2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
9 * Portions Copyright (c) 2006 Sine Nomine Associates
13 * NEW callback package callback.c (replaces vicecb.c)
14 * Updated call back routines, NOW with:
16 * Faster DeleteVenus (Now called DeleteAllCallBacks)
17 * Call back breaking for volumes
18 * Adaptive timeouts on call backs
19 * Architected for Multi RPC
20 * No locks (currently implicit vnode locks--these will go, to)
21 * Delayed call back when rpc connection down.
22 * Bulk break of delayed call backs when rpc connection
24 * Strict limit on number of call backs.
26 * InitCallBack(nblocks)
27 * Initialize: nblocks is max number # of file entries + # of callback entries
28 * nblocks must be < 65536
29 * Space used is nblocks*16 bytes
30 * Note that space will be reclaimed by breaking callbacks of old hosts
32 * time = AddCallBack(host, fid)
34 * Returns the expiration time at the workstation.
36 * BreakCallBack(host, fid)
37 * Break all call backs for fid, except for the specified host.
40 * BreakVolumeCallBacks(volume)
41 * Break all call backs on volume, using single call to each host
42 * Delete all the call backs.
44 * DeleteCallBack(host,fid)
45 * Delete (do not break) single call back for fid.
47 * DeleteFileCallBacks(fid)
48 * Delete (do not break) all call backs for fid.
50 * DeleteAllCallBacks(host)
51 * Delete (do not break) all call backs for host.
53 * CleanupTimedOutCallBacks()
54 * Delete all timed out call back entries
55 * Must be called periodically by file server.
57 * BreakDelayedCallBacks(host)
58 * Break all delayed call backs for host.
59 * Returns 1: one or more failed, 0: success.
61 * PrintCallBackStats()
62 * Print statistics about call backs to stdout.
64 * DumpCallBacks() ---wishful thinking---
65 * Dump call back state to /tmp/callback.state.
66 * This is separately interpretable by the program pcb.
68 * Notes: In general, if a call back to a host doesn't get through,
69 * then HostDown, supplied elsewhere, is called. BreakDelayedCallBacks,
70 * however, does not call HostDown, but instead returns an indication of
71 * success if all delayed call backs were finally broken.
73 * BreakDelayedCallBacks MUST be called at the first sign of activity
74 * from the host after HostDown has been called (or a previous
75 * BreakDelayedCallBacks failed). The BreakDelayedCallBacks must be
76 * allowed to complete before any requests from that host are handled.
77 * If BreakDelayedCallBacks fails, then the host should remain
78 * down (and the request should be failed).
80 * CleanupCallBacks MUST be called periodically by the file server for
81 * this package to work correctly. Every 5 minutes is suggested.
84 #include <afsconfig.h>
85 #include <afs/param.h>
89 #include <stdlib.h> /* for malloc() */
90 #include <time.h> /* ANSI standard location for time stuff */
100 #include <afs/assert.h>
102 #include <afs/stds.h>
104 #include <afs/nfs.h> /* yuck. This is an abomination. */
107 #include <afs/afscbint.h>
108 #include <afs/afsutil.h>
110 #include <afs/ihandle.h>
111 #include <afs/vnode.h>
112 #include <afs/volume.h>
113 #include "viced_prototypes.h"
116 #include <afs/ptclient.h> /* need definition of prlist for host.h */
118 #include "callback.h"
119 #ifdef AFS_DEMAND_ATTACH_FS
120 #include "../tviced/serialize_state.h"
121 #endif /* AFS_DEMAND_ATTACH_FS */
124 extern afsUUID FS_HostUUID;
125 extern int hostCount;
127 #ifndef INTERPRET_DUMP
128 static int ShowProblems = 1;
131 struct cbcounters cbstuff;
133 static struct FileEntry * FE = NULL; /* don't use FE[0] */
134 static struct CallBack * CB = NULL; /* don't use CB[0] */
136 static struct CallBack * CBfree = NULL;
137 static struct FileEntry * FEfree = NULL;
140 /* Time to live for call backs depends upon number of users of the file.
141 * TimeOuts is indexed by this number/8 (using TimeOut macro). Times
142 * in this table are for the workstation; server timeouts, add
145 static int TimeOuts[] = {
146 /* Note: don't make the first entry larger than 4 hours (see above) */
147 4 * 60 * 60, /* 0-7 users */
148 1 * 60 * 60, /* 8-15 users */
149 30 * 60, /* 16-23 users */
150 15 * 60, /* 24-31 users */
151 15 * 60, /* 32-39 users */
152 10 * 60, /* 40-47 users */
153 10 * 60, /* 48-55 users */
154 10 * 60, /* 56-63 users */
155 }; /* Anything more: MinTimeOut */
157 /* minimum time given for a call back */
158 #ifndef INTERPRET_DUMP
159 static int MinTimeOut = (7 * 60);
162 /* Heads of CB queues; a timeout index is 1+index into this array */
163 static afs_uint32 timeout[CB_NUM_TIMEOUT_QUEUES];
165 static afs_int32 tfirst; /* cbtime of oldest unexpired call back time queue */
168 /* 16 byte object get/free routines */
173 /* Prototypes for static routines */
174 static struct FileEntry *FindFE(register AFSFid * fid);
176 #ifndef INTERPRET_DUMP
177 static struct CallBack *iGetCB(register int *nused);
178 static int iFreeCB(register struct CallBack *cb, register int *nused);
179 static struct FileEntry *iGetFE(register int *nused);
180 static int iFreeFE(register struct FileEntry *fe, register int *nused);
181 static int TAdd(register struct CallBack *cb, register afs_uint32 * thead);
182 static int TDel(register struct CallBack *cb);
183 static int HAdd(register struct CallBack *cb, register struct host *host);
184 static int HDel(register struct CallBack *cb);
185 static int CDel(struct CallBack *cb, int deletefe);
186 static int CDelPtr(register struct FileEntry *fe, register afs_uint32 * cbp,
188 static afs_uint32 *FindCBPtr(struct FileEntry *fe, struct host *host);
189 static int FDel(register struct FileEntry *fe);
190 static int AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead,
191 int type, int locked);
192 static void MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
193 struct AFSCBFids *afidp, struct host *xhost);
194 static int MultiBreakVolumeCallBack_r(struct host *host, int isheld,
195 struct VCBParams *parms, int deletefe);
196 static int MultiBreakVolumeCallBack(struct host *host, int isheld,
198 static int MultiBreakVolumeLaterCallBack(struct host *host, int isheld,
200 static int GetSomeSpace_r(struct host *hostp, int locked);
201 static int ClearHostCallbacks_r(struct host *hp, int locked);
204 #define GetCB() ((struct CallBack *)iGetCB(&cbstuff.nCBs))
205 #define GetFE() ((struct FileEntry *)iGetFE(&cbstuff.nFEs))
206 #define FreeCB(cb) iFreeCB((struct CallBack *)cb, &cbstuff.nCBs)
207 #define FreeFE(fe) iFreeFE((struct FileEntry *)fe, &cbstuff.nFEs)
210 /* Other protos - move out sometime */
211 void PrintCB(register struct CallBack *cb, afs_uint32 now);
213 static afs_uint32 HashTable[FEHASH_SIZE]; /* File entry hash table */
215 static struct FileEntry *
216 FindFE(register AFSFid * fid)
220 register struct FileEntry *fe;
222 hash = FEHash(fid->Volume, fid->Unique);
223 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
225 if (fe->volid == fid->Volume && fe->unique == fid->Unique
226 && fe->vnode == fid->Vnode && (fe->status & FE_LATER) != FE_LATER)
232 #ifndef INTERPRET_DUMP
234 static struct CallBack *
235 iGetCB(register int *nused)
237 register struct CallBack *ret;
239 if ((ret = CBfree)) {
240 CBfree = (struct CallBack *)(((struct object *)ret)->next);
247 iFreeCB(register struct CallBack *cb, register int *nused)
249 ((struct object *)cb)->next = (struct object *)CBfree;
255 static struct FileEntry *
256 iGetFE(register int *nused)
258 register struct FileEntry *ret;
260 if ((ret = FEfree)) {
261 FEfree = (struct FileEntry *)(((struct object *)ret)->next);
268 iFreeFE(register struct FileEntry *fe, register int *nused)
270 ((struct object *)fe)->next = (struct object *)FEfree;
276 /* Add cb to end of specified timeout list */
278 TAdd(register struct CallBack *cb, register afs_uint32 * thead)
281 (*thead) = cb->tnext = cb->tprev = cbtoi(cb);
283 register struct CallBack *thp = itocb(*thead);
285 cb->tprev = thp->tprev;
289 thp->tprev = (itocb(thp->tprev)->tnext = cbtoi(cb));
291 thp->tprev = cbtoi(cb);
294 cb->thead = ttoi(thead);
298 /* Delete call back entry from timeout list */
300 TDel(register struct CallBack *cb)
302 register afs_uint32 *thead = itot(cb->thead);
304 if (*thead == cbtoi(cb))
305 *thead = (*thead == cb->tnext ? 0 : cb->tnext);
306 if (itocb(cb->tprev))
307 itocb(cb->tprev)->tnext = cb->tnext;
308 if (itocb(cb->tnext))
309 itocb(cb->tnext)->tprev = cb->tprev;
313 /* Add cb to end of specified host list */
315 HAdd(register struct CallBack *cb, register struct host *host)
317 cb->hhead = h_htoi(host);
319 host->cblist = cb->hnext = cb->hprev = cbtoi(cb);
321 register struct CallBack *fcb = itocb(host->cblist);
323 cb->hprev = fcb->hprev;
324 cb->hnext = cbtoi(fcb);
325 fcb->hprev = (itocb(fcb->hprev)->hnext = cbtoi(cb));
330 /* Delete call back entry from host list */
332 HDel(register struct CallBack *cb)
334 register afs_uint32 *hhead = &h_itoh(cb->hhead)->cblist;
336 if (*hhead == cbtoi(cb))
337 *hhead = (*hhead == cb->hnext ? 0 : cb->hnext);
338 itocb(cb->hprev)->hnext = cb->hnext;
339 itocb(cb->hnext)->hprev = cb->hprev;
343 /* Delete call back entry from fid's chain of cb's */
344 /* N.B. This one also deletes the CB, and also possibly parent FE, so
345 * make sure that it is not on any other list before calling this
348 CDel(struct CallBack *cb, int deletefe)
351 struct FileEntry *fe = itofe(cb->fhead);
352 register afs_uint32 *cbp;
355 for (safety = 0, cbp = &fe->firstcb; *cbp && *cbp != cbi;
356 cbp = &itocb(*cbp)->cnext, safety++) {
357 if (safety > cbstuff.nblks + 10) {
360 ("CDel: Internal Error -- shutting down: wanted %d from %d, now at %d\n",
361 cbi, fe->firstcb, *cbp));
363 ShutDownAndCore(PANIC);
366 CDelPtr(fe, cbp, deletefe);
370 /* Same as CDel, but pointer to parent pointer to CB entry is passed,
371 * as well as file entry */
372 /* N.B. This one also deletes the CB, and also possibly parent FE, so
373 * make sure that it is not on any other list before calling this
375 static int Ccdelpt = 0, CcdelB = 0;
378 CDelPtr(register struct FileEntry *fe, register afs_uint32 * cbp,
381 register struct CallBack *cb;
391 if ((--fe->ncbs == 0) && deletefe)
397 FindCBPtr(struct FileEntry *fe, struct host *host)
399 register afs_uint32 hostindex = h_htoi(host);
400 register struct CallBack *cb;
401 register afs_uint32 *cbp;
404 for (safety = 0, cbp = &fe->firstcb; *cbp; cbp = &cb->cnext, safety++) {
405 if (safety > cbstuff.nblks) {
406 ViceLog(0, ("FindCBPtr: Internal Error -- shutting down.\n"));
408 ShutDownAndCore(PANIC);
411 if (cb->hhead == hostindex)
417 /* Delete file entry from hash table */
419 FDel(register struct FileEntry *fe)
421 register int fei = fetoi(fe);
422 register afs_uint32 *p = &HashTable[FEHash(fe->volid, fe->unique)];
424 while (*p && *p != fei)
425 p = &itofe(*p)->fnext;
432 /* initialize the callback package */
434 InitCallBack(int nblks)
437 tfirst = CBtime(FT_ApproxTime());
438 /* N.B. The "-1", below, is because
439 * FE[0] and CB[0] are not used--and not allocated */
440 FE = ((struct FileEntry *)(calloc(nblks, sizeof(struct FileEntry))));
442 ViceLog(0, ("Failed malloc in InitCallBack\n"));
445 FE--; /* FE[0] is supposed to point to junk */
446 cbstuff.nFEs = nblks;
448 FreeFE(&FE[cbstuff.nFEs]); /* This is correct */
449 CB = ((struct CallBack *)(calloc(nblks, sizeof(struct CallBack))));
451 ViceLog(0, ("Failed malloc in InitCallBack\n"));
454 CB--; /* CB[0] is supposed to point to junk */
455 cbstuff.nCBs = nblks;
457 FreeCB(&CB[cbstuff.nCBs]); /* This is correct */
458 cbstuff.nblks = nblks;
459 cbstuff.nbreakers = 0;
465 XCallBackBulk_r(struct host * ahost, struct AFSFid * fids, afs_int32 nfids)
467 struct AFSCallBack tcbs[AFSCBMAX];
473 struct rx_connection *cb_conn = NULL;
476 rx_SetConnDeadTime(ahost->callback_rxcon, 4);
477 rx_SetConnHardDeadTime(ahost->callback_rxcon, AFS_HARDDEADTIME);
484 for (i = 0; i < nfids && i < AFSCBMAX; i++) {
485 tcbs[i].CallBackVersion = CALLBACK_VERSION;
486 tcbs[i].ExpirationTime = 0;
487 tcbs[i].CallBackType = CB_DROPPED;
489 tf.AFSCBFids_len = i;
490 tf.AFSCBFids_val = &(fids[j]);
494 tc.AFSCBs_val = tcbs;
496 cb_conn = ahost->callback_rxcon;
497 rx_GetConnection(cb_conn);
499 code |= RXAFSCB_CallBack(cb_conn, &tf, &tc);
500 rx_PutConnection(cb_conn);
508 /* the locked flag tells us if the host entry has already been locked
509 * by our parent. I don't think anybody actually calls us with the
510 * host locked, but here's how to make that work: GetSomeSpace has to
511 * change so that it doesn't attempt to lock any hosts < "host". That
512 * means that it might be unable to free any objects, so it has to
513 * return an exit status. If it fails, then AddCallBack1 might fail,
514 * as well. If so, the host->ResetDone should probably be set to 0,
515 * and we probably don't want to return a callback promise to the
516 * cache manager, either. */
518 AddCallBack1(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
526 if (!(host->hostFlags & HOSTDELETED))
527 retVal = AddCallBack1_r(host, fid, thead, type, 1);
537 AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
540 struct FileEntry *fe;
541 struct CallBack *cb = 0, *lastcb = 0;
542 struct FileEntry *newfe = 0;
543 afs_uint32 time_out = 0;
544 afs_uint32 *Thead = thead;
545 struct CallBack *newcb = 0;
548 cbstuff.AddCallBacks++;
552 /* allocate these guys first, since we can't call the allocator with
553 * the host structure locked -- or we might deadlock. However, we have
554 * to avoid races with FindFE... */
555 while (!(newcb = GetCB())) {
556 GetSomeSpace_r(host, locked);
558 while (!(newfe = GetFE())) { /* Get it now, so we don't have to call */
559 /* GetSomeSpace with the host locked, later. This might turn out to */
560 /* have been unneccessary, but that's actually kind of unlikely, since */
561 /* most files are not shared. */
562 GetSomeSpace_r(host, locked);
566 h_Lock_r(host); /* this can yield, so do it before we get any */
568 if (host->hostFlags & HOSTDELETED) {
576 if (type == CB_NORMAL) {
578 TimeCeiling(FT_ApproxTime() + TimeOut(fe ? fe->ncbs : 0) +
580 Thead = THead(CBtime(time_out));
581 } else if (type == CB_VOLUME) {
582 time_out = TimeCeiling((60 * 120 + FT_ApproxTime()) + ServerBias);
583 Thead = THead(CBtime(time_out));
584 } else if (type == CB_BULK) {
585 /* bulk status can get so many callbacks all at once, and most of them
586 * are probably not for things that will be used for long.
589 TimeCeiling(FT_ApproxTime() + ServerBias +
590 TimeOut(22 + (fe ? fe->ncbs : 0)));
591 Thead = THead(CBtime(time_out));
597 register afs_uint32 hash;
602 fe->volid = fid->Volume;
603 fe->vnode = fid->Vnode;
604 fe->unique = fid->Unique;
607 hash = FEHash(fid->Volume, fid->Unique);
608 fe->fnext = HashTable[hash];
609 HashTable[hash] = fetoi(fe);
611 for (safety = 0, lastcb = cb = itocb(fe->firstcb); cb;
612 lastcb = cb, cb = itocb(cb->cnext), safety++) {
613 if (safety > cbstuff.nblks) {
614 ViceLog(0, ("AddCallBack1: Internal Error -- shutting down.\n"));
616 ShutDownAndCore(PANIC);
618 if (cb->hhead == h_htoi(host))
621 if (cb) { /* Already have call back: move to new timeout list */
622 /* don't change delayed callbacks back to normal ones */
623 if (cb->status != CB_DELAYED)
625 /* Only move if new timeout is longer */
626 if (TNorm(ttoi(Thead)) > TNorm(cb->thead)) {
630 if (newfe == NULL) { /* we are using the new FE */
631 fe->firstcb = cbtoi(cb);
633 cb->fhead = fetoi(fe);
638 *(lastcb ? &lastcb->cnext : &fe->firstcb) = cbtoi(cb);
641 cb->fhead = fetoi(fe);
647 /* now free any still-unused callback or host entries */
653 if (!locked) /* freecb and freefe might(?) yield */
656 if (type == CB_NORMAL || type == CB_VOLUME || type == CB_BULK)
657 return time_out - ServerBias; /* Expires sooner at workstation */
663 CompareCBA(const void *e1, const void *e2)
665 const struct cbstruct *cba1 = (const struct cbstruct *)e1;
666 const struct cbstruct *cba2 = (const struct cbstruct *)e2;
667 return ((cba1->hp)->index - (cba2->hp)->index);
670 /* Take an array full of hosts, all held. Break callbacks to them, and
671 * release the holds once you're done, except don't release xhost. xhost
672 * may be NULL. Currently only works for a single Fid in afidp array.
673 * If you want to make this work with multiple fids, you need to fix
674 * the error handling. One approach would be to force a reset if a
675 * multi-fid call fails, or you could add delayed callbacks for each
676 * fid. You probably also need to sort and remove duplicate hosts.
677 * When this is called from the BreakVolumeCallBacks path, it does NOT
678 * force a reset if the RPC fails, it just marks the host down and tries
679 * to create a delayed callback. */
680 /* N.B. be sure that code works when ncbas == 0 */
681 /* N.B. requires all the cba[*].hp pointers to be valid... */
682 /* This routine does not hold a lock on the host for the duration of
683 * the BreakCallBack RPC, which is a significant deviation from tradition.
684 * It _does_ get a lock on the host before setting VenusDown = 1,
685 * which is sufficient only if VenusDown = 0 only happens when the
686 * lock is held over the RPC and the subsequent VenusDown == 0
687 * wherever that is done. */
689 MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
690 struct AFSCBFids *afidp, struct host *xhost)
693 struct rx_connection *conns[MAX_CB_HOSTS];
694 static struct AFSCBs tc = { 0, 0 };
695 int multi_to_cba_map[MAX_CB_HOSTS];
697 assert(ncbas <= MAX_CB_HOSTS);
699 /* sort cba list to avoid makecall issues */
700 qsort(cba, ncbas, sizeof(struct cbstruct), CompareCBA);
702 /* set up conns for multi-call */
703 for (i = 0, j = 0; i < ncbas; i++) {
704 struct host *thishost = cba[i].hp;
705 if (!thishost || (thishost->hostFlags & HOSTDELETED)) {
708 rx_GetConnection(thishost->callback_rxcon);
709 multi_to_cba_map[j] = i;
710 conns[j++] = thishost->callback_rxcon;
713 rx_SetConnDeadTime(thishost->callback_rxcon, 4);
714 rx_SetConnHardDeadTime(thishost->callback_rxcon, AFS_HARDDEADTIME);
718 if (j) { /* who knows what multi would do with 0 conns? */
722 multi_RXAFSCB_CallBack(afidp, &tc);
728 i = multi_to_cba_map[multi_i];
734 ("BCB: INTERNAL ERROR: hp=%x, cba=%x, thead=%u\n",
738 ** try breaking callbacks on alternate interface addresses
740 if (MultiBreakCallBackAlternateAddress(hp, afidp)) {
743 ("BCB: Failed on file %u.%u.%u, Host %x (%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++) {
776 if (hp && xhost != hp) {
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];
816 ("BCB: BreakCallBack(Host %x all but %s:%d, (%u,%u,%u))\n",
817 xhost, afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
818 fid->Volume, fid->Vnode, fid->Unique));
821 cbstuff.BreakCallBacks++;
826 hostindex = h_htoi(xhost);
827 cb = itocb(fe->firstcb);
828 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
829 /* the most common case is what follows the || */
832 tf.AFSCBFids_len = 1;
833 tf.AFSCBFids_val = fid;
836 for (ncbas = 0; cb && ncbas < MAX_CB_HOSTS; cb = nextcb) {
837 nextcb = itocb(cb->cnext);
838 if ((cb->hhead != hostindex || flag)
839 && (cb->status == CB_BULK || cb->status == CB_NORMAL
840 || cb->status == CB_VOLUME)) {
841 struct host *thishost = h_itoh(cb->hhead);
843 ViceLog(0, ("BCB: BOGUS! cb->hhead is NULL!\n"));
844 } else if (thishost->hostFlags & VENUSDOWN) {
846 ("BCB: %x (%s:%d) is down; delaying break call back\n",
847 thishost, afs_inet_ntoa_r(thishost->host, hoststr),
848 ntohs(thishost->port)));
849 cb->status = CB_DELAYED;
852 cba[ncbas].hp = thishost;
853 cba[ncbas].thead = cb->thead;
857 CDel(cb, 1); /* Usually first; so this delete
858 * is reasonably inexpensive */
864 MultiBreakCallBack_r(cba, ncbas, &tf, xhost);
866 /* we need to to all these initializations again because MultiBreakCallBack may block */
871 cb = itocb(fe->firstcb);
872 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
873 /* the most common case is what follows the || */
884 /* Delete (do not break) single call back for fid */
886 DeleteCallBack(struct host *host, AFSFid * fid)
888 register struct FileEntry *fe;
889 register afs_uint32 *pcb;
893 cbstuff.DeleteCallBacks++;
896 /* do not care if the host has been HOSTDELETED */
902 ("DCB: No call backs for fid (%u, %u, %u)\n", fid->Volume,
903 fid->Vnode, fid->Unique));
906 pcb = FindCBPtr(fe, host);
909 ("DCB: No call back for host %x (%s:%d), (%u, %u, %u)\n",
910 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
911 fid->Volume, fid->Vnode, fid->Unique));
925 * Delete (do not break) all call backs for fid. This call doesn't
926 * set all of the various host locks, but it shouldn't really matter
927 * since we're not adding callbacks, but deleting them. I'm not sure
928 * why it doesn't set the lock, however; perhaps it should.
931 DeleteFileCallBacks(AFSFid * fid)
933 register struct FileEntry *fe;
934 register struct CallBack *cb;
935 register afs_uint32 cbi;
939 cbstuff.DeleteFiles++;
944 ("DF: No fid (%u,%u,%u) to delete\n", fid->Volume, fid->Vnode,
948 for (n = 0, cbi = fe->firstcb; cbi; n++) {
961 /* Delete (do not break) all call backs for host. The host should be
964 DeleteAllCallBacks_r(struct host *host, int deletefe)
966 register struct CallBack *cb;
967 register int cbi, first;
969 cbstuff.DeleteAllCallBacks++;
970 cbi = first = host->cblist;
972 ViceLog(8, ("DV: no call backs\n"));
980 } while (cbi != first);
986 * Break all delayed call backs for host. Returns 1 if all call backs
987 * successfully broken; 0 otherwise. Assumes host is h_Held and h_Locked.
988 * Must be called with VenusDown set for this host
991 BreakDelayedCallBacks(struct host *host)
995 retVal = BreakDelayedCallBacks_r(host);
1001 BreakDelayedCallBacks_r(struct host *host)
1003 struct AFSFid fids[AFSCBMAX];
1004 u_byte thead[AFSCBMAX]; /* This should match thead in struct Callback */
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) for Host %x (%s:%d)\n",
1032 host, afs_inet_ntoa_r(host->host, hoststr),
1033 ntohs(host->port)));
1035 host->hostFlags |= VENUSDOWN;
1038 ("InitCallBackState success on %x (%s:%d)\n",
1039 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
1040 /* reset was done successfully */
1041 host->hostFlags |= RESETDONE;
1042 host->hostFlags &= ~VENUSDOWN;
1045 while (!(host->hostFlags & HOSTDELETED)) {
1047 host->hostFlags &= ~VENUSDOWN; /* presume up */
1048 cbi = first = host->cblist;
1052 first = host->cblist;
1055 if (cb->status == CB_DELAYED) {
1056 register struct FileEntry *fe = itofe(cb->fhead);
1057 thead[nfids] = cb->thead;
1058 fids[nfids].Volume = fe->volid;
1059 fids[nfids].Vnode = fe->vnode;
1060 fids[nfids].Unique = fe->unique;
1066 } while (cbi && cbi != first && nfids < AFSCBMAX);
1072 if (XCallBackBulk_r(host, fids, nfids)) {
1073 /* Failed, again: put them back, probably with old
1078 ("CB: XCallBackBulk failed, Host %x (%s:%d); callback list follows:\n",
1079 host, afs_inet_ntoa_r(host->host, hoststr),
1080 ntohs(host->port)));
1082 for (i = 0; i < nfids; i++) {
1085 ("CB: Host %x (%s:%d), file %u.%u.%u (part of bulk callback)\n",
1086 host, afs_inet_ntoa_r(host->host, hoststr),
1087 ntohs(host->port), fids[i].Volume,
1088 fids[i].Vnode, fids[i].Unique));
1091 * AddCallBack1_r(host, &fids[i], itot(thead[i]), CB_DELAYED, 1);
1092 * * but it turns out to cause too many tricky locking problems.
1093 * * now, if break delayed fails, screw it. */
1095 host->hostFlags |= VENUSDOWN; /* Failed */
1096 ClearHostCallbacks_r(host, 1 /* locked */ );
1100 if (nfids < AFSCBMAX)
1104 cbstuff.nbreakers--;
1105 /* If we succeeded it's always ok to unset HFE_LATER */
1106 if (!(host->hostFlags & VENUSDOWN))
1107 host->hostFlags &= ~HFE_LATER;
1108 return (host->hostFlags & VENUSDOWN);
1112 ** isheld is 0 if the host is held in h_Enumerate
1113 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1116 MultiBreakVolumeCallBack_r(struct host *host, int isheld,
1117 struct VCBParams *parms, int deletefe)
1122 return isheld; /* host is held only by h_Enumerate, do nothing */
1123 if (host->hostFlags & HOSTDELETED)
1124 return 0; /* host is deleted, release hold */
1126 if (host->hostFlags & VENUSDOWN) {
1128 /* Do not care if the host is now HOSTDELETED */
1131 ("BVCB: volume callback for Host %x (%s:%d) failed\n",
1132 host, afs_inet_ntoa_r(host->host, hoststr),
1133 ntohs(host->port)));
1135 DeleteAllCallBacks_r(host, deletefe); /* Delete all callback state
1136 * rather than attempting to
1137 * selectively remember to
1138 * delete the volume callbacks
1140 host->hostFlags &= ~RESETDONE; /* Do InitCallBackState when host returns */
1142 return 0; /* release hold */
1144 assert(parms->ncbas <= MAX_CB_HOSTS);
1146 /* Do not call MultiBreakCallBack on the current host structure
1147 ** because it would prematurely release the hold on the host
1149 if (parms->ncbas == MAX_CB_HOSTS) {
1150 struct AFSCBFids tf;
1152 tf.AFSCBFids_len = 1;
1153 tf.AFSCBFids_val = parms->fid;
1155 /* this releases all the hosts */
1156 MultiBreakCallBack_r(parms->cba, parms->ncbas, &tf, 0 /* xhost */ );
1160 parms->cba[parms->ncbas].hp = host;
1161 parms->cba[(parms->ncbas)++].thead = parms->thead;
1162 return 1; /* DON'T release hold, because we still need it. */
1166 ** isheld is 0 if the host is held in h_Enumerate
1167 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1170 MultiBreakVolumeCallBack(struct host *host, int isheld, void *rock)
1172 struct VCBParams *parms = (struct VCBParams *) rock;
1176 retval = MultiBreakVolumeCallBack_r(host, isheld, parms, 1);
1182 ** isheld is 0 if the host is held in h_Enumerate
1183 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1186 MultiBreakVolumeLaterCallBack(struct host *host, int isheld, void *rock)
1188 struct VCBParams *parms = (struct VCBParams *)rock;
1191 retval = MultiBreakVolumeCallBack_r(host, isheld, parms, 0);
1197 * Break all call backs on a single volume. Don't call this with any
1198 * hosts h_held. Note that this routine clears the callbacks before
1199 * actually breaking them, and that the vnode isn't locked during this
1200 * operation, so that people might see temporary callback loss while
1201 * this function is executing. It is just a temporary state, however,
1202 * since the callback will be broken later by this same function.
1204 * Now uses multi-RX for CallBack RPC. Note that the
1205 * multiBreakCallBacks routine does not force a reset if the RPC
1206 * fails, unlike the previous version of this routine, but does create
1207 * a delayed callback. Resets will be forced if the host is
1208 * determined to be down before the RPC is executed.
1211 BreakVolumeCallBacks(afs_uint32 volume)
1216 struct CallBack *cb;
1217 struct FileEntry *fe;
1219 struct VCBParams henumParms;
1220 afs_uint32 tthead = 0; /* zero is illegal value */
1223 fid.Volume = volume, fid.Vnode = fid.Unique = 0;
1224 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1225 for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
1226 if (fe->volid == volume) {
1227 register struct CallBack *cbnext;
1228 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1229 host = h_itoh(cb->hhead);
1231 cbnext = itocb(cb->cnext);
1232 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1238 /* leave hold for MultiBreakVolumeCallBack to clear */
1249 /* didn't find any callbacks, so return right away. */
1253 henumParms.ncbas = 0;
1254 henumParms.fid = &fid;
1255 henumParms.thead = tthead;
1257 h_Enumerate(MultiBreakVolumeCallBack, &henumParms);
1259 if (henumParms.ncbas) { /* do left-overs */
1260 struct AFSCBFids tf;
1261 tf.AFSCBFids_len = 1;
1262 tf.AFSCBFids_val = &fid;
1264 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf, 0);
1266 henumParms.ncbas = 0;
1272 #ifdef AFS_PTHREAD_ENV
1273 extern pthread_cond_t fsync_cond;
1275 extern char fsync_wait[];
1279 BreakVolumeCallBacksLater(afs_uint32 volume)
1283 struct FileEntry *fe;
1284 struct CallBack *cb;
1288 ViceLog(25, ("Setting later on volume %u\n", volume));
1290 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1291 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1292 if (fe->volid == volume) {
1293 register struct CallBack *cbnext;
1294 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1295 host = h_itoh(cb->hhead);
1296 host->hostFlags |= HFE_LATER;
1297 cb->status = CB_DELAYED;
1298 cbnext = itocb(cb->cnext);
1301 fe->status |= FE_LATER;
1310 /* didn't find any callbacks, so return right away. */
1314 ViceLog(25, ("Fsync thread wakeup\n"));
1315 #ifdef AFS_PTHREAD_ENV
1317 assert(pthread_cond_broadcast(&fsync_cond) == 0);
1320 LWP_NoYieldSignal(fsync_wait);
1326 BreakLaterCallBacks(void)
1331 struct CallBack *cb;
1332 struct FileEntry *fe = NULL;
1333 struct FileEntry *myfe = NULL;
1335 struct VCBParams henumParms;
1336 unsigned short tthead = 0; /* zero is illegal value */
1340 ViceLog(25, ("Looking for FileEntries to unchain\n"));
1343 /* Pick the first volume we see to clean up */
1344 fid.Volume = fid.Vnode = fid.Unique = 0;
1346 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1347 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1348 if (fe && (fe->status & FE_LATER)
1349 && (fid.Volume == 0 || fid.Volume == fe->volid)) {
1350 /* Ugly, but used to avoid left side casting */
1351 struct object *tmpfe;
1353 ("Unchaining for %u:%u:%u\n", fe->vnode, fe->unique,
1355 fid.Volume = fe->volid;
1357 fe->status &= ~FE_LATER; /* not strictly needed */
1358 /* Works since volid is deeper than the largest pointer */
1359 tmpfe = (struct object *)fe;
1360 tmpfe->next = (struct object *)myfe;
1373 /* loop over FEs from myfe and free/break */
1375 for (fe = myfe; fe;) {
1376 register struct CallBack *cbnext;
1377 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1378 cbnext = itocb(cb->cnext);
1379 host = h_itoh(cb->hhead);
1380 if (cb->status == CB_DELAYED) {
1382 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1387 CDel(cb, 0); /* Don't let CDel clean up the fe */
1388 /* leave hold for MultiBreakVolumeCallBack to clear */
1391 ("Found host %x (%s:%d) non-DELAYED cb for %u:%u:%u\n",
1392 host, afs_inet_ntoa_r(host->host, hoststr),
1393 ntohs(host->port), fe->vnode, fe->unique, fe->volid));
1397 fe = (struct FileEntry *)((struct object *)fe)->next;
1402 ViceLog(125, ("Breaking volume %u\n", fid.Volume));
1403 henumParms.ncbas = 0;
1404 henumParms.fid = &fid;
1405 henumParms.thead = tthead;
1407 h_Enumerate(MultiBreakVolumeLaterCallBack, (char *)&henumParms);
1409 if (henumParms.ncbas) { /* do left-overs */
1410 struct AFSCBFids tf;
1411 tf.AFSCBFids_len = 1;
1412 tf.AFSCBFids_val = &fid;
1414 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf, 0);
1415 henumParms.ncbas = 0;
1420 /* Arrange to be called again */
1425 * Delete all timed-out call back entries (to be called periodically by file
1429 CleanupTimedOutCallBacks(void)
1432 CleanupTimedOutCallBacks_r();
1438 CleanupTimedOutCallBacks_r(void)
1440 afs_uint32 now = CBtime(FT_ApproxTime());
1441 register afs_uint32 *thead;
1442 register struct CallBack *cb;
1443 register int ntimedout = 0;
1446 while (tfirst <= now) {
1448 cbi = *(thead = THead(tfirst));
1454 ("CCB: deleting timed out call back %x (%s:%d), (%u,%u,%u)\n",
1455 h_itoh(cb->hhead)->host,
1456 afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
1457 h_itoh(cb->hhead)->port, itofe(cb->fhead)->volid,
1458 itofe(cb->fhead)->vnode, itofe(cb->fhead)->unique));
1462 if (ntimedout > cbstuff.nblks) {
1463 ViceLog(0, ("CCB: Internal Error -- shutting down...\n"));
1464 DumpCallBackState();
1465 ShutDownAndCore(PANIC);
1467 } while (cbi != *thead);
1472 cbstuff.CBsTimedOut += ntimedout;
1473 ViceLog(7, ("CCB: deleted %d timed out callbacks\n", ntimedout));
1474 return (ntimedout > 0);
1477 static struct host *lih_host;
1478 static int lih_host_held;
1480 /* Value of host->refCount that allows us to reliably infer that
1481 * host may be held by some other thread */
1482 #define OTHER_MUSTHOLD_LIH 2
1484 /* This version does not allow 'host' to be selected unless its ActiveCall
1485 * is newer than 'hostp' which is the host with the oldest ActiveCall from
1486 * the last pass (if it is provided). We filter out any hosts that are
1487 * are held by other threads.
1490 lih0_r(register struct host *host, register int flags, void *rock)
1492 struct host *hostp = (struct host *) rock;
1494 && (hostp && host != hostp)
1495 && (host->refCount < OTHER_MUSTHOLD_LIH)
1496 && (!lih_host || host->ActiveCall < lih_host->ActiveCall)
1497 && (!hostp || host->ActiveCall > hostp->ActiveCall)) {
1498 if (lih_host != NULL && lih_host_held) {
1499 h_Release_r(lih_host); /* release prev host */
1502 lih_host_held = !flags; /* on i==1, this === (lih_host_held = 1) */
1503 flags = 1; /* now flags is 1, but at next(i), it will be 0 again */
1508 /* This version does not allow 'host' to be selected unless its ActiveCall
1509 * is newer than 'hostp' which is the host with the oldest ActiveCall from
1510 * the last pass (if it is provided). In this second varient, we do not
1511 * prevent held hosts from being selected.
1514 lih1_r(register struct host *host, register int flags, void *rock)
1516 struct host *hostp = (struct host *) rock;
1519 && (hostp && host != hostp)
1520 && (!lih_host || host->ActiveCall < lih_host->ActiveCall)
1521 && (!hostp || host->ActiveCall > hostp->ActiveCall)) {
1522 if (lih_host != NULL && lih_host_held) {
1523 h_Release_r(lih_host); /* really? */
1526 lih_host_held = !flags; /* see note above */
1527 flags = 1; /* see note above */
1532 /* This could be upgraded to get more space each time */
1533 /* first pass: sequentially find the oldest host which isn't held by
1534 anyone for which we can clear callbacks;
1536 /* second pass: sequentially find the oldest host regardless of
1537 whether or not the host is held; skipping 'hostp' */
1538 /* third pass: attempt to clear callbacks from 'hostp' */
1539 /* always called with hostp unlocked */
1541 /* Note: hostlist is ordered most recently created host first and
1542 * its order has no relationship to the most recently used. */
1543 extern struct host *hostList;
1545 GetSomeSpace_r(struct host *hostp, int locked)
1547 register struct host *hp, *hp1, *hp2;
1550 cbstuff.GotSomeSpaces++;
1552 ("GSS: First looking for timed out call backs via CleanupCallBacks\n"));
1553 if (CleanupTimedOutCallBacks_r()) {
1563 h_Enumerate_r(i == 0 ? lih0_r : lih1_r, hp2, (char *)hp1);
1566 /* set in lih_r! private copy before giving up H_LOCK */
1567 int lih_host_held2=lih_host_held;
1569 if ((hp != hostp) && !ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
1574 if (lih_host_held2) {
1582 * Next time try getting callbacks from any host even if
1583 * it's deleted (that's actually great since we can freely
1584 * remove its callbacks) or it's held since the only other
1585 * option is starvation for the file server (i.e. until the
1586 * callback timeout arrives).
1593 ("GSS: Try harder for longest inactive host cnt= %d\n",
1598 /* Could not obtain space from other hosts, clear hostp's callback state */
1603 ClearHostCallbacks_r(hostp, 1 /*already locked */ );
1610 /* locked - set if caller has already locked the host */
1612 ClearHostCallbacks_r(struct host *hp, int locked)
1616 struct rx_connection *cb_conn = NULL;
1619 ("GSS: Delete longest inactive host %x (%s:%d)\n",
1620 hp, afs_inet_ntoa_r(hp->host, hoststr), ntohs(hp->port)));
1624 /** Try a non-blocking lock. If the lock is already held return
1625 * after releasing hold on hp
1628 if (h_NBLock_r(hp)) {
1633 if (hp->Console & 2) {
1635 * If the special console field is set it means that a thread
1636 * is waiting in AddCallBack1 after it set pointers to the
1637 * file entry and/or callback entry. Because of the bogus
1638 * usage of h_hold it won't prevent from another thread, this
1639 * one, to remove all the callbacks so just to be safe we keep
1640 * a reference. NOTE, on the last phase we'll free the calling
1641 * host's callbacks but that's ok...
1645 DeleteAllCallBacks_r(hp, 1);
1646 if (hp->hostFlags & VENUSDOWN) {
1647 hp->hostFlags &= ~RESETDONE; /* remember that we must do a reset */
1648 } else if (!(hp->hostFlags & HOSTDELETED)) {
1649 /* host is up, try a call */
1650 hp->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1651 cb_conn = hp->callback_rxcon;
1652 rx_GetConnection(hp->callback_rxcon);
1653 if (hp->interface) {
1656 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1659 code = RXAFSCB_InitCallBackState(cb_conn);
1661 rx_PutConnection(cb_conn);
1664 hp->hostFlags |= ALTADDR; /* alternate addresses are valid */
1666 /* failed, mark host down and need reset */
1667 hp->hostFlags |= VENUSDOWN;
1668 hp->hostFlags &= ~RESETDONE;
1670 /* reset succeeded, we're done */
1671 hp->hostFlags |= RESETDONE;
1680 #endif /* INTERPRET_DUMP */
1684 PrintCallBackStats(void)
1687 "%d add CB, %d break CB, %d del CB, %d del FE, %d CB's timed out, %d space reclaim, %d del host\n",
1688 cbstuff.AddCallBacks, cbstuff.BreakCallBacks,
1689 cbstuff.DeleteCallBacks, cbstuff.DeleteFiles, cbstuff.CBsTimedOut,
1690 cbstuff.GotSomeSpaces, cbstuff.DeleteAllCallBacks);
1691 fprintf(stderr, "%d CBs, %d FEs, (%d of total of %d 16-byte blocks)\n",
1692 cbstuff.nCBs, cbstuff.nFEs, cbstuff.nCBs + cbstuff.nFEs,
1698 #define MAGIC 0x12345678 /* To check byte ordering of dump when it is read in */
1699 #define MAGICV2 0x12345679 /* To check byte ordering & version of dump when it is read in */
1702 #ifndef INTERPRET_DUMP
1704 #ifdef AFS_DEMAND_ATTACH_FS
1707 * callback state serialization
1709 static int cb_stateSaveTimeouts(struct fs_dump_state * state);
1710 static int cb_stateSaveFEHash(struct fs_dump_state * state);
1711 static int cb_stateSaveFEs(struct fs_dump_state * state);
1712 static int cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe);
1713 static int cb_stateRestoreTimeouts(struct fs_dump_state * state);
1714 static int cb_stateRestoreFEHash(struct fs_dump_state * state);
1715 static int cb_stateRestoreFEs(struct fs_dump_state * state);
1716 static int cb_stateRestoreFE(struct fs_dump_state * state);
1717 static int cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
1718 struct iovec * iov, int niovecs);
1720 static int cb_stateVerifyFEHash(struct fs_dump_state * state);
1721 static int cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe);
1722 static int cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe);
1723 static int cb_stateVerifyTimeoutQueues(struct fs_dump_state * state);
1725 static int cb_stateFEToDiskEntry(struct FileEntry *, struct FEDiskEntry *);
1726 static int cb_stateDiskEntryToFE(struct fs_dump_state * state,
1727 struct FEDiskEntry *, struct FileEntry *);
1729 static int cb_stateCBToDiskEntry(struct CallBack *, struct CBDiskEntry *);
1730 static int cb_stateDiskEntryToCB(struct fs_dump_state * state,
1731 struct CBDiskEntry *, struct CallBack *);
1733 static int cb_stateFillHeader(struct callback_state_header * hdr);
1734 static int cb_stateCheckHeader(struct callback_state_header * hdr);
1736 static int cb_stateAllocMap(struct fs_dump_state * state);
1739 cb_stateSave(struct fs_dump_state * state)
1743 AssignInt64(state->eof_offset, &state->hdr->cb_offset);
1745 /* invalidate callback state header */
1746 memset(state->cb_hdr, 0, sizeof(struct callback_state_header));
1747 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1748 sizeof(struct callback_state_header))) {
1753 fs_stateIncEOF(state, sizeof(struct callback_state_header));
1755 /* dump timeout state */
1756 if (cb_stateSaveTimeouts(state)) {
1761 /* dump fe hashtable state */
1762 if (cb_stateSaveFEHash(state)) {
1767 /* dump callback state */
1768 if (cb_stateSaveFEs(state)) {
1773 /* write the callback state header to disk */
1774 cb_stateFillHeader(state->cb_hdr);
1775 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1776 sizeof(struct callback_state_header))) {
1786 cb_stateRestore(struct fs_dump_state * state)
1790 if (fs_stateReadHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1791 sizeof(struct callback_state_header))) {
1796 if (cb_stateCheckHeader(state->cb_hdr)) {
1801 if (cb_stateAllocMap(state)) {
1806 if (cb_stateRestoreTimeouts(state)) {
1811 if (cb_stateRestoreFEHash(state)) {
1816 /* restore FEs and CBs from disk */
1817 if (cb_stateRestoreFEs(state)) {
1822 /* restore the timeout queue heads */
1823 tfirst = state->cb_hdr->tfirst;
1830 cb_stateRestoreIndices(struct fs_dump_state * state)
1833 struct FileEntry * fe;
1834 struct CallBack * cb;
1836 /* restore indices in the FileEntry structures */
1837 for (i = 1; i < state->fe_map.len; i++) {
1838 if (state->fe_map.entries[i].new_idx) {
1839 fe = itofe(state->fe_map.entries[i].new_idx);
1841 /* restore the fe->fnext entry */
1842 if (fe_OldToNew(state, fe->fnext, &fe->fnext)) {
1847 /* restore the fe->firstcb entry */
1848 if (cb_OldToNew(state, fe->firstcb, &fe->firstcb)) {
1855 /* restore indices in the CallBack structures */
1856 for (i = 1; i < state->cb_map.len; i++) {
1857 if (state->cb_map.entries[i].new_idx) {
1858 cb = itocb(state->cb_map.entries[i].new_idx);
1860 /* restore the cb->cnext entry */
1861 if (cb_OldToNew(state, cb->cnext, &cb->cnext)) {
1866 /* restore the cb->fhead entry */
1867 if (fe_OldToNew(state, cb->fhead, &cb->fhead)) {
1872 /* restore the cb->hhead entry */
1873 if (h_OldToNew(state, cb->hhead, &cb->hhead)) {
1878 /* restore the cb->tprev entry */
1879 if (cb_OldToNew(state, cb->tprev, &cb->tprev)) {
1884 /* restore the cb->tnext entry */
1885 if (cb_OldToNew(state, cb->tnext, &cb->tnext)) {
1890 /* restore the cb->hprev entry */
1891 if (cb_OldToNew(state, cb->hprev, &cb->hprev)) {
1896 /* restore the cb->hnext entry */
1897 if (cb_OldToNew(state, cb->hnext, &cb->hnext)) {
1904 /* restore the timeout queue head indices */
1905 for (i = 0; i < state->cb_timeout_hdr->records; i++) {
1906 if (cb_OldToNew(state, timeout[i], &timeout[i])) {
1912 /* restore the FE hash table queue heads */
1913 for (i = 0; i < state->cb_fehash_hdr->records; i++) {
1914 if (fe_OldToNew(state, HashTable[i], &HashTable[i])) {
1925 cb_stateVerify(struct fs_dump_state * state)
1929 if (cb_stateVerifyFEHash(state)) {
1933 if (cb_stateVerifyTimeoutQueues(state)) {
1941 cb_stateVerifyFEHash(struct fs_dump_state * state)
1944 struct FileEntry * fe;
1945 afs_uint32 fei, chain_len;
1947 for (i = 0; i < FEHASH_SIZE; i++) {
1949 for (fei = HashTable[i], fe = itofe(fei);
1951 fei = fe->fnext, fe = itofe(fei)) {
1952 if (fei > cbstuff.nblks) {
1953 ViceLog(0, ("cb_stateVerifyFEHash: error: index out of range (fei=%d)\n", fei));
1957 if (cb_stateVerifyFE(state, fe)) {
1960 if (chain_len > FS_STATE_FE_MAX_HASH_CHAIN_LEN) {
1961 ViceLog(0, ("cb_stateVerifyFEHash: error: hash chain %d length exceeds %d; assuming there's a loop\n",
1962 i, FS_STATE_FE_MAX_HASH_CHAIN_LEN));
1974 cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe)
1978 if ((fe->firstcb && !fe->ncbs) ||
1979 (!fe->firstcb && fe->ncbs)) {
1980 ViceLog(0, ("cb_stateVerifyFE: error: fe->firstcb does not agree with fe->ncbs (fei=%d, fe->firstcb=%d, fe->ncbs=%d)\n",
1981 fetoi(fe), fe->firstcb, fe->ncbs));
1984 if (cb_stateVerifyFCBList(state, fe)) {
1985 ViceLog(0, ("cb_stateVerifyFE: error: FCBList failed verification (fei=%d)\n", 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)
2299 register int fei, hash;
2300 register 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 register struct CallBack * cb;
2490 struct CBDiskEntry * cbdsk;
2495 for (idx = 0; idx < niovecs; idx++) {
2496 cbdsk = (struct CBDiskEntry *) iov[idx].iov_base;
2497 if ((cb = GetCB()) == NULL) {
2498 ViceLog(0, ("cb_stateRestoreCBs: ran out of free CallBack structures\n"));
2502 if (cb_stateDiskEntryToCB(state, cbdsk, cb)) {
2503 ViceLog(0, ("cb_stateRestoreCBs: corrupt CallBack disk entry\n"));
2515 cb_stateFillHeader(struct callback_state_header * hdr)
2517 hdr->stamp.magic = CALLBACK_STATE_MAGIC;
2518 hdr->stamp.version = CALLBACK_STATE_VERSION;
2519 hdr->tfirst = tfirst;
2524 cb_stateCheckHeader(struct callback_state_header * hdr)
2528 if (hdr->stamp.magic != CALLBACK_STATE_MAGIC) {
2530 } else if (hdr->stamp.version != CALLBACK_STATE_VERSION) {
2532 } else if ((hdr->nFEs > cbstuff.nblks) || (hdr->nCBs > cbstuff.nblks)) {
2533 ViceLog(0, ("cb_stateCheckHeader: saved callback state larger than callback memory allocation\n"));
2539 /* disk entry conversion routines */
2541 cb_stateFEToDiskEntry(struct FileEntry * in, struct FEDiskEntry * out)
2543 memcpy(&out->fe, in, sizeof(struct FileEntry));
2544 out->index = fetoi(in);
2549 cb_stateDiskEntryToFE(struct fs_dump_state * state,
2550 struct FEDiskEntry * in, struct FileEntry * out)
2554 memcpy(out, &in->fe, sizeof(struct FileEntry));
2556 /* setup FE map entry */
2557 if (!in->index || (in->index >= state->fe_map.len)) {
2558 ViceLog(0, ("cb_stateDiskEntryToFE: index (%d) out of range",
2563 state->fe_map.entries[in->index].old_idx = in->index;
2564 state->fe_map.entries[in->index].new_idx = fetoi(out);
2571 cb_stateCBToDiskEntry(struct CallBack * in, struct CBDiskEntry * out)
2573 memcpy(&out->cb, in, sizeof(struct CallBack));
2574 out->index = cbtoi(in);
2579 cb_stateDiskEntryToCB(struct fs_dump_state * state,
2580 struct CBDiskEntry * in, struct CallBack * out)
2584 memcpy(out, &in->cb, sizeof(struct CallBack));
2586 /* setup CB map entry */
2587 if (!in->index || (in->index >= state->cb_map.len)) {
2588 ViceLog(0, ("cb_stateDiskEntryToCB: index (%d) out of range\n",
2593 state->cb_map.entries[in->index].old_idx = in->index;
2594 state->cb_map.entries[in->index].new_idx = cbtoi(out);
2600 /* index map routines */
2602 cb_stateAllocMap(struct fs_dump_state * state)
2604 state->fe_map.len = state->cb_hdr->fe_max + 1;
2605 state->cb_map.len = state->cb_hdr->cb_max + 1;
2606 state->fe_map.entries = (struct idx_map_entry_t *)
2607 calloc(state->fe_map.len, sizeof(struct idx_map_entry_t));
2608 state->cb_map.entries = (struct idx_map_entry_t *)
2609 calloc(state->cb_map.len, sizeof(struct idx_map_entry_t));
2610 return ((state->fe_map.entries != NULL) && (state->cb_map.entries != NULL)) ? 0 : 1;
2614 fe_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2618 /* FEs use a one-based indexing system, so old==0 implies no mapping */
2624 if (old >= state->fe_map.len) {
2625 ViceLog(0, ("fe_OldToNew: index %d is out of range\n", old));
2627 } else if (state->fe_map.entries[old].old_idx != old) { /* sanity check */
2628 ViceLog(0, ("fe_OldToNew: index %d points to an invalid FileEntry record\n", old));
2631 *new = state->fe_map.entries[old].new_idx;
2639 cb_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2643 /* CBs use a one-based indexing system, so old==0 implies no mapping */
2649 if (old >= state->cb_map.len) {
2650 ViceLog(0, ("cb_OldToNew: index %d is out of range\n", old));
2652 } else if (state->cb_map.entries[old].old_idx != old) { /* sanity check */
2653 ViceLog(0, ("cb_OldToNew: index %d points to an invalid CallBack record\n", old));
2656 *new = state->cb_map.entries[old].new_idx;
2662 #endif /* AFS_DEMAND_ATTACH_FS */
2665 DumpCallBackState(void)
2668 afs_uint32 magic = MAGICV2, now = (afs_int32) FT_ApproxTime(), freelisthead;
2670 oflag = O_WRONLY | O_CREAT | O_TRUNC;
2674 fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, oflag, 0666);
2677 ("Couldn't create callback dump file %s\n",
2678 AFSDIR_SERVER_CBKDUMP_FILEPATH));
2681 (void)write(fd, &magic, sizeof(magic));
2682 (void)write(fd, &now, sizeof(now));
2683 (void)write(fd, &cbstuff, sizeof(cbstuff));
2684 (void)write(fd, TimeOuts, sizeof(TimeOuts));
2685 (void)write(fd, timeout, sizeof(timeout));
2686 (void)write(fd, &tfirst, sizeof(tfirst));
2687 freelisthead = cbtoi((struct CallBack *)CBfree);
2688 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2689 freelisthead = fetoi((struct FileEntry *)FEfree);
2690 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2691 (void)write(fd, HashTable, sizeof(HashTable));
2692 (void)write(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2693 (void)write(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2699 #endif /* !INTERPRET_DUMP */
2701 #ifdef INTERPRET_DUMP
2703 /* This is only compiled in for the callback analyzer program */
2704 /* Returns the time of the dump */
2706 ReadDump(char *file, int timebits)
2709 afs_uint32 magic, freelisthead;
2711 #ifdef AFS_64BIT_ENV
2719 fd = open(file, oflag);
2721 fprintf(stderr, "Couldn't read dump file %s\n", file);
2724 read(fd, &magic, sizeof(magic));
2725 if (magic == MAGICV2) {
2728 if (magic != MAGIC) {
2730 "Magic number of %s is invalid. You might be trying to\n",
2733 "run this program on a machine type with a different byte ordering.\n");
2737 #ifdef AFS_64BIT_ENV
2738 if (timebits == 64) {
2739 read(fd, &now64, sizeof(afs_int64));
2740 now = (afs_int32) now64;
2743 read(fd, &now, sizeof(afs_int32));
2744 read(fd, &cbstuff, sizeof(cbstuff));
2745 read(fd, TimeOuts, sizeof(TimeOuts));
2746 read(fd, timeout, sizeof(timeout));
2747 read(fd, &tfirst, sizeof(tfirst));
2748 read(fd, &freelisthead, sizeof(freelisthead));
2749 CB = ((struct CallBack
2750 *)(calloc(cbstuff.nblks, sizeof(struct CallBack)))) - 1;
2751 FE = ((struct FileEntry
2752 *)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
2753 CBfree = (struct CallBack *)itocb(freelisthead);
2754 read(fd, &freelisthead, sizeof(freelisthead));
2755 FEfree = (struct FileEntry *)itofe(freelisthead);
2756 read(fd, HashTable, sizeof(HashTable));
2757 read(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2758 read(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2760 perror("Error reading dumpfile");
2767 #include "AFS_component_version_number.h"
2769 #include "AFS_component_version_number.c"
2773 main(int argc, char **argv)
2775 int err = 0, cbi = 0, stats = 0, noptions = 0, all = 0, vol = 0, raw = 0;
2777 register struct FileEntry *fe;
2778 register struct CallBack *cb;
2782 memset(&fid, 0, sizeof(fid));
2785 while (argc && **argv == '-') {
2788 if (!strcmp(*argv, "-host")) {
2794 cbi = atoi(*++argv);
2795 } else if (!strcmp(*argv, "-fid")) {
2801 fid.Volume = atoi(*++argv);
2802 fid.Vnode = atoi(*++argv);
2803 fid.Unique = atoi(*++argv);
2804 } else if (!strcmp(*argv, "-time")) {
2805 fprintf(stderr, "-time not supported\n");
2807 } else if (!strcmp(*argv, "-stats")) {
2809 } else if (!strcmp(*argv, "-all")) {
2811 } else if (!strcmp(*argv, "-raw")) {
2813 } else if (!strcmp(*argv, "-timebits")) {
2819 timebits = atoi(*++argv);
2820 if ((timebits != 32)
2821 #ifdef AFS_64BIT_ENV
2826 } else if (!strcmp(*argv, "-volume")) {
2832 vol = atoi(*++argv);
2837 if (err || argc != 1) {
2839 "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] [-timebits 32"
2840 #ifdef AFS_64BIT_ENV
2843 "] callbackdumpfile\n");
2845 "[cbid is shown for each host in the hosts.dump file]\n");
2848 now = ReadDump(*argv, timebits);
2849 if (stats || noptions == 0) {
2850 time_t uxtfirst = UXtime(tfirst), tnow = now;
2851 printf("The time of the dump was %u %s", (unsigned int) now, ctime(&tnow));
2852 printf("The last time cleanup ran was %u %s", (unsigned int) uxtfirst,
2854 PrintCallBackStats();
2859 struct CallBack *cb;
2860 struct FileEntry *fe;
2862 for (hash = 0; hash < FEHASH_SIZE; hash++) {
2863 for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
2864 if (!vol || (fe->volid == vol)) {
2865 register struct CallBack *cbnext;
2866 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
2868 cbnext = itocb(cb->cnext);
2878 afs_uint32 cfirst = cbi;
2883 } while (cbi != cfirst);
2888 printf("No callback entries for %u.%u\n", fid.Volume, fid.Vnode);
2891 cb = itocb(fe->firstcb);
2894 cb = itocb(cb->cnext);
2899 for (i = 1; i < cbstuff.nblks; i++) {
2900 p = (afs_int32 *) & FE[i];
2901 printf("%d:%12x%12x%12x%12x\n", i, p[0], p[1], p[2], p[3]);
2908 PrintCB(register struct CallBack *cb, afs_uint32 now)
2910 struct FileEntry *fe = itofe(cb->fhead);
2911 time_t expires = TIndexToTime(cb->thead);
2916 printf("vol=%u vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %lu secs at %s",
2917 fe->volid, fe->vnode, fe->ncbs, cb->hhead, cb->status, fe->status,
2918 expires - now, ctime(&expires));
2923 #if !defined(INTERPRET_DUMP)
2925 ** try breaking calbacks on afidp from host. Use multi_rx.
2926 ** return 0 on success, non-zero on failure
2929 MultiBreakCallBackAlternateAddress(struct host *host, struct AFSCBFids *afidp)
2933 retVal = MultiBreakCallBackAlternateAddress_r(host, afidp);
2939 MultiBreakCallBackAlternateAddress_r(struct host *host,
2940 struct AFSCBFids *afidp)
2943 struct rx_connection **conns;
2944 struct rx_connection *connSuccess = 0;
2945 struct AddrPort *interfaces;
2946 static struct rx_securityClass *sc = 0;
2947 static struct AFSCBs tc = { 0, 0 };
2950 /* nothing more can be done */
2951 if (!host->interface)
2952 return 1; /* failure */
2954 /* the only address is the primary interface */
2955 if (host->interface->numberOfInterfaces <= 1)
2956 return 1; /* failure */
2958 /* initialise a security object only once */
2960 sc = rxnull_NewClientSecurityObject();
2962 i = host->interface->numberOfInterfaces;
2963 interfaces = calloc(i, sizeof(struct AddrPort));
2964 conns = calloc(i, sizeof(struct rx_connection *));
2965 if (!interfaces || !conns) {
2967 ("Failed malloc in MultiBreakCallBackAlternateAddress_r\n"));
2971 /* initialize alternate rx connections */
2972 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
2973 /* this is the current primary address */
2974 if (host->host == host->interface->interface[i].addr &&
2975 host->port == host->interface->interface[i].port)
2978 interfaces[j] = host->interface->interface[i];
2980 rx_NewConnection(interfaces[j].addr,
2981 interfaces[j].port, 1, sc, 0);
2982 rx_SetConnDeadTime(conns[j], 2);
2983 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
2987 assert(j); /* at least one alternate address */
2989 ("Starting multibreakcall back on all addr for host %x (%s:%d)\n",
2990 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
2992 multi_Rx(conns, j) {
2993 multi_RXAFSCB_CallBack(afidp, &tc);
2997 if (host->callback_rxcon)
2998 rx_DestroyConnection(host->callback_rxcon);
2999 host->callback_rxcon = conns[multi_i];
3000 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
3001 host->host = interfaces[multi_i].addr;
3002 host->port = interfaces[multi_i].port;
3003 h_AddHostToAddrHashTable_r(host->host, host->port, host);
3004 connSuccess = conns[multi_i];
3005 rx_SetConnDeadTime(host->callback_rxcon, 50);
3006 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3008 ("multibreakcall success with addr %s:%d\n",
3009 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3010 ntohs(interfaces[multi_i].port)));
3017 /* Destroy all connections except the one on which we succeeded */
3018 for (i = 0; i < j; i++)
3019 if (conns[i] != connSuccess)
3020 rx_DestroyConnection(conns[i]);
3026 return 0; /* success */
3028 return 1; /* failure */
3033 ** try multi_RX probes to host.
3034 ** return 0 on success, non-0 on failure
3037 MultiProbeAlternateAddress_r(struct host *host)
3040 struct rx_connection **conns;
3041 struct rx_connection *connSuccess = 0;
3042 struct AddrPort *interfaces;
3043 static struct rx_securityClass *sc = 0;
3046 /* nothing more can be done */
3047 if (!host->interface)
3048 return 1; /* failure */
3050 /* the only address is the primary interface */
3051 if (host->interface->numberOfInterfaces <= 1)
3052 return 1; /* failure */
3054 /* initialise a security object only once */
3056 sc = rxnull_NewClientSecurityObject();
3058 i = host->interface->numberOfInterfaces;
3059 interfaces = calloc(i, sizeof(struct AddrPort));
3060 conns = calloc(i, sizeof(struct rx_connection *));
3061 if (!interfaces || !conns) {
3062 ViceLog(0, ("Failed malloc in MultiProbeAlternateAddress_r\n"));
3066 /* initialize alternate rx connections */
3067 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3068 /* this is the current primary address */
3069 if (host->host == host->interface->interface[i].addr &&
3070 host->port == host->interface->interface[i].port)
3073 interfaces[j] = host->interface->interface[i];
3075 rx_NewConnection(interfaces[j].addr,
3076 interfaces[j].port, 1, sc, 0);
3077 rx_SetConnDeadTime(conns[j], 2);
3078 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3082 assert(j); /* at least one alternate address */
3084 ("Starting multiprobe on all addr for host %x (%s:%d)\n",
3085 host, afs_inet_ntoa_r(host->host, hoststr),
3086 ntohs(host->port)));
3088 multi_Rx(conns, j) {
3089 multi_RXAFSCB_ProbeUuid(&host->interface->uuid);
3093 if (host->callback_rxcon)
3094 rx_DestroyConnection(host->callback_rxcon);
3095 host->callback_rxcon = conns[multi_i];
3096 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
3097 host->host = interfaces[multi_i].addr;
3098 host->port = interfaces[multi_i].port;
3099 h_AddHostToAddrHashTable_r(host->host, host->port, host);
3100 connSuccess = conns[multi_i];
3101 rx_SetConnDeadTime(host->callback_rxcon, 50);
3102 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3104 ("multiprobe success with addr %s:%d\n",
3105 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3106 ntohs(interfaces[multi_i].port)));
3111 ("multiprobe failure with addr %s:%d\n",
3112 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3113 ntohs(interfaces[multi_i].port)));
3115 /* This is less than desirable but its the best we can do.
3116 * The AFS Cache Manager will return either 0 for a Uuid
3117 * match and a 1 for a non-match. If the error is 1 we
3118 * therefore know that our mapping of IP address to Uuid
3119 * is wrong. We should attempt to find the correct
3120 * Uuid and fix the host tables.
3122 if (multi_error == 1) {
3123 /* remove the current alternate address from this host */
3125 removeInterfaceAddr_r(host, interfaces[multi_i].addr, interfaces[multi_i].port);
3129 #ifdef AFS_DEMAND_ATTACH_FS
3130 /* try to bail ASAP if the fileserver is shutting down */
3132 if (fs_state.mode == FS_MODE_SHUTDOWN) {
3141 /* Destroy all connections except the one on which we succeeded */
3142 for (i = 0; i < j; i++)
3143 if (conns[i] != connSuccess)
3144 rx_DestroyConnection(conns[i]);
3150 return 0; /* success */
3152 return 1; /* failure */
3155 #endif /* !defined(INTERPRET_DUMP) */