAK;
1322 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1328 CDel(cb, 0); /* Don't let CDel clean up the fe */
1329 /* leave flag for MultiBreakVolumeCallBack to clear */
1332 ("Found host %p (%s:%d) non-DELAYED cb for %u:%u:%" AFS_VOLID_FMT "\n",
1333 host, afs_inet_ntoa_r(host->host, hoststr),
1334 ntohs(host->port), fe->vnode, fe->unique,
1335 afs_printable_VolumeId_lu(fe->volid)));
1339 fe = (struct FileEntry *)((struct object *)fe)->next;
1344 ViceLog(125, ("Breaking volume %u\n", fid.Volume));
1345 henumParms.ncbas = 0;
1346 henumParms.fid = &fid;
1347 henumParms.thead = tthead;
1349 h_Enumerate(MultiBreakVolumeLaterCallBack, (char *)&henumParms);
1351 if (henumParms.ncbas) { /* do left-overs */
1352 struct AFSCBFids tf;
1353 tf.AFSCBFids_len = 1;
1354 tf.AFSCBFids_val = &fid;
1356 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf);
1357 henumParms.ncbas = 0;
1362 /* Arrange to be called again */
1367 * Delete all timed-out call back entries (to be called periodically by file
1371 CleanupTimedOutCallBacks(void)
1374 CleanupTimedOutCallBacks_r();
1380 CleanupTimedOutCallBacks_r(void)
1382 afs_uint32 now = CBtime(time(NULL));
1384 struct CallBack *cb;
1388 while (tfirst <= now) {
1390 cbi = *(thead = THead(tfirst));
1396 ("CCB: deleting timed out call back %x (%s:%d), (%" AFS_VOLID_FMT ",%u,%u)\n",
1397 h_itoh(cb->hhead)->host,
1398 afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
1399 h_itoh(cb->hhead)->port,
1400 afs_printable_VolumeId_lu(itofe(cb->fhead)->volid),
1401 itofe(cb->fhead)->vnode, itofe(cb->fhead)->unique));
1405 if (ntimedout > cbstuff.nblks) {
1406 ViceLog(0, ("CCB: Internal Error -- shutting down...\n"));
1407 DumpCallBackState_r();
1408 ShutDownAndCore(PANIC);
1410 } while (cbi != *thead);
1415 cbstuff.CBsTimedOut += ntimedout;
1416 ViceLog(7, ("CCB: deleted %d timed out callbacks\n", ntimedout));
1417 return (ntimedout > 0);
1421 * parameters to pass to lih*_r from h_Enumerate_r when trying to find a host
1422 * from which to clear callbacks.
1426 * Points to the least interesting host found; try to clear callbacks on
1427 * this host after h_Enumerate_r(lih*_r)'ing.
1432 * The last host we got from lih*_r, but we couldn't clear its callbacks
1433 * for some reason. Choose the next-best host after this one (with the
1434 * current lih*_r, this means to only select hosts that have an ActiveCall
1435 * newer than lastlih).
1437 struct host *lastlih;
1440 /* Value of host->refCount that allows us to reliably infer that
1441 * host may be held by some other thread */
1442 #define OTHER_MUSTHOLD_LIH 2
1444 /* This version does not allow 'host' to be selected unless its ActiveCall
1445 * is newer than 'params->lastlih' which is the host with the oldest
1446 * ActiveCall from the last pass (if it is provided). We filter out any hosts
1447 * that are are held by other threads.
1449 * There is a small problem here, but it may not be easily fixable. Say we
1450 * select some host A, and give it back to GetSomeSpace_r. GSS_r for some
1451 * reason cannot clear the callbacks on A, and so calls us again with
1452 * lastlih = A. Suppose there is another host B that has the same ActiveCall
1453 * time as A. We will now skip over host B, since
1454 * 'hostB->ActiveCall > hostA->ActiveCall' is not true. This could result in
1455 * us prematurely going to the GSS_r 2nd or 3rd pass, and making us a little
1456 * inefficient. This should be pretty rare, though, except perhaps in cases
1457 * with very small numbers of hosts.
1459 * Also filter out any hosts with HOSTDELETED set. h_Enumerate_r should in
1460 * theory not give these to us anyway, but be paranoid.
1463 lih0_r(struct host *host, void *rock)
1465 struct lih_params *params = (struct lih_params *)rock;
1467 /* OTHER_MUSTHOLD_LIH is because the h_Enum loop holds us once */
1469 && (!(host->hostFlags & HOSTDELETED))
1470 && (host->refCount < OTHER_MUSTHOLD_LIH)
1471 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1472 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1475 h_Release_r(params->lih); /* release prev host */
1484 /* same as lih0_r, except we do not prevent held hosts from being selected. */
1486 lih1_r(struct host *host, void *rock)
1488 struct lih_params *params = (struct lih_params *)rock;
1491 && (!(host->hostFlags & HOSTDELETED))
1492 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1493 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1496 h_Release_r(params->lih); /* release prev host */
1505 /* This could be upgraded to get more space each time */
1506 /* first pass: sequentially find the oldest host which isn't held by
1507 anyone for which we can clear callbacks;
1509 /* second pass: sequentially find the oldest host regardless of
1510 whether or not the host is held; skipping 'hostp' */
1511 /* third pass: attempt to clear callbacks from 'hostp' */
1512 /* always called with hostp unlocked */
1514 /* Note: hostlist is ordered most recently created host first and
1515 * its order has no relationship to the most recently used. */
1516 extern struct host *hostList;
1518 GetSomeSpace_r(struct host *hostp, int locked)
1521 struct lih_params params;
1524 if (cbstuff.GotSomeSpaces == 0) {
1525 /* only log this once; if GSS is getting called constantly, that's not
1526 * good but don't make things worse by spamming the log. */
1527 ViceLog(0, ("We have run out of callback space; forcing callback revocation. "
1528 "This suggests the fileserver is configured with insufficient "
1529 "callbacks; you probably want to increase the -cb fileserver "
1530 "parameter (current setting: %u). The fileserver will continue "
1531 "to operate, but this may indicate a severe performance problem\n",
1533 ViceLog(0, ("This message is logged at most once; for more information "
1534 "see the OpenAFS documentation and fileserver xstat collection 3\n"));
1537 cbstuff.GotSomeSpaces++;
1539 ("GSS: First looking for timed out call backs via CleanupCallBacks\n"));
1540 if (CleanupTimedOutCallBacks_r()) {
1546 params.lastlih = NULL;
1551 h_Enumerate_r(i == 0 ? lih0_r : lih1_r, hostList, ¶ms);
1554 if (params.lastlih) {
1555 h_Release_r(params.lastlih);
1556 params.lastlih = NULL;
1560 /* note that 'hp' was held by lih*_r; we will need to release it */
1562 if ((hp != hostp) && !ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
1567 params.lastlih = hp;
1568 /* params.lastlih will be released on the next iteration, after
1573 * Next time try getting callbacks from any host even if
1574 * it's held, since the only other option is starvation for
1575 * the file server (i.e. until the callback timeout arrives).
1578 params.lastlih = NULL;
1581 ("GSS: Try harder for longest inactive host cnt= %d\n",
1586 /* Could not obtain space from other hosts, clear hostp's callback state */
1591 ClearHostCallbacks_r(hostp, 1 /*already locked */ );
1598 /* locked - set if caller has already locked the host */
1600 ClearHostCallbacks_r(struct host *hp, int locked)
1604 struct rx_connection *cb_conn = NULL;
1607 ("GSS: Delete longest inactive host %p (%s:%d)\n",
1608 hp, afs_inet_ntoa_r(hp->host, hoststr), ntohs(hp->port)));
1610 if ((hp->hostFlags & HOSTDELETED)) {
1611 /* hp could go away after reacquiring H_LOCK in h_NBLock_r, so we can't
1612 * really use it; its callbacks will get cleared anyway when
1613 * h_TossStuff_r gets its hands on it */
1619 /** Try a non-blocking lock. If the lock is already held return
1620 * after releasing hold on hp
1623 if (h_NBLock_r(hp)) {
1628 if (hp->Console & 2) {
1630 * If the special console field is set it means that a thread
1631 * is waiting in AddCallBack1 after it set pointers to the
1632 * file entry and/or callback entry. Because of the bogus
1633 * usage of h_hold it won't prevent from another thread, this
1634 * one, to remove all the callbacks so just to be safe we keep
1635 * a reference. NOTE, on the last phase we'll free the calling
1636 * host's callbacks but that's ok...
1640 DeleteAllCallBacks_r(hp, 1);
1641 if (hp->hostFlags & VENUSDOWN) {
1642 hp->hostFlags &= ~RESETDONE; /* remember that we must do a reset */
1643 } else if (!(hp->hostFlags & HOSTDELETED)) {
1644 /* host is up, try a call */
1645 hp->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1646 cb_conn = hp->callback_rxcon;
1647 rx_GetConnection(hp->callback_rxcon);
1648 if (hp->interface) {
1651 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1654 code = RXAFSCB_InitCallBackState(cb_conn);
1656 rx_PutConnection(cb_conn);
1659 hp->hostFlags |= ALTADDR; /* alternate addresses are valid */
1661 /* failed, mark host down and need reset */
1662 hp->hostFlags |= VENUSDOWN;
1663 hp->hostFlags &= ~RESETDONE;
1665 /* reset succeeded, we're done */
1666 hp->hostFlags |= RESETDONE;
1675 #endif /* INTERPRET_DUMP */
1679 PrintCallBackStats(void)
1682 "%d add CB, %d break CB, %d del CB, %d del FE, %d CB's timed out, %d space reclaim, %d del host\n",
1683 cbstuff.AddCallBacks, cbstuff.BreakCallBacks,
1684 cbstuff.DeleteCallBacks, cbstuff.DeleteFiles, cbstuff.CBsTimedOut,
1685 cbstuff.GotSomeSpaces, cbstuff.DeleteAllCallBacks);
1686 fprintf(stderr, "%d CBs, %d FEs, (%d of total of %d 16-byte blocks)\n",
1687 cbstuff.nCBs, cbstuff.nFEs, cbstuff.nCBs + cbstuff.nFEs,
1689 fprintf(stderr, "%d GSS1, %d GSS2, %d GSS3, %d GSS4, %d GSS5 (internal counters)\n",
1690 cbstuff.GSS1, cbstuff.GSS2, cbstuff.GSS3, cbstuff.GSS4, cbstuff.GSS5);
1695 #define MAGIC 0x12345678 /* To check byte ordering of dump when it is read in */
1696 #define MAGICV2 0x12345679 /* To check byte ordering & version of dump when it is read in */
1699 #ifndef INTERPRET_DUMP
1701 #ifdef AFS_DEMAND_ATTACH_FS
1704 * callback state serialization
1706 static int cb_stateSaveTimeouts(struct fs_dump_state * state);
1707 static int cb_stateSaveFEHash(struct fs_dump_state * state);
1708 static int cb_stateSaveFEs(struct fs_dump_state * state);
1709 static int cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe);
1710 static int cb_stateRestoreTimeouts(struct fs_dump_state * state);
1711 static int cb_stateRestoreFEHash(struct fs_dump_state * state);
1712 static int cb_stateRestoreFEs(struct fs_dump_state * state);
1713 static int cb_stateRestoreFE(struct fs_dump_state * state);
1714 static int cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
1715 struct iovec * iov, int niovecs);
1717 static int cb_stateVerifyFEHash(struct fs_dump_state * state);
1718 static int cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe);
1719 static int cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe);
1720 static int cb_stateVerifyTimeoutQueues(struct fs_dump_state * state);
1722 static int cb_stateFEToDiskEntry(struct FileEntry *, struct FEDiskEntry *);
1723 static int cb_stateDiskEntryToFE(struct fs_dump_state * state,
1724 struct FEDiskEntry *, struct FileEntry *);
1726 static int cb_stateCBToDiskEntry(struct CallBack *, struct CBDiskEntry *);
1727 static int cb_stateDiskEntryToCB(struct fs_dump_state * state,
1728 struct CBDiskEntry *, struct CallBack *);
1730 static int cb_stateFillHeader(struct callback_state_header * hdr);
1731 static int cb_stateCheckHeader(struct callback_state_header * hdr);
1733 static int cb_stateAllocMap(struct fs_dump_state * state);
1736 cb_stateSave(struct fs_dump_state * state)
1740 AssignInt64(state->eof_offset, &state->hdr->cb_offset);
1742 /* invalidate callback state header */
1743 memset(state->cb_hdr, 0, sizeof(struct callback_state_header));
1744 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1745 sizeof(struct callback_state_header))) {
1750 fs_stateIncEOF(state, sizeof(struct callback_state_header));
1752 /* dump timeout state */
1753 if (cb_stateSaveTimeouts(state)) {
1758 /* dump fe hashtable state */
1759 if (cb_stateSaveFEHash(state)) {
1764 /* dump callback state */
1765 if (cb_stateSaveFEs(state)) {
1770 /* write the callback state header to disk */
1771 cb_stateFillHeader(state->cb_hdr);
1772 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1773 sizeof(struct callback_state_header))) {
1783 cb_stateRestore(struct fs_dump_state * state)
1787 if (fs_stateReadHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1788 sizeof(struct callback_state_header))) {
1793 if (cb_stateCheckHeader(state->cb_hdr)) {
1798 if (cb_stateAllocMap(state)) {
1803 if (cb_stateRestoreTimeouts(state)) {
1808 if (cb_stateRestoreFEHash(state)) {
1813 /* restore FEs and CBs from disk */
1814 if (cb_stateRestoreFEs(state)) {
1819 /* restore the timeout queue heads */
1820 tfirst = state->cb_hdr->tfirst;
1827 cb_stateRestoreIndices(struct fs_dump_state * state)
1830 struct FileEntry * fe;
1831 struct CallBack * cb;
1833 /* restore indices in the FileEntry structures */
1834 for (i = 1; i < state->fe_map.len; i++) {
1835 if (state->fe_map.entries[i].new_idx) {
1836 fe = itofe(state->fe_map.entries[i].new_idx);
1838 /* restore the fe->fnext entry */
1839 if (fe_OldToNew(state, fe->fnext, &fe->fnext)) {
1844 /* restore the fe->firstcb entry */
1845 if (cb_OldToNew(state, fe->firstcb, &fe->firstcb)) {
1852 /* restore indices in the CallBack structures */
1853 for (i = 1; i < state->cb_map.len; i++) {
1854 if (state->cb_map.entries[i].new_idx) {
1855 cb = itocb(state->cb_map.entries[i].new_idx);
1857 /* restore the cb->cnext entry */
1858 if (cb_OldToNew(state, cb->cnext, &cb->cnext)) {
1863 /* restore the cb->fhead entry */
1864 if (fe_OldToNew(state, cb->fhead, &cb->fhead)) {
1869 /* restore the cb->hhead entry */
1870 if (h_OldToNew(state, cb->hhead, &cb->hhead)) {
1875 /* restore the cb->tprev entry */
1876 if (cb_OldToNew(state, cb->tprev, &cb->tprev)) {
1881 /* restore the cb->tnext entry */
1882 if (cb_OldToNew(state, cb->tnext, &cb->tnext)) {
1887 /* restore the cb->hprev entry */
1888 if (cb_OldToNew(state, cb->hprev, &cb->hprev)) {
1893 /* restore the cb->hnext entry */
1894 if (cb_OldToNew(state, cb->hnext, &cb->hnext)) {
1901 /* restore the timeout queue head indices */
1902 for (i = 0; i < state->cb_timeout_hdr->records; i++) {
1903 if (cb_OldToNew(state, timeout[i], &timeout[i])) {
1909 /* restore the FE hash table queue heads */
1910 for (i = 0; i < state->cb_fehash_hdr->records; i++) {
1911 if (fe_OldToNew(state, HashTable[i], &HashTable[i])) {
1922 cb_stateVerify(struct fs_dump_state * state)
1926 if (cb_stateVerifyFEHash(state)) {
1930 if (cb_stateVerifyTimeoutQueues(state)) {
1938 cb_stateVerifyFEHash(struct fs_dump_state * state)
1941 struct FileEntry * fe;
1942 afs_uint32 fei, chain_len;
1944 for (i = 0; i < FEHASH_SIZE; i++) {
1946 for (fei = HashTable[i], fe = itofe(fei);
1948 fei = fe->fnext, fe = itofe(fei)) {
1949 if (fei > cbstuff.nblks) {
1950 ViceLog(0, ("cb_stateVerifyFEHash: error: index out of range (fei=%d)\n", fei));
1954 if (cb_stateVerifyFE(state, fe)) {
1957 if (chain_len > FS_STATE_FE_MAX_HASH_CHAIN_LEN) {
1958 ViceLog(0, ("cb_stateVerifyFEHash: error: hash chain %d length exceeds %d; assuming there's a loop\n",
1959 i, FS_STATE_FE_MAX_HASH_CHAIN_LEN));
1971 cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe)
1975 if ((fe->firstcb && !fe->ncbs) ||
1976 (!fe->firstcb && fe->ncbs)) {
1977 ViceLog(0, ("cb_stateVerifyFE: error: fe->firstcb does not agree with fe->ncbs (fei=%lu, fe->firstcb=%lu, fe->ncbs=%lu)\n",
1978 afs_printable_uint32_lu(fetoi(fe)),
1979 afs_printable_uint32_lu(fe->firstcb),
1980 afs_printable_uint32_lu(fe->ncbs)));
1983 if (cb_stateVerifyFCBList(state, fe)) {
1984 ViceLog(0, ("cb_stateVerifyFE: error: FCBList failed verification (fei=%lu)\n",
1985 afs_printable_uint32_lu(fetoi(fe))));
1993 cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe)
1996 afs_uint32 cbi, fei, chain_len = 0;
1997 struct CallBack * cb;
2001 for (cbi = fe->firstcb, cb = itocb(cbi);
2003 cbi = cb->cnext, cb = itocb(cbi)) {
2004 if (cbi > cbstuff.nblks) {
2005 ViceLog(0, ("cb_stateVerifyFCBList: error: list index out of range (cbi=%d, ncbs=%d)\n",
2006 cbi, cbstuff.nblks));
2010 if (cb->fhead != fei) {
2011 ViceLog(0, ("cb_stateVerifyFCBList: error: cb->fhead != fei (fei=%d, cb->fhead=%d)\n",
2015 if (chain_len > FS_STATE_FCB_MAX_LIST_LEN) {
2016 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (fei=%d); assuming there's a loop\n",
2017 FS_STATE_FCB_MAX_LIST_LEN, fei));
2024 if (fe->ncbs != chain_len) {
2025 ViceLog(0, ("cb_stateVerifyFCBList: error: list length mismatch (len=%d, fe->ncbs=%d)\n",
2026 chain_len, fe->ncbs));
2035 cb_stateVerifyHCBList(struct fs_dump_state * state, struct host * host)
2038 afs_uint32 hi, chain_len, cbi;
2039 struct CallBack *cb, *ncb;
2044 for (cbi = host->cblist, cb = itocb(cbi);
2046 cbi = cb->hnext, cb = ncb) {
2047 if (chain_len && (host->cblist == cbi)) {
2048 /* we've wrapped around the circular list, and everything looks ok */
2051 if (cb->hhead != hi) {
2052 ViceLog(0, ("cb_stateVerifyHCBList: error: incorrect cb->hhead (cbi=%d, h->index=%d, cb->hhead=%d)\n",
2053 cbi, hi, cb->hhead));
2056 if (!cb->hprev || !cb->hnext) {
2057 ViceLog(0, ("cb_stateVerifyHCBList: error: null index in circular list (cbi=%d, h->index=%d)\n",
2062 if ((cb->hprev > cbstuff.nblks) ||
2063 (cb->hnext > cbstuff.nblks)) {
2064 ViceLog(0, ("cb_stateVerifyHCBList: error: list index out of range (cbi=%d, h->index=%d, cb->hprev=%d, cb->hnext=%d, nCBs=%d)\n",
2065 cbi, hi, cb->hprev, cb->hnext, cbstuff.nblks));
2069 ncb = itocb(cb->hnext);
2070 if (cbi != ncb->hprev) {
2071 ViceLog(0, ("cb_stateVerifyHCBList: error: corrupt linked list (cbi=%d, h->index=%d)\n",
2076 if (chain_len > FS_STATE_HCB_MAX_LIST_LEN) {
2077 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (h->index=%d); assuming there's a loop\n",
2078 FS_STATE_HCB_MAX_LIST_LEN, hi));
2090 cb_stateVerifyTimeoutQueues(struct fs_dump_state * state)
2093 afs_uint32 cbi, chain_len;
2094 struct CallBack *cb, *ncb;
2096 for (i = 0; i < CB_NUM_TIMEOUT_QUEUES; i++) {
2098 for (cbi = timeout[i], cb = itocb(cbi);
2100 cbi = cb->tnext, cb = ncb) {
2101 if (chain_len && (cbi == timeout[i])) {
2102 /* we've wrapped around the circular list, and everything looks ok */
2105 if (cbi > cbstuff.nblks) {
2106 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: list index out of range (cbi=%d, tindex=%d)\n",
2111 if (itot(cb->thead) != &timeout[i]) {
2112 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: cb->thead points to wrong timeout queue (tindex=%d, cbi=%d, cb->thead=%d)\n",
2113 i, cbi, cb->thead));
2116 if (!cb->tprev || !cb->tnext) {
2117 ViceLog(0, ("cb_stateVerifyTimeoutQueues: null index in circular list (cbi=%d, tindex=%d)\n",
2122 if ((cb->tprev > cbstuff.nblks) ||
2123 (cb->tnext > cbstuff.nblks)) {
2124 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list index out of range (cbi=%d, tindex=%d, cb->tprev=%d, cb->tnext=%d, nCBs=%d)\n",
2125 cbi, i, cb->tprev, cb->tnext, cbstuff.nblks));
2129 ncb = itocb(cb->tnext);
2130 if (cbi != ncb->tprev) {
2131 ViceLog(0, ("cb_stateVerifyTimeoutQueues: corrupt linked list (cbi=%d, tindex=%d)\n",
2136 if (chain_len > FS_STATE_TCB_MAX_LIST_LEN) {
2137 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list length exceeds %d (tindex=%d); assuming there's a loop\n",
2138 FS_STATE_TCB_MAX_LIST_LEN, i));
2150 cb_stateSaveTimeouts(struct fs_dump_state * state)
2153 struct iovec iov[2];
2155 AssignInt64(state->eof_offset, &state->cb_hdr->timeout_offset);
2157 memset(state->cb_timeout_hdr, 0, sizeof(struct callback_state_fehash_header));
2158 state->cb_timeout_hdr->magic = CALLBACK_STATE_TIMEOUT_MAGIC;
2159 state->cb_timeout_hdr->records = CB_NUM_TIMEOUT_QUEUES;
2160 state->cb_timeout_hdr->len = sizeof(struct callback_state_timeout_header) +
2161 (state->cb_timeout_hdr->records * sizeof(afs_uint32));
2163 iov[0].iov_base = (char *)state->cb_timeout_hdr;
2164 iov[0].iov_len = sizeof(struct callback_state_timeout_header);
2165 iov[1].iov_base = (char *)timeout;
2166 iov[1].iov_len = sizeof(timeout);
2168 if (fs_stateSeek(state, &state->cb_hdr->timeout_offset)) {
2173 if (fs_stateWriteV(state, iov, 2)) {
2178 fs_stateIncEOF(state, state->cb_timeout_hdr->len);
2185 cb_stateRestoreTimeouts(struct fs_dump_state * state)
2189 if (fs_stateReadHeader(state, &state->cb_hdr->timeout_offset,
2190 state->cb_timeout_hdr,
2191 sizeof(struct callback_state_timeout_header))) {
2196 if (state->cb_timeout_hdr->magic != CALLBACK_STATE_TIMEOUT_MAGIC) {
2200 if (state->cb_timeout_hdr->records != CB_NUM_TIMEOUT_QUEUES) {
2205 len = state->cb_timeout_hdr->records * sizeof(afs_uint32);
2207 if (state->cb_timeout_hdr->len !=
2208 (sizeof(struct callback_state_timeout_header) + len)) {
2213 if (fs_stateRead(state, timeout, len)) {
2223 cb_stateSaveFEHash(struct fs_dump_state * state)
2226 struct iovec iov[2];
2228 AssignInt64(state->eof_offset, &state->cb_hdr->fehash_offset);
2230 memset(state->cb_fehash_hdr, 0, sizeof(struct callback_state_fehash_header));
2231 state->cb_fehash_hdr->magic = CALLBACK_STATE_FEHASH_MAGIC;
2232 state->cb_fehash_hdr->records = FEHASH_SIZE;
2233 state->cb_fehash_hdr->len = sizeof(struct callback_state_fehash_header) +
2234 (state->cb_fehash_hdr->records * sizeof(afs_uint32));
2236 iov[0].iov_base = (char *)state->cb_fehash_hdr;
2237 iov[0].iov_len = sizeof(struct callback_state_fehash_header);
2238 iov[1].iov_base = (char *)HashTable;
2239 iov[1].iov_len = sizeof(HashTable);
2241 if (fs_stateSeek(state, &state->cb_hdr->fehash_offset)) {
2246 if (fs_stateWriteV(state, iov, 2)) {
2251 fs_stateIncEOF(state, state->cb_fehash_hdr->len);
2258 cb_stateRestoreFEHash(struct fs_dump_state * state)
2262 if (fs_stateReadHeader(state, &state->cb_hdr->fehash_offset,
2263 state->cb_fehash_hdr,
2264 sizeof(struct callback_state_fehash_header))) {
2269 if (state->cb_fehash_hdr->magic != CALLBACK_STATE_FEHASH_MAGIC) {
2273 if (state->cb_fehash_hdr->records != FEHASH_SIZE) {
2278 len = state->cb_fehash_hdr->records * sizeof(afs_uint32);
2280 if (state->cb_fehash_hdr->len !=
2281 (sizeof(struct callback_state_fehash_header) + len)) {
2286 if (fs_stateRead(state, HashTable, len)) {
2296 cb_stateSaveFEs(struct fs_dump_state * state)
2300 struct FileEntry *fe;
2302 AssignInt64(state->eof_offset, &state->cb_hdr->fe_offset);
2304 for (hash = 0; hash < FEHASH_SIZE ; hash++) {
2305 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
2307 if (cb_stateSaveFE(state, fe)) {
2319 cb_stateRestoreFEs(struct fs_dump_state * state)
2321 int count, nFEs, ret = 0;
2323 nFEs = state->cb_hdr->nFEs;
2325 for (count = 0; count < nFEs; count++) {
2326 if (cb_stateRestoreFE(state)) {
2337 cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe)
2339 int ret = 0, iovcnt, cbi, written = 0;
2341 struct callback_state_entry_header hdr;
2342 struct FEDiskEntry fedsk;
2343 struct CBDiskEntry cbdsk[16];
2344 struct iovec iov[16];
2345 struct CallBack *cb;
2348 if (fei > state->cb_hdr->fe_max) {
2349 state->cb_hdr->fe_max = fei;
2352 memset(&hdr, 0, sizeof(struct callback_state_entry_header));
2354 if (cb_stateFEToDiskEntry(fe, &fedsk)) {
2359 iov[0].iov_base = (char *)&hdr;
2360 iov[0].iov_len = sizeof(hdr);
2361 iov[1].iov_base = (char *)&fedsk;
2362 iov[1].iov_len = sizeof(struct FEDiskEntry);
2365 for (cbi = fe->firstcb, cb = itocb(cbi);
2367 cbi = cb->cnext, cb = itocb(cbi), hdr.nCBs++) {
2368 if (cbi > state->cb_hdr->cb_max) {
2369 state->cb_hdr->cb_max = cbi;
2371 if (cb_stateCBToDiskEntry(cb, &cbdsk[iovcnt])) {
2375 cbdsk[iovcnt].index = cbi;
2376 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2377 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2379 if ((iovcnt == 16) || (!cb->cnext)) {
2380 if (fs_stateWriteV(state, iov, iovcnt)) {
2389 hdr.magic = CALLBACK_STATE_ENTRY_MAGIC;
2390 hdr.len = sizeof(hdr) + sizeof(struct FEDiskEntry) +
2391 (hdr.nCBs * sizeof(struct CBDiskEntry));
2394 if (fs_stateWriteV(state, iov, iovcnt)) {
2399 if (fs_stateWriteHeader(state, &state->eof_offset, &hdr, sizeof(hdr))) {
2405 fs_stateIncEOF(state, hdr.len);
2408 if (fs_stateSeek(state, &state->eof_offset)) {
2414 state->cb_hdr->nFEs++;
2415 state->cb_hdr->nCBs += hdr.nCBs;
2422 cb_stateRestoreFE(struct fs_dump_state * state)
2424 int ret = 0, iovcnt, nCBs;
2425 struct callback_state_entry_header hdr;
2426 struct FEDiskEntry fedsk;
2427 struct CBDiskEntry cbdsk[16];
2428 struct iovec iov[16];
2429 struct FileEntry * fe;
2431 iov[0].iov_base = (char *)&hdr;
2432 iov[0].iov_len = sizeof(hdr);
2433 iov[1].iov_base = (char *)&fedsk;
2434 iov[1].iov_len = sizeof(fedsk);
2437 if (fs_stateReadV(state, iov, iovcnt)) {
2442 if (hdr.magic != CALLBACK_STATE_ENTRY_MAGIC) {
2449 ViceLog(0, ("cb_stateRestoreFE: ran out of free FileEntry structures\n"));
2454 if (cb_stateDiskEntryToFE(state, &fedsk, fe)) {
2460 for (iovcnt = 0, nCBs = 0;
2463 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2464 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2466 if ((iovcnt == 16) || (nCBs == hdr.nCBs - 1)) {
2467 if (fs_stateReadV(state, iov, iovcnt)) {
2471 if (cb_stateRestoreCBs(state, fe, iov, iovcnt)) {
2485 cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
2486 struct iovec * iov, int niovecs)
2489 struct CallBack * cb;
2490 struct CBDiskEntry * cbdsk;
2492 for (idx = 0; idx < niovecs; idx++) {
2493 cbdsk = (struct CBDiskEntry *) iov[idx].iov_base;
2495 if (cbdsk->cb.hhead < state->h_map.len &&
2496 state->h_map.entries[cbdsk->cb.hhead].valid == FS_STATE_IDX_SKIPPED) {
2500 if ((cb = GetCB()) == NULL) {
2501 ViceLog(0, ("cb_stateRestoreCBs: ran out of free CallBack structures\n"));
2505 if (cb_stateDiskEntryToCB(state, cbdsk, cb)) {
2506 ViceLog(0, ("cb_stateRestoreCBs: corrupt CallBack disk entry\n"));
2518 cb_stateFillHeader(struct callback_state_header * hdr)
2520 hdr->stamp.magic = CALLBACK_STATE_MAGIC;
2521 hdr->stamp.version = CALLBACK_STATE_VERSION;
2522 hdr->tfirst = tfirst;
2527 cb_stateCheckHeader(struct callback_state_header * hdr)
2531 if (hdr->stamp.magic != CALLBACK_STATE_MAGIC) {
2533 } else if (hdr->stamp.version != CALLBACK_STATE_VERSION) {
2535 } else if ((hdr->nFEs > cbstuff.nblks) || (hdr->nCBs > cbstuff.nblks)) {
2536 ViceLog(0, ("cb_stateCheckHeader: saved callback state larger than callback memory allocation\n"));
2542 /* disk entry conversion routines */
2544 cb_stateFEToDiskEntry(struct FileEntry * in, struct FEDiskEntry * out)
2546 memcpy(&out->fe, in, sizeof(struct FileEntry));
2547 out->index = fetoi(in);
2552 cb_stateDiskEntryToFE(struct fs_dump_state * state,
2553 struct FEDiskEntry * in, struct FileEntry * out)
2557 memcpy(out, &in->fe, sizeof(struct FileEntry));
2559 /* setup FE map entry */
2560 if (!in->index || (in->index >= state->fe_map.len)) {
2561 ViceLog(0, ("cb_stateDiskEntryToFE: index (%d) out of range",
2566 state->fe_map.entries[in->index].valid = FS_STATE_IDX_VALID;
2567 state->fe_map.entries[in->index].old_idx = in->index;
2568 state->fe_map.entries[in->index].new_idx = fetoi(out);
2575 cb_stateCBToDiskEntry(struct CallBack * in, struct CBDiskEntry * out)
2577 memcpy(&out->cb, in, sizeof(struct CallBack));
2578 out->index = cbtoi(in);
2583 cb_stateDiskEntryToCB(struct fs_dump_state * state,
2584 struct CBDiskEntry * in, struct CallBack * out)
2588 memcpy(out, &in->cb, sizeof(struct CallBack));
2590 /* setup CB map entry */
2591 if (!in->index || (in->index >= state->cb_map.len)) {
2592 ViceLog(0, ("cb_stateDiskEntryToCB: index (%d) out of range\n",
2597 state->cb_map.entries[in->index].valid = FS_STATE_IDX_VALID;
2598 state->cb_map.entries[in->index].old_idx = in->index;
2599 state->cb_map.entries[in->index].new_idx = cbtoi(out);
2605 /* index map routines */
2607 cb_stateAllocMap(struct fs_dump_state * state)
2609 state->fe_map.len = state->cb_hdr->fe_max + 1;
2610 state->cb_map.len = state->cb_hdr->cb_max + 1;
2611 state->fe_map.entries = (struct idx_map_entry_t *)
2612 calloc(state->fe_map.len, sizeof(struct idx_map_entry_t));
2613 state->cb_map.entries = (struct idx_map_entry_t *)
2614 calloc(state->cb_map.len, sizeof(struct idx_map_entry_t));
2615 return ((state->fe_map.entries != NULL) && (state->cb_map.entries != NULL)) ? 0 : 1;
2619 fe_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2623 /* FEs use a one-based indexing system, so old==0 implies no mapping */
2629 if (old >= state->fe_map.len) {
2630 ViceLog(0, ("fe_OldToNew: index %d is out of range\n", old));
2632 } else if (state->fe_map.entries[old].valid != FS_STATE_IDX_VALID ||
2633 state->fe_map.entries[old].old_idx != old) { /* sanity check */
2634 ViceLog(0, ("fe_OldToNew: index %d points to an invalid FileEntry record\n", old));
2637 *new = state->fe_map.entries[old].new_idx;
2645 cb_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2649 /* CBs use a one-based indexing system, so old==0 implies no mapping */
2655 if (old >= state->cb_map.len) {
2656 ViceLog(0, ("cb_OldToNew: index %d is out of range\n", old));
2658 } else if (state->cb_map.entries[old].valid != FS_STATE_IDX_VALID ||
2659 state->cb_map.entries[old].old_idx != old) { /* sanity check */
2660 ViceLog(0, ("cb_OldToNew: index %d points to an invalid CallBack record\n", old));
2663 *new = state->cb_map.entries[old].new_idx;
2669 #endif /* AFS_DEMAND_ATTACH_FS */
2671 #define DumpBytes(fd,buf,req) if (write(fd, buf, req) < 0) ; /* don't care */
2674 DumpCallBackState_r(void)
2677 afs_uint32 magic = MAGICV2, now = (afs_int32) time(NULL), freelisthead;
2679 oflag = O_WRONLY | O_CREAT | O_TRUNC;
2683 fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, oflag, 0666);
2686 ("Couldn't create callback dump file %s\n",
2687 AFSDIR_SERVER_CBKDUMP_FILEPATH));
2691 * Collect but ignoring the return value of write(2) here,
2692 * to avoid compiler warnings on some platforms.
2694 DumpBytes(fd, &magic, sizeof(magic));
2695 DumpBytes(fd, &now, sizeof(now));
2696 DumpBytes(fd, &cbstuff, sizeof(cbstuff));
2697 DumpBytes(fd, TimeOuts, sizeof(TimeOuts));
2698 DumpBytes(fd, timeout, sizeof(timeout));
2699 DumpBytes(fd, &tfirst, sizeof(tfirst));
2700 freelisthead = cbtoi((struct CallBack *)CBfree);
2701 DumpBytes(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2702 freelisthead = fetoi((struct FileEntry *)FEfree);
2703 DumpBytes(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2704 DumpBytes(fd, HashTable, sizeof(HashTable));
2705 DumpBytes(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2706 DumpBytes(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2713 DumpCallBackState(void) {
2717 rc = DumpCallBackState_r();
2723 #endif /* !INTERPRET_DUMP */
2725 #ifdef INTERPRET_DUMP
2728 ReadBytes(int fd, void *buf, size_t req)
2732 count = read(fd, buf, req);
2736 } else if (count != req) {
2737 fprintf(stderr, "read: premature EOF (expected %lu, got %lu)\n",
2738 (unsigned long)req, (unsigned long)count);
2743 /* This is only compiled in for the callback analyzer program */
2744 /* Returns the time of the dump */
2746 ReadDump(char *file, int timebits)
2749 afs_uint32 magic, freelisthead;
2757 fd = open(file, oflag);
2759 fprintf(stderr, "Couldn't read dump file %s\n", file);
2762 ReadBytes(fd, &magic, sizeof(magic));
2763 if (magic == MAGICV2) {
2766 if (magic != MAGIC) {
2768 "Magic number of %s is invalid. You might be trying to\n",
2771 "run this program on a machine type with a different byte ordering.\n");
2775 if (timebits == 64) {
2776 ReadBytes(fd, &now64, sizeof(afs_int64));
2777 now = (afs_int32) now64;
2779 ReadBytes(fd, &now, sizeof(afs_int32));
2781 ReadBytes(fd, &cbstuff, sizeof(cbstuff));
2782 ReadBytes(fd, TimeOuts, sizeof(TimeOuts));
2783 ReadBytes(fd, timeout, sizeof(timeout));
2784 ReadBytes(fd, &tfirst, sizeof(tfirst));
2785 ReadBytes(fd, &freelisthead, sizeof(freelisthead));
2786 CB = ((struct CallBack
2787 *)(calloc(cbstuff.nblks, sizeof(struct CallBack)))) - 1;
2788 FE = ((struct FileEntry
2789 *)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
2790 CBfree = (struct CallBack *)itocb(freelisthead);
2791 ReadBytes(fd, &freelisthead, sizeof(freelisthead));
2792 FEfree = (struct FileEntry *)itofe(freelisthead);
2793 ReadBytes(fd, HashTable, sizeof(HashTable));
2794 ReadBytes(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2795 ReadBytes(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2797 perror("Error reading dumpfile");
2804 #include "AFS_component_version_number.h"
2806 #include "AFS_component_version_number.c"
2809 static afs_uint32 *cbTrack;
2812 main(int argc, char **argv)
2814 int err = 0, cbi = 0, stats = 0, noptions = 0, all = 0, vol = 0, raw = 0;
2816 struct FileEntry *fe;
2817 struct CallBack *cb;
2821 memset(&fid, 0, sizeof(fid));
2824 while (argc && **argv == '-') {
2827 if (!strcmp(*argv, "-host")) {
2833 cbi = atoi(*++argv);
2834 } else if (!strcmp(*argv, "-fid")) {
2840 fid.Volume = atoi(*++argv);
2841 fid.Vnode = atoi(*++argv);
2842 fid.Unique = atoi(*++argv);
2843 } else if (!strcmp(*argv, "-time")) {
2844 fprintf(stderr, "-time not supported\n");
2846 } else if (!strcmp(*argv, "-stats")) {
2848 } else if (!strcmp(*argv, "-all")) {
2850 } else if (!strcmp(*argv, "-raw")) {
2852 } else if (!strcmp(*argv, "-timebits")) {
2858 timebits = atoi(*++argv);
2859 if ((timebits != 32)
2863 } else if (!strcmp(*argv, "-volume")) {
2869 vol = atoi(*++argv);
2874 if (err || argc != 1) {
2876 "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] [-timebits 32"
2878 "] callbackdumpfile\n");
2880 "[cbid is shown for each host in the hosts.dump file]\n");
2883 now = ReadDump(*argv, timebits);
2884 if (stats || noptions == 0) {
2885 time_t uxtfirst = UXtime(tfirst), tnow = now;
2886 printf("The time of the dump was %u %s", (unsigned int) now, ctime(&tnow));
2887 printf("The last time cleanup ran was %u %s", (unsigned int) uxtfirst,
2889 PrintCallBackStats();
2892 cbTrack = calloc(cbstuff.nblks, sizeof(cbTrack[0]));
2897 struct CallBack *cb;
2898 struct FileEntry *fe;
2900 for (hash = 0; hash < FEHASH_SIZE; hash++) {
2901 for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
2902 if (!vol || (fe->volid == vol)) {
2903 afs_uint32 fe_i = fetoi(fe);
2905 for (cb = itocb(fe->firstcb); cb; cb = itocb(cb->cnext)) {
2906 afs_uint32 cb_i = cbtoi(cb);
2908 if (cb_i > cbstuff.nblks) {
2909 printf("CB index out of range (%u > %d), stopped for this FE\n",
2910 cb_i, cbstuff.nblks);
2914 if (cbTrack[cb_i]) {
2915 printf("CB entry already claimed for FE[%u] (this is FE[%u]), stopped\n",
2916 cbTrack[cb_i], fe_i);
2919 cbTrack[cb_i] = fe_i;
2931 afs_uint32 cfirst = cbi;
2936 } while (cbi != cfirst);
2941 printf("No callback entries for %u.%u\n", fid.Volume, fid.Vnode);
2944 cb = itocb(fe->firstcb);
2947 cb = itocb(cb->cnext);
2952 for (i = 1; i < cbstuff.nblks; i++) {
2953 p = (afs_int32 *) & FE[i];
2954 printf("%d:%12x%12x%12x%12x\n", i, p[0], p[1], p[2], p[3]);
2963 PrintCB(struct CallBack *cb, afs_uint32 now)
2965 struct FileEntry *fe = itofe(cb->fhead);
2966 time_t expires = TIndexToTime(cb->thead);
2971 printf("vol=%" AFS_VOLID_FMT " vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %lu secs at %s",
2972 afs_printable_VolumeId_lu(fe->volid), fe->vnode, fe->ncbs,
2973 cb->hhead, cb->status, fe->status, expires - now, ctime(&expires));
2978 #if !defined(INTERPRET_DUMP)
2980 ** try breaking calbacks on afidp from host. Use multi_rx.
2981 ** return 0 on success, non-zero on failure
2984 MultiBreakCallBackAlternateAddress(struct host *host, struct AFSCBFids *afidp)
2988 retVal = MultiBreakCallBackAlternateAddress_r(host, afidp);
2994 MultiBreakCallBackAlternateAddress_r(struct host *host,
2995 struct AFSCBFids *afidp)
2998 struct rx_connection **conns;
2999 struct rx_connection *connSuccess = 0;
3000 struct AddrPort *interfaces;
3001 static struct rx_securityClass *sc = 0;
3002 static struct AFSCBs tc = { 0, 0 };
3005 /* nothing more can be done */
3006 if (!host->interface)
3007 return 1; /* failure */
3009 /* the only address is the primary interface */
3010 if (host->interface->numberOfInterfaces <= 1)
3011 return 1; /* failure */
3013 /* initialise a security object only once */
3015 sc = rxnull_NewClientSecurityObject();
3017 i = host->interface->numberOfInterfaces;
3018 interfaces = calloc(i, sizeof(struct AddrPort));
3019 conns = calloc(i, sizeof(struct rx_connection *));
3020 if (!interfaces || !conns) {
3021 ViceLogThenPanic(0, ("Failed malloc in "
3022 "MultiBreakCallBackAlternateAddress_r\n"));
3025 /* initialize alternate rx connections */
3026 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3027 /* this is the current primary address */
3028 if (host->host == host->interface->interface[i].addr &&
3029 host->port == host->interface->interface[i].port)
3032 interfaces[j] = host->interface->interface[i];
3034 rx_NewConnection(interfaces[j].addr,
3035 interfaces[j].port, 1, sc, 0);
3036 rx_SetConnDeadTime(conns[j], 2);
3037 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3041 opr_Assert(j); /* at least one alternate address */
3043 ("Starting multibreakcall back on all addr for host %p (%s:%d)\n",
3044 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
3046 multi_Rx(conns, j) {
3047 multi_RXAFSCB_CallBack(afidp, &tc);
3051 if (host->callback_rxcon)
3052 rx_DestroyConnection(host->callback_rxcon);
3053 host->callback_rxcon = conns[multi_i];
3054 /* add then remove */
3055 addInterfaceAddr_r(host, interfaces[multi_i].addr,
3056 interfaces[multi_i].port);
3057 removeInterfaceAddr_r(host, host->host, host->port);
3058 host->host = interfaces[multi_i].addr;
3059 host->port = interfaces[multi_i].port;
3060 connSuccess = conns[multi_i];
3061 rx_SetConnDeadTime(host->callback_rxcon, 50);
3062 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3064 ("multibreakcall success with addr %s:%d\n",
3065 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3066 ntohs(interfaces[multi_i].port)));
3073 /* Destroy all connections except the one on which we succeeded */
3074 for (i = 0; i < j; i++)
3075 if (conns[i] != connSuccess)
3076 rx_DestroyConnection(conns[i]);
3082 return 0; /* success */
3084 return 1; /* failure */
3089 ** try multi_RX probes to host.
3090 ** return 0 on success, non-0 on failure
3093 MultiProbeAlternateAddress_r(struct host *host)
3096 struct rx_connection **conns;
3097 struct rx_connection *connSuccess = 0;
3098 struct AddrPort *interfaces;
3099 static struct rx_securityClass *sc = 0;
3102 /* nothing more can be done */
3103 if (!host->interface)
3104 return 1; /* failure */
3106 /* the only address is the primary interface */
3107 if (host->interface->numberOfInterfaces <= 1)
3108 return 1; /* failure */
3110 /* initialise a security object only once */
3112 sc = rxnull_NewClientSecurityObject();
3114 i = host->interface->numberOfInterfaces;
3115 interfaces = calloc(i, sizeof(struct AddrPort));
3116 conns = calloc(i, sizeof(struct rx_connection *));
3117 if (!interfaces || !conns) {
3118 ViceLogThenPanic(0, ("Failed malloc in "
3119 "MultiProbeAlternateAddress_r\n"));
3122 /* initialize alternate rx connections */
3123 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3124 /* this is the current primary address */
3125 if (host->host == host->interface->interface[i].addr &&
3126 host->port == host->interface->interface[i].port)
3129 interfaces[j] = host->interface->interface[i];
3131 rx_NewConnection(interfaces[j].addr,
3132 interfaces[j].port, 1, sc, 0);
3133 rx_SetConnDeadTime(conns[j], 2);
3134 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3138 opr_Assert(j); /* at least one alternate address */
3140 ("Starting multiprobe on all addr for host %p (%s:%d)\n",
3141 host, afs_inet_ntoa_r(host->host, hoststr),
3142 ntohs(host->port)));
3144 multi_Rx(conns, j) {
3145 multi_RXAFSCB_ProbeUuid(&host->interface->uuid);
3149 if (host->callback_rxcon)
3150 rx_DestroyConnection(host->callback_rxcon);
3151 host->callback_rxcon = conns[multi_i];
3152 /* add then remove */
3153 addInterfaceAddr_r(host, interfaces[multi_i].addr,
3154 interfaces[multi_i].port);
3155 removeInterfaceAddr_r(host, host->host, host->port);
3156 host->host = interfaces[multi_i].addr;
3157 host->port = interfaces[multi_i].port;
3158 connSuccess = conns[multi_i];
3159 rx_SetConnDeadTime(host->callback_rxcon, 50);
3160 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3162 ("multiprobe success with addr %s:%d\n",
3163 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3164 ntohs(interfaces[multi_i].port)));
3169 ("multiprobe failure with addr %s:%d\n",
3170 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3171 ntohs(interfaces[multi_i].port)));
3173 /* This is less than desirable but its the best we can do.
3174 * The AFS Cache Manager will return either 0 for a Uuid
3175 * match and a 1 for a non-match. If the error is 1 we
3176 * therefore know that our mapping of IP address to Uuid
3177 * is wrong. We should attempt to find the correct
3178 * Uuid and fix the host tables.
3180 if (multi_error == 1) {
3181 /* remove the current alternate address from this host */
3183 removeInterfaceAddr_r(host, interfaces[multi_i].addr, interfaces[multi_i].port);
3187 #ifdef AFS_DEMAND_ATTACH_FS
3188 /* try to bail ASAP if the fileserver is shutting down */
3190 if (fs_state.mode == FS_MODE_SHUTDOWN) {
3199 /* Destroy all connections except the one on which we succeeded */
3200 for (i = 0; i < j; i++)
3201 if (conns[i] != connSuccess)
3202 rx_DestroyConnection(conns[i]);
3208 return 0; /* success */
3210 return 1; /* failure */
3213 #endif /* !defined(INTERPRET_DUMP) */