2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
9 * Portions Copyright (c) 2006 Sine Nomine Associates
13 * NEW callback package callback.c (replaces vicecb.c)
14 * Updated call back routines, NOW with:
16 * Faster DeleteVenus (Now called DeleteAllCallBacks)
17 * Call back breaking for volumes
18 * Adaptive timeouts on call backs
19 * Architected for Multi RPC
20 * No locks (currently implicit vnode locks--these will go, to)
21 * Delayed call back when rpc connection down.
22 * Bulk break of delayed call backs when rpc connection
24 * Strict limit on number of call backs.
26 * InitCallBack(nblocks)
27 * Initialize: nblocks is max number # of file entries + # of callback entries
28 * nblocks must be < 65536
29 * Space used is nblocks*16 bytes
30 * Note that space will be reclaimed by breaking callbacks of old hosts
32 * time = AddCallBack(host, fid)
34 * Returns the expiration time at the workstation.
36 * BreakCallBack(host, fid)
37 * Break all call backs for fid, except for the specified host.
40 * BreakVolumeCallBacks(volume)
41 * Break all call backs on volume, using single call to each host
42 * Delete all the call backs.
44 * DeleteCallBack(host,fid)
45 * Delete (do not break) single call back for fid.
47 * DeleteFileCallBacks(fid)
48 * Delete (do not break) all call backs for fid.
50 * DeleteAllCallBacks(host)
51 * Delete (do not break) all call backs for host.
53 * CleanupTimedOutCallBacks()
54 * Delete all timed out call back entries
55 * Must be called periodically by file server.
57 * BreakDelayedCallBacks(host)
58 * Break all delayed call backs for host.
59 * Returns 1: one or more failed, 0: success.
61 * PrintCallBackStats()
62 * Print statistics about call backs to stdout.
64 * DumpCallBacks() ---wishful thinking---
65 * Dump call back state to /tmp/callback.state.
66 * This is separately interpretable by the program pcb.
68 * Notes: In general, if a call back to a host doesn't get through,
69 * then HostDown, supplied elsewhere, is called. BreakDelayedCallBacks,
70 * however, does not call HostDown, but instead returns an indication of
71 * success if all delayed call backs were finally broken.
73 * BreakDelayedCallBacks MUST be called at the first sign of activity
74 * from the host after HostDown has been called (or a previous
75 * BreakDelayedCallBacks failed). The BreakDelayedCallBacks must be
76 * allowed to complete before any requests from that host are handled.
77 * If BreakDelayedCallBacks fails, then the host should remain
78 * down (and the request should be failed).
80 * CleanupCallBacks MUST be called periodically by the file server for
81 * this package to work correctly. Every 5 minutes is suggested.
84 #include <afsconfig.h>
85 #include <afs/param.h>
91 #include <stdlib.h> /* for malloc() */
92 #include <time.h> /* ANSI standard location for time stuff */
101 #include <afs/assert.h>
103 #include <afs/stds.h>
105 #include <afs/nfs.h> /* yuck. This is an abomination. */
108 #include <afscbint.h>
109 #include <afs/afsutil.h>
111 #include <afs/ihandle.h>
112 #include <afs/vnode.h>
113 #include <afs/volume.h>
114 #include "viced_prototypes.h"
117 #include <afs/ptclient.h> /* need definition of prlist for host.h */
119 #include "callback.h"
120 #ifdef AFS_DEMAND_ATTACH_FS
121 #include "../tviced/serialize_state.h"
122 #endif /* AFS_DEMAND_ATTACH_FS */
125 extern afsUUID FS_HostUUID;
126 extern int hostCount;
127 static int ShowProblems = 1;
129 struct cbcounters cbstuff;
131 static struct FileEntry * FE = NULL; /* don't use FE[0] */
132 static struct CallBack * CB = NULL; /* don't use CB[0] */
134 static struct CallBack * CBfree = NULL;
135 static struct FileEntry * FEfree = NULL;
138 /* Time to live for call backs depends upon number of users of the file.
139 * TimeOuts is indexed by this number/8 (using TimeOut macro). Times
140 * in this table are for the workstation; server timeouts, add
143 static int TimeOuts[] = {
144 /* Note: don't make the first entry larger than 4 hours (see above) */
145 4 * 60 * 60, /* 0-7 users */
146 1 * 60 * 60, /* 8-15 users */
147 30 * 60, /* 16-23 users */
148 15 * 60, /* 24-31 users */
149 15 * 60, /* 32-39 users */
150 10 * 60, /* 40-47 users */
151 10 * 60, /* 48-55 users */
152 10 * 60, /* 56-63 users */
153 }; /* Anything more: MinTimeOut */
155 /* minimum time given for a call back */
156 static int MinTimeOut = (7 * 60);
158 /* Heads of CB queues; a timeout index is 1+index into this array */
159 static afs_uint32 timeout[CB_NUM_TIMEOUT_QUEUES];
161 static afs_int32 tfirst; /* cbtime of oldest unexpired call back time queue */
164 /* 16 byte object get/free routines */
169 /* Prototypes for static routines */
170 static struct FileEntry *FindFE(register AFSFid * fid);
171 static struct CallBack *iGetCB(register int *nused);
172 static int iFreeCB(register struct CallBack *cb, register int *nused);
173 static struct FileEntry *iGetFE(register int *nused);
174 static int iFreeFE(register struct FileEntry *fe, register int *nused);
175 static int TAdd(register struct CallBack *cb, register afs_uint32 * thead);
176 static int TDel(register struct CallBack *cb);
177 static int HAdd(register struct CallBack *cb, register struct host *host);
178 static int HDel(register struct CallBack *cb);
179 static int CDel(struct CallBack *cb, int deletefe);
180 static int CDelPtr(register struct FileEntry *fe, register afs_uint32 * cbp,
182 static afs_uint32 *FindCBPtr(struct FileEntry *fe, struct host *host);
183 static int FDel(register struct FileEntry *fe);
184 static int AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead,
185 int type, int locked);
186 static void MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
187 struct AFSCBFids *afidp, struct host *xhost);
188 static int MultiBreakVolumeCallBack_r(struct host *host, int isheld,
189 struct VCBParams *parms, int deletefe);
190 static int MultiBreakVolumeCallBack(struct host *host, int isheld,
191 struct VCBParams *parms);
192 static int MultiBreakVolumeLaterCallBack(struct host *host, int isheld,
193 struct VCBParams *parms);
194 static int GetSomeSpace_r(struct host *hostp, int locked);
195 static int ClearHostCallbacks_r(struct host *hp, int locked);
197 #define GetCB() ((struct CallBack *)iGetCB(&cbstuff.nCBs))
198 #define GetFE() ((struct FileEntry *)iGetFE(&cbstuff.nFEs))
199 #define FreeCB(cb) iFreeCB((struct CallBack *)cb, &cbstuff.nCBs)
200 #define FreeFE(fe) iFreeFE((struct FileEntry *)fe, &cbstuff.nFEs)
203 /* Other protos - move out sometime */
204 void PrintCB(register struct CallBack *cb, afs_uint32 now);
206 static afs_uint32 HashTable[FEHASH_SIZE]; /* File entry hash table */
208 static struct FileEntry *
209 FindFE(register AFSFid * fid)
213 register struct FileEntry *fe;
215 hash = FEHash(fid->Volume, fid->Unique);
216 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
218 if (fe->volid == fid->Volume && fe->unique == fid->Unique
219 && fe->vnode == fid->Vnode && (fe->status & FE_LATER) != FE_LATER)
225 #ifndef INTERPRET_DUMP
227 static struct CallBack *
228 iGetCB(register int *nused)
230 register struct CallBack *ret;
232 if ((ret = CBfree)) {
233 CBfree = (struct CallBack *)(((struct object *)ret)->next);
240 iFreeCB(register struct CallBack *cb, register int *nused)
242 ((struct object *)cb)->next = (struct object *)CBfree;
248 static struct FileEntry *
249 iGetFE(register int *nused)
251 register struct FileEntry *ret;
253 if ((ret = FEfree)) {
254 FEfree = (struct FileEntry *)(((struct object *)ret)->next);
261 iFreeFE(register struct FileEntry *fe, register int *nused)
263 ((struct object *)fe)->next = (struct object *)FEfree;
269 /* Add cb to end of specified timeout list */
271 TAdd(register struct CallBack *cb, register afs_uint32 * thead)
274 (*thead) = cb->tnext = cb->tprev = cbtoi(cb);
276 register struct CallBack *thp = itocb(*thead);
278 cb->tprev = thp->tprev;
282 thp->tprev = (itocb(thp->tprev)->tnext = cbtoi(cb));
284 thp->tprev = cbtoi(cb);
287 cb->thead = ttoi(thead);
291 /* Delete call back entry from timeout list */
293 TDel(register struct CallBack *cb)
295 register afs_uint32 *thead = itot(cb->thead);
297 if (*thead == cbtoi(cb))
298 *thead = (*thead == cb->tnext ? 0 : cb->tnext);
299 if (itocb(cb->tprev))
300 itocb(cb->tprev)->tnext = cb->tnext;
301 if (itocb(cb->tnext))
302 itocb(cb->tnext)->tprev = cb->tprev;
306 /* Add cb to end of specified host list */
308 HAdd(register struct CallBack *cb, register struct host *host)
310 cb->hhead = h_htoi(host);
312 host->cblist = cb->hnext = cb->hprev = cbtoi(cb);
314 register struct CallBack *fcb = itocb(host->cblist);
316 cb->hprev = fcb->hprev;
317 cb->hnext = cbtoi(fcb);
318 fcb->hprev = (itocb(fcb->hprev)->hnext = cbtoi(cb));
323 /* Delete call back entry from host list */
325 HDel(register struct CallBack *cb)
327 register afs_uint32 *hhead = &h_itoh(cb->hhead)->cblist;
329 if (*hhead == cbtoi(cb))
330 *hhead = (*hhead == cb->hnext ? 0 : cb->hnext);
331 itocb(cb->hprev)->hnext = cb->hnext;
332 itocb(cb->hnext)->hprev = cb->hprev;
336 /* Delete call back entry from fid's chain of cb's */
337 /* N.B. This one also deletes the CB, and also possibly parent FE, so
338 * make sure that it is not on any other list before calling this
341 CDel(struct CallBack *cb, int deletefe)
344 struct FileEntry *fe = itofe(cb->fhead);
345 register afs_uint32 *cbp;
348 for (safety = 0, cbp = &fe->firstcb; *cbp && *cbp != cbi;
349 cbp = &itocb(*cbp)->cnext, safety++) {
350 if (safety > cbstuff.nblks + 10) {
353 ("CDel: Internal Error -- shutting down: wanted %d from %d, now at %d\n",
354 cbi, fe->firstcb, *cbp));
356 ShutDownAndCore(PANIC);
359 CDelPtr(fe, cbp, deletefe);
363 /* Same as CDel, but pointer to parent pointer to CB entry is passed,
364 * as well as file entry */
365 /* N.B. This one also deletes the CB, and also possibly parent FE, so
366 * make sure that it is not on any other list before calling this
368 static int Ccdelpt = 0, CcdelB = 0;
371 CDelPtr(register struct FileEntry *fe, register afs_uint32 * cbp,
374 register struct CallBack *cb;
384 if ((--fe->ncbs == 0) && deletefe)
390 FindCBPtr(struct FileEntry *fe, struct host *host)
392 register afs_uint32 hostindex = h_htoi(host);
393 register struct CallBack *cb;
394 register afs_uint32 *cbp;
397 for (safety = 0, cbp = &fe->firstcb; *cbp; cbp = &cb->cnext, safety++) {
398 if (safety > cbstuff.nblks) {
399 ViceLog(0, ("FindCBPtr: Internal Error -- shutting down.\n"));
401 ShutDownAndCore(PANIC);
404 if (cb->hhead == hostindex)
410 /* Delete file entry from hash table */
412 FDel(register struct FileEntry *fe)
414 register int fei = fetoi(fe);
415 register afs_uint32 *p = &HashTable[FEHash(fe->volid, fe->unique)];
417 while (*p && *p != fei)
418 p = &itofe(*p)->fnext;
425 /* initialize the callback package */
427 InitCallBack(int nblks)
430 tfirst = CBtime(FT_ApproxTime());
431 /* N.B. The "-1", below, is because
432 * FE[0] and CB[0] are not used--and not allocated */
433 FE = ((struct FileEntry *)(calloc(nblks, sizeof(struct FileEntry))));
435 ViceLog(0, ("Failed malloc in InitCallBack\n"));
438 FE--; /* FE[0] is supposed to point to junk */
439 cbstuff.nFEs = nblks;
441 FreeFE(&FE[cbstuff.nFEs]); /* This is correct */
442 CB = ((struct CallBack *)(calloc(nblks, sizeof(struct CallBack))));
444 ViceLog(0, ("Failed malloc in InitCallBack\n"));
447 CB--; /* CB[0] is supposed to point to junk */
448 cbstuff.nCBs = nblks;
450 FreeCB(&CB[cbstuff.nCBs]); /* This is correct */
451 cbstuff.nblks = nblks;
452 cbstuff.nbreakers = 0;
458 XCallBackBulk_r(struct host * ahost, struct AFSFid * fids, afs_int32 nfids)
460 struct AFSCallBack tcbs[AFSCBMAX];
466 struct rx_connection *cb_conn = NULL;
469 rx_SetConnDeadTime(ahost->callback_rxcon, 4);
470 rx_SetConnHardDeadTime(ahost->callback_rxcon, AFS_HARDDEADTIME);
477 for (i = 0; i < nfids && i < AFSCBMAX; i++) {
478 tcbs[i].CallBackVersion = CALLBACK_VERSION;
479 tcbs[i].ExpirationTime = 0;
480 tcbs[i].CallBackType = CB_DROPPED;
482 tf.AFSCBFids_len = i;
483 tf.AFSCBFids_val = &(fids[j]);
487 tc.AFSCBs_val = tcbs;
489 cb_conn = ahost->callback_rxcon;
490 rx_GetConnection(cb_conn);
492 code |= RXAFSCB_CallBack(cb_conn, &tf, &tc);
493 rx_PutConnection(cb_conn);
501 /* the locked flag tells us if the host entry has already been locked
502 * by our parent. I don't think anybody actually calls us with the
503 * host locked, but here's how to make that work: GetSomeSpace has to
504 * change so that it doesn't attempt to lock any hosts < "host". That
505 * means that it might be unable to free any objects, so it has to
506 * return an exit status. If it fails, then AddCallBack1 might fail,
507 * as well. If so, the host->ResetDone should probably be set to 0,
508 * and we probably don't want to return a callback promise to the
509 * cache manager, either. */
511 AddCallBack1(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
519 retVal = AddCallBack1_r(host, fid, thead, type, 1);
529 AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
532 struct FileEntry *fe;
533 struct CallBack *cb = 0, *lastcb = 0;
534 struct FileEntry *newfe = 0;
536 afs_uint32 *Thead = thead;
537 struct CallBack *newcb = 0;
540 cbstuff.AddCallBacks++;
544 /* allocate these guys first, since we can't call the allocator with
545 * the host structure locked -- or we might deadlock. However, we have
546 * to avoid races with FindFE... */
547 while (!(newcb = GetCB())) {
548 GetSomeSpace_r(host, locked);
550 while (!(newfe = GetFE())) { /* Get it now, so we don't have to call */
551 /* GetSomeSpace with the host locked, later. This might turn out to */
552 /* have been unneccessary, but that's actually kind of unlikely, since */
553 /* most files are not shared. */
554 GetSomeSpace_r(host, locked);
558 h_Lock_r(host); /* this can yield, so do it before we get any */
563 if (type == CB_NORMAL) {
565 TimeCeiling(FT_ApproxTime() + TimeOut(fe ? fe->ncbs : 0) +
567 Thead = THead(CBtime(time_out));
568 } else if (type == CB_VOLUME) {
569 time_out = TimeCeiling((60 * 120 + FT_ApproxTime()) + 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(FT_ApproxTime() + ServerBias +
577 TimeOut(22 + (fe ? fe->ncbs : 0)));
578 Thead = THead(CBtime(time_out));
584 register afs_uint32 hash;
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"));
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);
634 /* now free any still-unused callback or host entries */
640 if (!locked) /* freecb and freefe might(?) yield */
643 if (type == CB_NORMAL || type == CB_VOLUME || type == CB_BULK)
644 return time_out - ServerBias; /* Expires sooner at workstation */
650 CompareCBA(const void *e1, const void *e2)
652 const struct cbstruct *cba1 = (const struct cbstruct *)e1;
653 const struct cbstruct *cba2 = (const struct cbstruct *)e2;
654 return ((cba1->hp)->index - (cba2->hp)->index);
657 /* Take an array full of hosts, all held. Break callbacks to them, and
658 * release the holds once you're done, except don't release xhost. xhost
659 * may be NULL. Currently only works for a single Fid in afidp array.
660 * If you want to make this work with multiple fids, you need to fix
661 * the error handling. One approach would be to force a reset if a
662 * multi-fid call fails, or you could add delayed callbacks for each
663 * fid. You probably also need to sort and remove duplicate hosts.
664 * When this is called from the BreakVolumeCallBacks path, it does NOT
665 * force a reset if the RPC fails, it just marks the host down and tries
666 * to create a delayed callback. */
667 /* N.B. be sure that code works when ncbas == 0 */
668 /* N.B. requires all the cba[*].hp pointers to be valid... */
669 /* This routine does not hold a lock on the host for the duration of
670 * the BreakCallBack RPC, which is a significant deviation from tradition.
671 * It _does_ get a lock on the host before setting VenusDown = 1,
672 * which is sufficient only if VenusDown = 0 only happens when the
673 * lock is held over the RPC and the subsequent VenusDown == 0
674 * wherever that is done. */
676 MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
677 struct AFSCBFids *afidp, struct host *xhost)
680 struct rx_connection *conns[MAX_CB_HOSTS];
681 static struct AFSCBs tc = { 0, 0 };
682 int multi_to_cba_map[MAX_CB_HOSTS];
684 assert(ncbas <= MAX_CB_HOSTS);
686 /* sort cba list to avoid makecall issues */
687 qsort(cba, ncbas, sizeof(struct cbstruct), CompareCBA);
689 /* set up conns for multi-call */
690 for (i = 0, j = 0; i < ncbas; i++) {
691 struct host *thishost = cba[i].hp;
692 if (!thishost || (thishost->hostFlags & HOSTDELETED)) {
695 rx_GetConnection(thishost->callback_rxcon);
696 multi_to_cba_map[j] = i;
697 conns[j++] = thishost->callback_rxcon;
700 rx_SetConnDeadTime(thishost->callback_rxcon, 4);
701 rx_SetConnHardDeadTime(thishost->callback_rxcon, AFS_HARDDEADTIME);
705 if (j) { /* who knows what multi would do with 0 conns? */
709 multi_RXAFSCB_CallBack(afidp, &tc);
715 i = multi_to_cba_map[multi_i];
721 ("BCB: INTERNAL ERROR: hp=%x, cba=%x, thead=%u\n",
725 ** try breaking callbacks on alternate interface addresses
727 if (MultiBreakCallBackAlternateAddress(hp, afidp)) {
730 ("BCB: Failed on file %u.%u.%u, Host %x (%s:%d) is down\n",
731 afidp->AFSCBFids_val->Volume,
732 afidp->AFSCBFids_val->Vnode,
733 afidp->AFSCBFids_val->Unique,
735 afs_inet_ntoa_r(hp->host, hoststr),
741 hp->hostFlags |= VENUSDOWN;
743 * We always go into AddCallBack1_r with the host locked
745 AddCallBack1_r(hp, afidp->AFSCBFids_val, itot(idx),
758 for (i = 0; i < ncbas; i++) {
761 if (hp && xhost != hp) {
766 /* H_UNLOCK around this so h_FreeConnection does not deadlock.
767 h_FreeConnection should *never* be called on a callback connection,
768 but on 10/27/04 a deadlock occurred where it was, when we know why,
769 this should be reverted. -- shadow */
771 for (i = 0; i < j; i++) {
772 rx_PutConnection(conns[i]);
780 * Break all call backs for fid, except for the specified host (unless flag
781 * is true, in which case all get a callback message. Assumption: the specified
782 * host is h_Held, by the caller; the others aren't.
783 * Specified host may be bogus, that's ok. This used to check to see if the
784 * host was down in two places, once right after the host was h_held, and
785 * again after it was locked. That race condition is incredibly rare and
786 * relatively harmless even when it does occur, so we don't check for it now.
788 /* if flag is true, send a break callback msg to "host", too */
790 BreakCallBack(struct host *xhost, AFSFid * fid, int flag)
792 struct FileEntry *fe;
793 struct CallBack *cb, *nextcb;
794 struct cbstruct cba[MAX_CB_HOSTS];
801 ("BCB: BreakCallBack(Host %x all but %s:%d, (%u,%u,%u))\n",
802 xhost, afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
803 fid->Volume, fid->Vnode, fid->Unique));
806 cbstuff.BreakCallBacks++;
811 hostindex = h_htoi(xhost);
812 cb = itocb(fe->firstcb);
813 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
814 /* the most common case is what follows the || */
817 tf.AFSCBFids_len = 1;
818 tf.AFSCBFids_val = fid;
821 for (ncbas = 0; cb && ncbas < MAX_CB_HOSTS; cb = nextcb) {
822 nextcb = itocb(cb->cnext);
823 if ((cb->hhead != hostindex || flag)
824 && (cb->status == CB_BULK || cb->status == CB_NORMAL
825 || cb->status == CB_VOLUME)) {
826 struct host *thishost = h_itoh(cb->hhead);
828 ViceLog(0, ("BCB: BOGUS! cb->hhead is NULL!\n"));
829 } else if (thishost->hostFlags & VENUSDOWN) {
831 ("BCB: %x (%s:%d) is down; delaying break call back\n",
832 thishost, afs_inet_ntoa_r(thishost->host, hoststr),
833 ntohs(thishost->port)));
834 cb->status = CB_DELAYED;
837 cba[ncbas].hp = thishost;
838 cba[ncbas].thead = cb->thead;
842 CDel(cb, 1); /* Usually first; so this delete
843 * is reasonably inexpensive */
849 MultiBreakCallBack_r(cba, ncbas, &tf, xhost);
851 /* we need to to all these initializations again because MultiBreakCallBack may block */
856 cb = itocb(fe->firstcb);
857 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
858 /* the most common case is what follows the || */
869 /* Delete (do not break) single call back for fid */
871 DeleteCallBack(struct host *host, AFSFid * fid)
873 register struct FileEntry *fe;
874 register afs_uint32 *pcb;
878 cbstuff.DeleteCallBacks++;
886 ("DCB: No call backs for fid (%u, %u, %u)\n", fid->Volume,
887 fid->Vnode, fid->Unique));
890 pcb = FindCBPtr(fe, host);
893 ("DCB: No call back for host %x (%s:%d), (%u, %u, %u)\n",
894 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
895 fid->Volume, fid->Vnode, fid->Unique));
909 * Delete (do not break) all call backs for fid. This call doesn't
910 * set all of the various host locks, but it shouldn't really matter
911 * since we're not adding callbacks, but deleting them. I'm not sure
912 * why it doesn't set the lock, however; perhaps it should.
915 DeleteFileCallBacks(AFSFid * fid)
917 register struct FileEntry *fe;
918 register struct CallBack *cb;
919 register afs_uint32 cbi;
923 cbstuff.DeleteFiles++;
928 ("DF: No fid (%u,%u,%u) to delete\n", fid->Volume, fid->Vnode,
932 for (n = 0, cbi = fe->firstcb; cbi; n++) {
945 /* Delete (do not break) all call backs for host. The host should be
948 DeleteAllCallBacks_r(struct host *host, int deletefe)
950 register struct CallBack *cb;
951 register int cbi, first;
953 cbstuff.DeleteAllCallBacks++;
954 cbi = first = host->cblist;
956 ViceLog(8, ("DV: no call backs\n"));
964 } while (cbi != first);
970 * Break all delayed call backs for host. Returns 1 if all call backs
971 * successfully broken; 0 otherwise. Assumes host is h_Held and h_Locked.
972 * Must be called with VenusDown set for this host
975 BreakDelayedCallBacks(struct host *host)
979 retVal = BreakDelayedCallBacks_r(host);
985 BreakDelayedCallBacks_r(struct host *host)
987 struct AFSFid fids[AFSCBMAX];
988 u_byte thead[AFSCBMAX]; /* This should match thead in struct Callback */
989 int cbi, first, nfids;
993 struct rx_connection *cb_conn;
996 if (!(host->hostFlags & RESETDONE) && !(host->hostFlags & HOSTDELETED)) {
997 host->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
998 cb_conn = host->callback_rxcon;
999 rx_GetConnection(cb_conn);
1000 if (host->interface) {
1003 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1006 code = RXAFSCB_InitCallBackState(cb_conn);
1008 rx_PutConnection(cb_conn);
1011 host->hostFlags |= ALTADDR; /* alternate addresses are valid */
1015 ("CB: Call back connect back failed (in break delayed) for Host %x (%s:%d)\n",
1016 host, afs_inet_ntoa_r(host->host, hoststr),
1017 ntohs(host->port)));
1019 host->hostFlags |= VENUSDOWN;
1022 ("InitCallBackState success on %x (%s:%d)\n",
1023 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
1024 /* reset was done successfully */
1025 host->hostFlags |= RESETDONE;
1026 host->hostFlags &= ~VENUSDOWN;
1029 while (!(host->hostFlags & HOSTDELETED)) {
1031 host->hostFlags &= ~VENUSDOWN; /* presume up */
1032 cbi = first = host->cblist;
1036 first = host->cblist;
1039 if (cb->status == CB_DELAYED) {
1040 register struct FileEntry *fe = itofe(cb->fhead);
1041 thead[nfids] = cb->thead;
1042 fids[nfids].Volume = fe->volid;
1043 fids[nfids].Vnode = fe->vnode;
1044 fids[nfids].Unique = fe->unique;
1050 } while (cbi && cbi != first && nfids < AFSCBMAX);
1056 if (XCallBackBulk_r(host, fids, nfids)) {
1057 /* Failed, again: put them back, probably with old
1062 ("CB: XCallBackBulk failed, Host %x (%s:%d); callback list follows:\n",
1063 host, afs_inet_ntoa_r(host->host, hoststr),
1064 ntohs(host->port)));
1066 for (i = 0; i < nfids; i++) {
1069 ("CB: Host %x (%s:%d), file %u.%u.%u (part of bulk callback)\n",
1070 host, afs_inet_ntoa_r(host->host, hoststr),
1071 ntohs(host->port), fids[i].Volume,
1072 fids[i].Vnode, fids[i].Unique));
1075 * AddCallBack1_r(host, &fids[i], itot(thead[i]), CB_DELAYED, 1);
1076 * * but it turns out to cause too many tricky locking problems.
1077 * * now, if break delayed fails, screw it. */
1079 host->hostFlags |= VENUSDOWN; /* Failed */
1080 ClearHostCallbacks_r(host, 1 /* locked */ );
1084 if (nfids < AFSCBMAX)
1088 cbstuff.nbreakers--;
1089 /* If we succeeded it's always ok to unset HFE_LATER */
1090 if (!(host->hostFlags & VENUSDOWN))
1091 host->hostFlags &= ~HFE_LATER;
1092 return (host->hostFlags & VENUSDOWN);
1096 ** isheld is 0 if the host is held in h_Enumerate
1097 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1100 MultiBreakVolumeCallBack_r(struct host *host, int isheld,
1101 struct VCBParams *parms, int deletefe)
1106 return isheld; /* host is held only by h_Enumerate, do nothing */
1107 if (host->hostFlags & HOSTDELETED)
1108 return 0; /* host is deleted, release hold */
1110 if (host->hostFlags & VENUSDOWN) {
1112 if (host->hostFlags & HOSTDELETED) {
1114 return 0; /* Release hold */
1117 ("BVCB: volume call back for Host %x (%s:%d) failed\n",
1118 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
1121 ("CB: volume callback for Host %x (%s:%d) failed\n",
1122 host, afs_inet_ntoa_r(host->host, hoststr),
1123 ntohs(host->port)));
1125 DeleteAllCallBacks_r(host, deletefe); /* Delete all callback state
1126 * rather than attempting to
1127 * selectively remember to
1128 * delete the volume callbacks
1130 host->hostFlags &= ~RESETDONE; /* Do InitCallBackState when host returns */
1132 return 0; /* release hold */
1134 assert(parms->ncbas <= MAX_CB_HOSTS);
1136 /* Do not call MultiBreakCallBack on the current host structure
1137 ** because it would prematurely release the hold on the host
1139 if (parms->ncbas == MAX_CB_HOSTS) {
1140 struct AFSCBFids tf;
1142 tf.AFSCBFids_len = 1;
1143 tf.AFSCBFids_val = parms->fid;
1145 /* this releases all the hosts */
1146 MultiBreakCallBack_r(parms->cba, parms->ncbas, &tf, 0 /* xhost */ );
1150 parms->cba[parms->ncbas].hp = host;
1151 parms->cba[(parms->ncbas)++].thead = parms->thead;
1152 return 1; /* DON'T release hold, because we still need it. */
1156 ** isheld is 0 if the host is held in h_Enumerate
1157 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1160 MultiBreakVolumeCallBack(struct host *host, int isheld,
1161 struct VCBParams *parms)
1165 retval = MultiBreakVolumeCallBack_r(host, isheld, parms, 1);
1171 ** isheld is 0 if the host is held in h_Enumerate
1172 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1175 MultiBreakVolumeLaterCallBack(struct host *host, int isheld,
1176 struct VCBParams *parms)
1180 retval = MultiBreakVolumeCallBack_r(host, isheld, parms, 0);
1186 * Break all call backs on a single volume. Don't call this with any
1187 * hosts h_held. Note that this routine clears the callbacks before
1188 * actually breaking them, and that the vnode isn't locked during this
1189 * operation, so that people might see temporary callback loss while
1190 * this function is executing. It is just a temporary state, however,
1191 * since the callback will be broken later by this same function.
1193 * Now uses multi-RX for CallBack RPC. Note that the
1194 * multiBreakCallBacks routine does not force a reset if the RPC
1195 * fails, unlike the previous version of this routine, but does create
1196 * a delayed callback. Resets will be forced if the host is
1197 * determined to be down before the RPC is executed.
1200 BreakVolumeCallBacks(afs_uint32 volume)
1205 struct CallBack *cb;
1206 struct FileEntry *fe;
1208 struct VCBParams henumParms;
1209 afs_uint32 tthead = 0; /* zero is illegal value */
1212 fid.Volume = volume, fid.Vnode = fid.Unique = 0;
1213 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1214 for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
1215 if (fe->volid == volume) {
1216 register struct CallBack *cbnext;
1217 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1218 host = h_itoh(cb->hhead);
1220 cbnext = itocb(cb->cnext);
1221 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1227 /* leave hold for MultiBreakVolumeCallBack to clear */
1238 /* didn't find any callbacks, so return right away. */
1242 henumParms.ncbas = 0;
1243 henumParms.fid = &fid;
1244 henumParms.thead = tthead;
1246 h_Enumerate(MultiBreakVolumeCallBack, (char *)&henumParms);
1248 if (henumParms.ncbas) { /* do left-overs */
1249 struct AFSCBFids tf;
1250 tf.AFSCBFids_len = 1;
1251 tf.AFSCBFids_val = &fid;
1253 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf, 0);
1255 henumParms.ncbas = 0;
1261 #ifdef AFS_PTHREAD_ENV
1262 extern pthread_cond_t fsync_cond;
1264 extern char fsync_wait[];
1268 BreakVolumeCallBacksLater(afs_uint32 volume)
1272 struct FileEntry *fe;
1273 struct CallBack *cb;
1277 ViceLog(25, ("Setting later on volume %u\n", volume));
1279 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1280 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1281 if (fe->volid == volume) {
1282 register struct CallBack *cbnext;
1283 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1284 host = h_itoh(cb->hhead);
1285 host->hostFlags |= HFE_LATER;
1286 cb->status = CB_DELAYED;
1287 cbnext = itocb(cb->cnext);
1290 fe->status |= FE_LATER;
1299 /* didn't find any callbacks, so return right away. */
1303 ViceLog(25, ("Fsync thread wakeup\n"));
1304 #ifdef AFS_PTHREAD_ENV
1306 assert(pthread_cond_broadcast(&fsync_cond) == 0);
1309 LWP_NoYieldSignal(fsync_wait);
1315 BreakLaterCallBacks(void)
1320 struct CallBack *cb;
1321 struct FileEntry *fe = NULL;
1322 struct FileEntry *myfe = NULL;
1324 struct VCBParams henumParms;
1325 unsigned short tthead = 0; /* zero is illegal value */
1329 ViceLog(25, ("Looking for FileEntries to unchain\n"));
1332 /* Pick the first volume we see to clean up */
1333 fid.Volume = fid.Vnode = fid.Unique = 0;
1335 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1336 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1337 if (fe && (fe->status & FE_LATER)
1338 && (fid.Volume == 0 || fid.Volume == fe->volid)) {
1339 /* Ugly, but used to avoid left side casting */
1340 struct object *tmpfe;
1342 ("Unchaining for %u:%u:%u\n", fe->vnode, fe->unique,
1344 fid.Volume = fe->volid;
1346 /* Works since volid is deeper than the largest pointer */
1347 tmpfe = (struct object *)fe;
1348 tmpfe->next = (struct object *)myfe;
1361 /* loop over FEs from myfe and free/break */
1363 for (fe = myfe; fe;) {
1364 register struct CallBack *cbnext;
1365 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1366 cbnext = itocb(cb->cnext);
1367 host = h_itoh(cb->hhead);
1368 if (cb->status == CB_DELAYED) {
1370 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1375 CDel(cb, 0); /* Don't let CDel clean up the fe */
1376 /* leave hold for MultiBreakVolumeCallBack to clear */
1379 ("Found host %x (%s:%d) non-DELAYED cb for %u:%u:%u\n",
1380 host, afs_inet_ntoa_r(host->host, hoststr),
1381 ntohs(host->port), fe->vnode, fe->unique, fe->volid));
1385 fe = (struct FileEntry *)((struct object *)fe)->next;
1390 ViceLog(125, ("Breaking volume %u\n", fid.Volume));
1391 henumParms.ncbas = 0;
1392 henumParms.fid = &fid;
1393 henumParms.thead = tthead;
1395 h_Enumerate(MultiBreakVolumeLaterCallBack, (char *)&henumParms);
1397 if (henumParms.ncbas) { /* do left-overs */
1398 struct AFSCBFids tf;
1399 tf.AFSCBFids_len = 1;
1400 tf.AFSCBFids_val = &fid;
1402 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf, 0);
1403 henumParms.ncbas = 0;
1408 /* Arrange to be called again */
1413 * Delete all timed-out call back entries (to be called periodically by file
1417 CleanupTimedOutCallBacks(void)
1420 CleanupTimedOutCallBacks_r();
1426 CleanupTimedOutCallBacks_r(void)
1428 afs_uint32 now = CBtime(FT_ApproxTime());
1429 register afs_uint32 *thead;
1430 register struct CallBack *cb;
1431 register int ntimedout = 0;
1434 while (tfirst <= now) {
1436 cbi = *(thead = THead(tfirst));
1442 ("CCB: deleting timed out call back %x (%s:%d), (%u,%u,%u)\n",
1443 h_itoh(cb->hhead)->host,
1444 afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
1445 h_itoh(cb->hhead)->port, itofe(cb->fhead)->volid,
1446 itofe(cb->fhead)->vnode, itofe(cb->fhead)->unique));
1450 if (ntimedout > cbstuff.nblks) {
1451 ViceLog(0, ("CCB: Internal Error -- shutting down...\n"));
1452 DumpCallBackState();
1453 ShutDownAndCore(PANIC);
1455 } while (cbi != *thead);
1460 cbstuff.CBsTimedOut += ntimedout;
1461 ViceLog(7, ("CCB: deleted %d timed out callbacks\n", ntimedout));
1462 return (ntimedout > 0);
1465 static struct host *lih_host;
1466 static int lih_host_held;
1468 /* This version does not allow 'host' to be selected unless its ActiveCall
1469 * is newer than 'hostp' which is the host with the oldest ActiveCall from
1470 * the last pass (if it is provided). We filter out any hosts that are
1471 * are held by other threads.
1474 lih0_r(register struct host *host, register int held,
1475 register struct host *hostp)
1478 && (hostp && host != hostp)
1479 && (!held && !h_OtherHolds_r(host))
1480 && (!lih_host || host->ActiveCall < lih_host->ActiveCall)
1481 && (!hostp || host->ActiveCall > hostp->ActiveCall)) {
1482 if (lih_host != NULL && lih_host_held) {
1483 h_Release_r(lih_host);
1486 lih_host_held = !held;
1492 /* This version does not allow 'host' to be selected unless its ActiveCall
1493 * is newer than 'hostp' which is the host with the oldest ActiveCall from
1494 * the last pass (if it is provided). In this second varient, we do not
1495 * prevent held hosts from being selected.
1498 lih1_r(register struct host *host, register int held,
1499 register struct host *hostp)
1502 && (hostp && host != hostp)
1503 && (!lih_host || host->ActiveCall < lih_host->ActiveCall)
1504 && (!hostp || host->ActiveCall > hostp->ActiveCall)) {
1505 if (lih_host != NULL && lih_host_held) {
1506 h_Release_r(lih_host);
1509 lih_host_held = !held;
1515 /* This could be upgraded to get more space each time */
1516 /* first pass: sequentially find the oldest host which isn't held by
1517 anyone for which we can clear callbacks;
1519 /* second pass: sequentially find the oldest host regardless of
1520 whether or not the host is held; skipping 'hostp' */
1521 /* third pass: attempt to clear callbacks from 'hostp' */
1522 /* always called with hostp unlocked */
1524 /* Note: hostlist is ordered most recently created host first and
1525 * its order has no relationship to the most recently used. */
1526 extern struct host *hostList;
1528 GetSomeSpace_r(struct host *hostp, int locked)
1530 register struct host *hp, *hp1, *hp2;
1533 cbstuff.GotSomeSpaces++;
1535 ("GSS: First looking for timed out call backs via CleanupCallBacks\n"));
1536 if (CleanupTimedOutCallBacks_r()) {
1546 h_Enumerate_r(i == 0 ? lih0_r : lih1_r, hp2, (char *)hp1);
1549 /* set in lih_r! private copy before giving up H_LOCK */
1550 int lih_host_held2=lih_host_held;
1552 if ((hp != hostp) && !ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
1563 * Next time try getting callbacks from any host even if
1564 * it's deleted (that's actually great since we can freely
1565 * remove its callbacks) or it's held since the only other
1566 * option is starvation for the file server (i.e. until the
1567 * callback timeout arrives).
1574 ("GSS: Try harder for longest inactive host cnt= %d\n",
1579 /* Could not obtain space from other hosts, clear hostp's callback state */
1584 ClearHostCallbacks_r(hostp, 1 /*already locked */ );
1591 /* locked - set if caller has already locked the host */
1593 ClearHostCallbacks_r(struct host *hp, int locked)
1598 struct rx_connection *cb_conn = NULL;
1601 ("GSS: Delete longest inactive host %x (%s:%d)\n",
1602 hp, afs_inet_ntoa_r(hp->host, hoststr), ntohs(hp->port)));
1603 if (!(held = h_Held_r(hp)))
1606 /** Try a non-blocking lock. If the lock is already held return
1607 * after releasing hold on hp
1610 if (h_NBLock_r(hp)) {
1616 if (hp->Console & 2) {
1618 * If the special console field is set it means that a thread
1619 * is waiting in AddCallBack1 after it set pointers to the
1620 * file entry and/or callback entry. Because of the bogus
1621 * usage of h_hold it won't prevent from another thread, this
1622 * one, to remove all the callbacks so just to be safe we keep
1623 * a reference. NOTE, on the last phase we'll free the calling
1624 * host's callbacks but that's ok...
1628 DeleteAllCallBacks_r(hp, 1);
1629 if (hp->hostFlags & VENUSDOWN) {
1630 hp->hostFlags &= ~RESETDONE; /* remember that we must do a reset */
1632 /* host is up, try a call */
1633 hp->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1634 cb_conn = hp->callback_rxcon;
1635 rx_GetConnection(hp->callback_rxcon);
1636 if (hp->interface) {
1639 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1642 code = RXAFSCB_InitCallBackState(cb_conn);
1644 rx_PutConnection(cb_conn);
1647 hp->hostFlags |= ALTADDR; /* alternate addresses are valid */
1649 /* failed, mark host down and need reset */
1650 hp->hostFlags |= VENUSDOWN;
1651 hp->hostFlags &= ~RESETDONE;
1653 /* reset succeeded, we're done */
1654 hp->hostFlags |= RESETDONE;
1665 #endif /* INTERPRET_DUMP */
1669 PrintCallBackStats(void)
1672 "%d add CB, %d break CB, %d del CB, %d del FE, %d CB's timed out, %d space reclaim, %d del host\n",
1673 cbstuff.AddCallBacks, cbstuff.BreakCallBacks,
1674 cbstuff.DeleteCallBacks, cbstuff.DeleteFiles, cbstuff.CBsTimedOut,
1675 cbstuff.GotSomeSpaces, cbstuff.DeleteAllCallBacks);
1676 fprintf(stderr, "%d CBs, %d FEs, (%d of total of %d 16-byte blocks)\n",
1677 cbstuff.nCBs, cbstuff.nFEs, cbstuff.nCBs + cbstuff.nFEs,
1683 #define MAGIC 0x12345678 /* To check byte ordering of dump when it is read in */
1685 #ifndef INTERPRET_DUMP
1687 #ifdef AFS_DEMAND_ATTACH_FS
1690 * callback state serialization
1692 static int cb_stateSaveTimeouts(struct fs_dump_state * state);
1693 static int cb_stateSaveFEHash(struct fs_dump_state * state);
1694 static int cb_stateSaveFEs(struct fs_dump_state * state);
1695 static int cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe);
1696 static int cb_stateRestoreTimeouts(struct fs_dump_state * state);
1697 static int cb_stateRestoreFEHash(struct fs_dump_state * state);
1698 static int cb_stateRestoreFEs(struct fs_dump_state * state);
1699 static int cb_stateRestoreFE(struct fs_dump_state * state);
1700 static int cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
1701 struct iovec * iov, int niovecs);
1703 static int cb_stateVerifyFEHash(struct fs_dump_state * state);
1704 static int cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe);
1705 static int cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe);
1706 static int cb_stateVerifyTimeoutQueues(struct fs_dump_state * state);
1708 static int cb_stateFEToDiskEntry(struct FileEntry *, struct FEDiskEntry *);
1709 static int cb_stateDiskEntryToFE(struct fs_dump_state * state,
1710 struct FEDiskEntry *, struct FileEntry *);
1712 static int cb_stateCBToDiskEntry(struct CallBack *, struct CBDiskEntry *);
1713 static int cb_stateDiskEntryToCB(struct fs_dump_state * state,
1714 struct CBDiskEntry *, struct CallBack *);
1716 static int cb_stateFillHeader(struct callback_state_header * hdr);
1717 static int cb_stateCheckHeader(struct callback_state_header * hdr);
1719 static int cb_stateAllocMap(struct fs_dump_state * state);
1722 cb_stateSave(struct fs_dump_state * state)
1726 AssignInt64(state->eof_offset, &state->hdr->cb_offset);
1728 /* invalidate callback state header */
1729 memset(state->cb_hdr, 0, sizeof(struct callback_state_header));
1730 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1731 sizeof(struct callback_state_header))) {
1736 fs_stateIncEOF(state, sizeof(struct callback_state_header));
1738 /* dump timeout state */
1739 if (cb_stateSaveTimeouts(state)) {
1744 /* dump fe hashtable state */
1745 if (cb_stateSaveFEHash(state)) {
1750 /* dump callback state */
1751 if (cb_stateSaveFEs(state)) {
1756 /* write the callback state header to disk */
1757 cb_stateFillHeader(state->cb_hdr);
1758 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1759 sizeof(struct callback_state_header))) {
1769 cb_stateRestore(struct fs_dump_state * state)
1773 if (fs_stateReadHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1774 sizeof(struct callback_state_header))) {
1779 if (cb_stateCheckHeader(state->cb_hdr)) {
1784 if (cb_stateAllocMap(state)) {
1789 if (cb_stateRestoreTimeouts(state)) {
1794 if (cb_stateRestoreFEHash(state)) {
1799 /* restore FEs and CBs from disk */
1800 if (cb_stateRestoreFEs(state)) {
1805 /* restore the timeout queue heads */
1806 tfirst = state->cb_hdr->tfirst;
1813 cb_stateRestoreIndices(struct fs_dump_state * state)
1816 struct FileEntry * fe;
1817 struct CallBack * cb;
1819 /* restore indices in the FileEntry structures */
1820 for (i = 1; i < state->fe_map.len; i++) {
1821 if (state->fe_map.entries[i].new_idx) {
1822 fe = itofe(state->fe_map.entries[i].new_idx);
1824 /* restore the fe->fnext entry */
1825 if (fe_OldToNew(state, fe->fnext, &fe->fnext)) {
1830 /* restore the fe->firstcb entry */
1831 if (cb_OldToNew(state, fe->firstcb, &fe->firstcb)) {
1838 /* restore indices in the CallBack structures */
1839 for (i = 1; i < state->cb_map.len; i++) {
1840 if (state->cb_map.entries[i].new_idx) {
1841 cb = itocb(state->cb_map.entries[i].new_idx);
1843 /* restore the cb->cnext entry */
1844 if (cb_OldToNew(state, cb->cnext, &cb->cnext)) {
1849 /* restore the cb->fhead entry */
1850 if (fe_OldToNew(state, cb->fhead, &cb->fhead)) {
1855 /* restore the cb->hhead entry */
1856 if (h_OldToNew(state, cb->hhead, &cb->hhead)) {
1861 /* restore the cb->tprev entry */
1862 if (cb_OldToNew(state, cb->tprev, &cb->tprev)) {
1867 /* restore the cb->tnext entry */
1868 if (cb_OldToNew(state, cb->tnext, &cb->tnext)) {
1873 /* restore the cb->hprev entry */
1874 if (cb_OldToNew(state, cb->hprev, &cb->hprev)) {
1879 /* restore the cb->hnext entry */
1880 if (cb_OldToNew(state, cb->hnext, &cb->hnext)) {
1887 /* restore the timeout queue head indices */
1888 for (i = 0; i < state->cb_timeout_hdr->records; i++) {
1889 if (cb_OldToNew(state, timeout[i], &timeout[i])) {
1895 /* restore the FE hash table queue heads */
1896 for (i = 0; i < state->cb_fehash_hdr->records; i++) {
1897 if (fe_OldToNew(state, HashTable[i], &HashTable[i])) {
1908 cb_stateVerify(struct fs_dump_state * state)
1912 if (cb_stateVerifyFEHash(state)) {
1916 if (cb_stateVerifyTimeoutQueues(state)) {
1925 cb_stateVerifyFEHash(struct fs_dump_state * state)
1928 struct FileEntry * fe;
1929 afs_uint32 fei, chain_len;
1931 for (i = 0; i < FEHASH_SIZE; i++) {
1933 for (fei = HashTable[i], fe = itofe(fei);
1935 fei = fe->fnext, fe = itofe(fei)) {
1936 if (fei > cbstuff.nblks) {
1937 ViceLog(0, ("cb_stateVerifyFEHash: error: index out of range (fei=%d)\n", fei));
1941 if (cb_stateVerifyFE(state, fe)) {
1944 if (chain_len > FS_STATE_FE_MAX_HASH_CHAIN_LEN) {
1945 ViceLog(0, ("cb_stateVerifyFEHash: error: hash chain %d length exceeds %d; assuming there's a loop\n",
1946 i, FS_STATE_FE_MAX_HASH_CHAIN_LEN));
1959 cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe)
1963 if ((fe->firstcb && !fe->ncbs) ||
1964 (!fe->firstcb && fe->ncbs)) {
1965 ViceLog(0, ("cb_stateVerifyFE: error: fe->firstcb does not agree with fe->ncbs (fei=%d, fe->firstcb=%d, fe->ncbs=%d)\n",
1966 fetoi(fe), fe->firstcb, fe->ncbs));
1969 if (cb_stateVerifyFCBList(state, fe)) {
1970 ViceLog(0, ("cb_stateVerifyFE: error: FCBList failed verification (fei=%d)\n", fetoi(fe)));
1979 cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe)
1982 afs_uint32 cbi, fei, chain_len = 0;
1983 struct CallBack * cb;
1987 for (cbi = fe->firstcb, cb = itocb(cbi);
1989 cbi = cb->cnext, cb = itocb(cbi)) {
1990 if (cbi > cbstuff.nblks) {
1991 ViceLog(0, ("cb_stateVerifyFCBList: error: list index out of range (cbi=%d, ncbs=%d)\n",
1992 cbi, cbstuff.nblks));
1996 if (cb->fhead != fei) {
1997 ViceLog(0, ("cb_stateVerifyFCBList: error: cb->fhead != fei (fei=%d, cb->fhead=%d)\n",
2001 if (chain_len > FS_STATE_FCB_MAX_LIST_LEN) {
2002 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (fei=%d); assuming there's a loop\n",
2003 FS_STATE_FCB_MAX_LIST_LEN, fei));
2010 if (fe->ncbs != chain_len) {
2011 ViceLog(0, ("cb_stateVerifyFCBList: error: list length mismatch (len=%d, fe->ncbs=%d)\n",
2012 chain_len, fe->ncbs));
2021 cb_stateVerifyHCBList(struct fs_dump_state * state, struct host * host)
2024 afs_uint32 hi, chain_len, cbi;
2025 struct CallBack *cb, *ncb;
2030 for (cbi = host->cblist, cb = itocb(cbi);
2032 cbi = cb->hnext, cb = ncb) {
2033 if (chain_len && (host->cblist == cbi)) {
2034 /* we've wrapped around the circular list, and everything looks ok */
2037 if (cb->hhead != hi) {
2038 ViceLog(0, ("cb_stateVerifyHCBList: error: incorrect cb->hhead (cbi=%d, h->index=%d, cb->hhead=%d)\n",
2039 cbi, hi, cb->hhead));
2042 if (!cb->hprev || !cb->hnext) {
2043 ViceLog(0, ("cb_stateVerifyHCBList: error: null index in circular list (cbi=%d, h->index=%d)\n",
2048 if ((cb->hprev > cbstuff.nblks) ||
2049 (cb->hnext > cbstuff.nblks)) {
2050 ViceLog(0, ("cb_stateVerifyHCBList: error: list index out of range (cbi=%d, h->index=%d, cb->hprev=%d, cb->hnext=%d, nCBs=%d)\n",
2051 cbi, hi, cb->hprev, cb->hnext, cbstuff.nblks));
2055 ncb = itocb(cb->hnext);
2056 if (cbi != ncb->hprev) {
2057 ViceLog(0, ("cb_stateVerifyHCBList: error: corrupt linked list (cbi=%d, h->index=%d)\n",
2062 if (chain_len > FS_STATE_HCB_MAX_LIST_LEN) {
2063 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (h->index=%d); assuming there's a loop\n",
2064 FS_STATE_HCB_MAX_LIST_LEN, hi));
2076 cb_stateVerifyTimeoutQueues(struct fs_dump_state * state)
2079 afs_uint32 cbi, chain_len;
2080 struct CallBack *cb, *ncb;
2082 for (i = 0; i < CB_NUM_TIMEOUT_QUEUES; i++) {
2084 for (cbi = timeout[i], cb = itocb(cbi);
2086 cbi = cb->tnext, cb = ncb) {
2087 if (chain_len && (cbi == timeout[i])) {
2088 /* we've wrapped around the circular list, and everything looks ok */
2091 if (cbi > cbstuff.nblks) {
2092 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: list index out of range (cbi=%d, tindex=%d)\n",
2097 if (itot(cb->thead) != &timeout[i]) {
2098 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: cb->thead points to wrong timeout queue (tindex=%d, cbi=%d, cb->thead=%d)\n",
2099 i, cbi, cb->thead));
2102 if (!cb->tprev || !cb->tnext) {
2103 ViceLog(0, ("cb_stateVerifyTimeoutQueues: null index in circular list (cbi=%d, tindex=%d)\n",
2108 if ((cb->tprev > cbstuff.nblks) ||
2109 (cb->tnext > cbstuff.nblks)) {
2110 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list index out of range (cbi=%d, tindex=%d, cb->tprev=%d, cb->tnext=%d, nCBs=%d)\n",
2111 cbi, i, cb->tprev, cb->tnext, cbstuff.nblks));
2115 ncb = itocb(cb->tnext);
2116 if (cbi != ncb->tprev) {
2117 ViceLog(0, ("cb_stateVerifyTimeoutQueues: corrupt linked list (cbi=%d, tindex=%d)\n",
2122 if (chain_len > FS_STATE_TCB_MAX_LIST_LEN) {
2123 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list length exceeds %d (tindex=%d); assuming there's a loop\n",
2124 FS_STATE_TCB_MAX_LIST_LEN, i));
2137 cb_stateSaveTimeouts(struct fs_dump_state * state)
2140 struct iovec iov[2];
2142 AssignInt64(state->eof_offset, &state->cb_hdr->timeout_offset);
2144 memset(state->cb_timeout_hdr, 0, sizeof(struct callback_state_fehash_header));
2145 state->cb_timeout_hdr->magic = CALLBACK_STATE_TIMEOUT_MAGIC;
2146 state->cb_timeout_hdr->records = CB_NUM_TIMEOUT_QUEUES;
2147 state->cb_timeout_hdr->len = sizeof(struct callback_state_timeout_header) +
2148 (state->cb_timeout_hdr->records * sizeof(afs_uint32));
2150 iov[0].iov_base = (char *)state->cb_timeout_hdr;
2151 iov[0].iov_len = sizeof(struct callback_state_timeout_header);
2152 iov[1].iov_base = (char *)timeout;
2153 iov[1].iov_len = sizeof(timeout);
2155 if (fs_stateSeek(state, &state->cb_hdr->timeout_offset)) {
2160 if (fs_stateWriteV(state, iov, 2)) {
2165 fs_stateIncEOF(state, state->cb_timeout_hdr->len);
2172 cb_stateRestoreTimeouts(struct fs_dump_state * state)
2176 if (fs_stateReadHeader(state, &state->cb_hdr->timeout_offset,
2177 state->cb_timeout_hdr,
2178 sizeof(struct callback_state_timeout_header))) {
2183 if (state->cb_timeout_hdr->magic != CALLBACK_STATE_TIMEOUT_MAGIC) {
2187 if (state->cb_timeout_hdr->records != CB_NUM_TIMEOUT_QUEUES) {
2192 len = state->cb_timeout_hdr->records * sizeof(afs_uint32);
2194 if (state->cb_timeout_hdr->len !=
2195 (sizeof(struct callback_state_timeout_header) + len)) {
2200 if (fs_stateRead(state, timeout, len)) {
2210 cb_stateSaveFEHash(struct fs_dump_state * state)
2213 struct iovec iov[2];
2215 AssignInt64(state->eof_offset, &state->cb_hdr->fehash_offset);
2217 memset(state->cb_fehash_hdr, 0, sizeof(struct callback_state_fehash_header));
2218 state->cb_fehash_hdr->magic = CALLBACK_STATE_FEHASH_MAGIC;
2219 state->cb_fehash_hdr->records = FEHASH_SIZE;
2220 state->cb_fehash_hdr->len = sizeof(struct callback_state_fehash_header) +
2221 (state->cb_fehash_hdr->records * sizeof(afs_uint32));
2223 iov[0].iov_base = (char *)state->cb_fehash_hdr;
2224 iov[0].iov_len = sizeof(struct callback_state_fehash_header);
2225 iov[1].iov_base = (char *)HashTable;
2226 iov[1].iov_len = sizeof(HashTable);
2228 if (fs_stateSeek(state, &state->cb_hdr->fehash_offset)) {
2233 if (fs_stateWriteV(state, iov, 2)) {
2238 fs_stateIncEOF(state, state->cb_fehash_hdr->len);
2245 cb_stateRestoreFEHash(struct fs_dump_state * state)
2249 if (fs_stateReadHeader(state, &state->cb_hdr->fehash_offset,
2250 state->cb_fehash_hdr,
2251 sizeof(struct callback_state_fehash_header))) {
2256 if (state->cb_fehash_hdr->magic != CALLBACK_STATE_FEHASH_MAGIC) {
2260 if (state->cb_fehash_hdr->records != FEHASH_SIZE) {
2265 len = state->cb_fehash_hdr->records * sizeof(afs_uint32);
2267 if (state->cb_fehash_hdr->len !=
2268 (sizeof(struct callback_state_fehash_header) + len)) {
2273 if (fs_stateRead(state, HashTable, len)) {
2283 cb_stateSaveFEs(struct fs_dump_state * state)
2286 register int fei, hash;
2287 register struct FileEntry *fe;
2289 AssignInt64(state->eof_offset, &state->cb_hdr->fe_offset);
2291 for (hash = 0; hash < FEHASH_SIZE ; hash++) {
2292 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
2294 if (cb_stateSaveFE(state, fe)) {
2306 cb_stateRestoreFEs(struct fs_dump_state * state)
2308 int count, nFEs, ret = 0;
2310 nFEs = state->cb_hdr->nFEs;
2312 for (count = 0; count < nFEs; count++) {
2313 if (cb_stateRestoreFE(state)) {
2324 cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe)
2326 int ret = 0, iovcnt, cbi, written = 0;
2328 struct callback_state_entry_header hdr;
2329 struct FEDiskEntry fedsk;
2330 struct CBDiskEntry cbdsk[16];
2331 struct iovec iov[16];
2332 struct CallBack *cb;
2335 if (fei > state->cb_hdr->fe_max) {
2336 state->cb_hdr->fe_max = fei;
2339 memset(&hdr, 0, sizeof(struct callback_state_entry_header));
2341 if (cb_stateFEToDiskEntry(fe, &fedsk)) {
2346 iov[0].iov_base = (char *)&hdr;
2347 iov[0].iov_len = sizeof(hdr);
2348 iov[1].iov_base = (char *)&fedsk;
2349 iov[1].iov_len = sizeof(struct FEDiskEntry);
2352 for (cbi = fe->firstcb, cb = itocb(cbi);
2354 cbi = cb->cnext, cb = itocb(cbi), hdr.nCBs++) {
2355 if (cbi > state->cb_hdr->cb_max) {
2356 state->cb_hdr->cb_max = cbi;
2358 if (cb_stateCBToDiskEntry(cb, &cbdsk[iovcnt])) {
2362 cbdsk[iovcnt].index = cbi;
2363 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2364 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2366 if ((iovcnt == 16) || (!cb->cnext)) {
2367 if (fs_stateWriteV(state, iov, iovcnt)) {
2376 hdr.magic = CALLBACK_STATE_ENTRY_MAGIC;
2377 hdr.len = sizeof(hdr) + sizeof(struct FEDiskEntry) +
2378 (hdr.nCBs * sizeof(struct CBDiskEntry));
2381 if (fs_stateWriteV(state, iov, iovcnt)) {
2386 if (fs_stateWriteHeader(state, &state->eof_offset, &hdr, sizeof(hdr))) {
2392 fs_stateIncEOF(state, hdr.len);
2395 if (fs_stateSeek(state, &state->eof_offset)) {
2401 state->cb_hdr->nFEs++;
2402 state->cb_hdr->nCBs += hdr.nCBs;
2409 cb_stateRestoreFE(struct fs_dump_state * state)
2411 int ret = 0, iovcnt, nCBs;
2412 struct callback_state_entry_header hdr;
2413 struct FEDiskEntry fedsk;
2414 struct CBDiskEntry cbdsk[16];
2415 struct iovec iov[16];
2416 struct FileEntry * fe;
2417 struct CallBack * cb;
2419 iov[0].iov_base = (char *)&hdr;
2420 iov[0].iov_len = sizeof(hdr);
2421 iov[1].iov_base = (char *)&fedsk;
2422 iov[1].iov_len = sizeof(fedsk);
2425 if (fs_stateReadV(state, iov, iovcnt)) {
2430 if (hdr.magic != CALLBACK_STATE_ENTRY_MAGIC) {
2437 ViceLog(0, ("cb_stateRestoreFE: ran out of free FileEntry structures\n"));
2442 if (cb_stateDiskEntryToFE(state, &fedsk, fe)) {
2448 for (iovcnt = 0, nCBs = 0;
2451 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2452 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2454 if ((iovcnt == 16) || (nCBs == hdr.nCBs - 1)) {
2455 if (fs_stateReadV(state, iov, iovcnt)) {
2459 if (cb_stateRestoreCBs(state, fe, iov, iovcnt)) {
2473 cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
2474 struct iovec * iov, int niovecs)
2477 register struct CallBack * cb;
2478 struct CBDiskEntry * cbdsk;
2483 for (idx = 0; idx < niovecs; idx++) {
2484 cbdsk = (struct CBDiskEntry *) iov[idx].iov_base;
2485 if ((cb = GetCB()) == NULL) {
2486 ViceLog(0, ("cb_stateRestoreCBs: ran out of free CallBack structures\n"));
2490 if (cb_stateDiskEntryToCB(state, cbdsk, cb)) {
2491 ViceLog(0, ("cb_stateRestoreCBs: corrupt CallBack disk entry\n"));
2503 cb_stateFillHeader(struct callback_state_header * hdr)
2505 hdr->stamp.magic = CALLBACK_STATE_MAGIC;
2506 hdr->stamp.version = CALLBACK_STATE_VERSION;
2507 hdr->tfirst = tfirst;
2512 cb_stateCheckHeader(struct callback_state_header * hdr)
2516 if (hdr->stamp.magic != CALLBACK_STATE_MAGIC) {
2518 } else if (hdr->stamp.version != CALLBACK_STATE_VERSION) {
2520 } else if ((hdr->nFEs > cbstuff.nblks) || (hdr->nCBs > cbstuff.nblks)) {
2521 ViceLog(0, ("cb_stateCheckHeader: saved callback state larger than callback memory allocation\n"));
2527 /* disk entry conversion routines */
2529 cb_stateFEToDiskEntry(struct FileEntry * in, struct FEDiskEntry * out)
2531 memcpy(&out->fe, in, sizeof(struct FileEntry));
2532 out->index = fetoi(in);
2537 cb_stateDiskEntryToFE(struct fs_dump_state * state,
2538 struct FEDiskEntry * in, struct FileEntry * out)
2542 memcpy(out, &in->fe, sizeof(struct FileEntry));
2544 /* setup FE map entry */
2545 if (!in->index || (in->index >= state->fe_map.len)) {
2546 ViceLog(0, ("cb_stateDiskEntryToFE: index (%d) out of range",
2551 state->fe_map.entries[in->index].old_idx = in->index;
2552 state->fe_map.entries[in->index].new_idx = fetoi(out);
2559 cb_stateCBToDiskEntry(struct CallBack * in, struct CBDiskEntry * out)
2561 memcpy(&out->cb, in, sizeof(struct CallBack));
2562 out->index = cbtoi(in);
2567 cb_stateDiskEntryToCB(struct fs_dump_state * state,
2568 struct CBDiskEntry * in, struct CallBack * out)
2572 memcpy(out, &in->cb, sizeof(struct CallBack));
2574 /* setup CB map entry */
2575 if (!in->index || (in->index >= state->cb_map.len)) {
2576 ViceLog(0, ("cb_stateDiskEntryToCB: index (%d) out of range\n",
2581 state->cb_map.entries[in->index].old_idx = in->index;
2582 state->cb_map.entries[in->index].new_idx = cbtoi(out);
2588 /* index map routines */
2590 cb_stateAllocMap(struct fs_dump_state * state)
2592 state->fe_map.len = state->cb_hdr->fe_max + 1;
2593 state->cb_map.len = state->cb_hdr->cb_max + 1;
2594 state->fe_map.entries = (struct idx_map_entry_t *)
2595 calloc(state->fe_map.len, sizeof(struct idx_map_entry_t));
2596 state->cb_map.entries = (struct idx_map_entry_t *)
2597 calloc(state->cb_map.len, sizeof(struct idx_map_entry_t));
2598 return ((state->fe_map.entries != NULL) && (state->cb_map.entries != NULL)) ? 0 : 1;
2602 fe_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2606 /* FEs use a one-based indexing system, so old==0 implies no mapping */
2612 if (old >= state->fe_map.len) {
2613 ViceLog(0, ("fe_OldToNew: index %d is out of range\n", old));
2615 } else if (state->fe_map.entries[old].old_idx != old) { /* sanity check */
2616 ViceLog(0, ("fe_OldToNew: index %d points to an invalid FileEntry record\n", old));
2619 *new = state->fe_map.entries[old].new_idx;
2627 cb_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2631 /* CBs use a one-based indexing system, so old==0 implies no mapping */
2637 if (old >= state->cb_map.len) {
2638 ViceLog(0, ("cb_OldToNew: index %d is out of range\n", old));
2640 } else if (state->cb_map.entries[old].old_idx != old) { /* sanity check */
2641 ViceLog(0, ("cb_OldToNew: index %d points to an invalid CallBack record\n", old));
2644 *new = state->cb_map.entries[old].new_idx;
2650 #endif /* AFS_DEMAND_ATTACH_FS */
2653 DumpCallBackState(void)
2656 afs_uint32 magic = MAGIC, now = FT_ApproxTime(), freelisthead;
2658 oflag = O_WRONLY | O_CREAT | O_TRUNC;
2662 fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, oflag, 0666);
2665 ("Couldn't create callback dump file %s\n",
2666 AFSDIR_SERVER_CBKDUMP_FILEPATH));
2669 (void)write(fd, &magic, sizeof(magic));
2670 (void)write(fd, &now, sizeof(now));
2671 (void)write(fd, &cbstuff, sizeof(cbstuff));
2672 (void)write(fd, TimeOuts, sizeof(TimeOuts));
2673 (void)write(fd, timeout, sizeof(timeout));
2674 (void)write(fd, &tfirst, sizeof(tfirst));
2675 freelisthead = cbtoi((struct CallBack *)CBfree);
2676 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2677 freelisthead = fetoi((struct FileEntry *)FEfree);
2678 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2679 (void)write(fd, HashTable, sizeof(HashTable));
2680 (void)write(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2681 (void)write(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2687 #endif /* !INTERPRET_DUMP */
2689 #ifdef INTERPRET_DUMP
2691 /* This is only compiled in for the callback analyzer program */
2692 /* Returns the time of the dump */
2694 ReadDump(char *file)
2697 afs_uint32 magic, freelisthead;
2704 fd = open(file, oflag);
2706 fprintf(stderr, "Couldn't read dump file %s\n", file);
2709 read(fd, &magic, sizeof(magic));
2710 if (magic != MAGIC) {
2712 "Magic number of %s is invalid. You might be trying to\n",
2715 "run this program on a machine type with a different byte ordering.\n");
2718 read(fd, &now, sizeof(now));
2719 read(fd, &cbstuff, sizeof(cbstuff));
2720 read(fd, TimeOuts, sizeof(TimeOuts));
2721 read(fd, timeout, sizeof(timeout));
2722 read(fd, &tfirst, sizeof(tfirst));
2723 read(fd, &freelisthead, sizeof(freelisthead));
2724 CB = ((struct CallBack
2725 *)(calloc(cbstuff.nblks, sizeof(struct CallBack)))) - 1;
2726 FE = ((struct FileEntry
2727 *)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
2728 CBfree = (struct CallBack *)itocb(freelisthead);
2729 read(fd, &freelisthead, sizeof(freelisthead));
2730 FEfree = (struct FileEntry *)itofe(freelisthead);
2731 read(fd, HashTable, sizeof(HashTable));
2732 read(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2733 read(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2735 perror("Error reading dumpfile");
2742 #include "AFS_component_version_number.h"
2744 #include "AFS_component_version_number.c"
2748 main(int argc, char **argv)
2750 int err = 0, cbi = 0, stats = 0, noptions = 0, all = 0, vol = 0, raw = 0;
2752 register struct FileEntry *fe;
2753 register struct CallBack *cb;
2756 memset(&fid, 0, sizeof(fid));
2759 while (argc && **argv == '-') {
2762 if (!strcmp(*argv, "-host")) {
2768 cbi = atoi(*++argv);
2769 } else if (!strcmp(*argv, "-fid")) {
2775 fid.Volume = atoi(*++argv);
2776 fid.Vnode = atoi(*++argv);
2777 fid.Unique = atoi(*++argv);
2778 } else if (!strcmp(*argv, "-time")) {
2779 fprintf(stderr, "-time not supported\n");
2781 } else if (!strcmp(*argv, "-stats")) {
2783 } else if (!strcmp(*argv, "-all")) {
2785 } else if (!strcmp(*argv, "-raw")) {
2787 } else if (!strcmp(*argv, "-volume")) {
2793 vol = atoi(*++argv);
2798 if (err || argc != 1) {
2800 "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] callbackdumpfile\n");
2802 "[cbid is shown for each host in the hosts.dump file]\n");
2805 now = ReadDump(*argv);
2806 if (stats || noptions == 0) {
2807 time_t uxtfirst = UXtime(tfirst);
2808 printf("The time of the dump was %u %s", now, ctime(&now));
2809 printf("The last time cleanup ran was %u %s", uxtfirst,
2811 PrintCallBackStats();
2816 struct CallBack *cb;
2817 struct FileEntry *fe;
2819 for (hash = 0; hash < FEHASH_SIZE; hash++) {
2820 for (feip = &HashTable[hash]; fe = itofe(*feip);) {
2821 if (!vol || (fe->volid == vol)) {
2822 register struct CallBack *cbnext;
2823 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
2825 cbnext = itocb(cb->cnext);
2835 afs_uint32 cfirst = cbi;
2840 } while (cbi != cfirst);
2845 printf("No callback entries for %u.%u\n", fid.Volume, fid.Vnode);
2848 cb = itocb(fe->firstcb);
2851 cb = itocb(cb->cnext);
2856 for (i = 1; i < cbstuff.nblks; i++) {
2857 p = (afs_int32 *) & FE[i];
2858 printf("%d:%12x%12x%12x%12x\n", i, p[0], p[1], p[2], p[3]);
2865 PrintCB(register struct CallBack *cb, afs_uint32 now)
2867 struct FileEntry *fe = itofe(cb->fhead);
2868 time_t expires = TIndexToTime(cb->thead);
2873 printf("vol=%u vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %d secs at %s",
2874 fe->volid, fe->vnode, fe->ncbs, cb->hhead, cb->status, fe->status,
2875 expires - now, ctime(&expires));
2880 #if !defined(INTERPRET_DUMP)
2882 ** try breaking calbacks on afidp from host. Use multi_rx.
2883 ** return 0 on success, non-zero on failure
2886 MultiBreakCallBackAlternateAddress(struct host *host, struct AFSCBFids *afidp)
2890 retVal = MultiBreakCallBackAlternateAddress_r(host, afidp);
2896 MultiBreakCallBackAlternateAddress_r(struct host *host,
2897 struct AFSCBFids *afidp)
2900 struct rx_connection **conns;
2901 struct rx_connection *connSuccess = 0;
2902 struct AddrPort *interfaces;
2903 static struct rx_securityClass *sc = 0;
2904 static struct AFSCBs tc = { 0, 0 };
2907 /* nothing more can be done */
2908 if (!host->interface)
2909 return 1; /* failure */
2911 /* the only address is the primary interface */
2912 if (host->interface->numberOfInterfaces <= 1)
2913 return 1; /* failure */
2915 /* initialise a security object only once */
2917 sc = rxnull_NewClientSecurityObject();
2919 i = host->interface->numberOfInterfaces;
2920 interfaces = calloc(i, sizeof(struct AddrPort));
2921 conns = calloc(i, sizeof(struct rx_connection *));
2922 if (!interfaces || !conns) {
2924 ("Failed malloc in MultiBreakCallBackAlternateAddress_r\n"));
2928 /* initialize alternate rx connections */
2929 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
2930 /* this is the current primary address */
2931 if (host->host == host->interface->interface[i].addr &&
2932 host->port == host->interface->interface[i].port)
2935 interfaces[j] = host->interface->interface[i];
2937 rx_NewConnection(interfaces[j].addr,
2938 interfaces[j].port, 1, sc, 0);
2939 rx_SetConnDeadTime(conns[j], 2);
2940 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
2944 assert(j); /* at least one alternate address */
2946 ("Starting multibreakcall back on all addr for host %x (%s:%d)\n",
2947 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
2949 multi_Rx(conns, j) {
2950 multi_RXAFSCB_CallBack(afidp, &tc);
2954 if (host->callback_rxcon)
2955 rx_DestroyConnection(host->callback_rxcon);
2956 host->callback_rxcon = conns[multi_i];
2957 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
2958 host->host = interfaces[multi_i].addr;
2959 host->port = interfaces[multi_i].port;
2960 h_AddHostToAddrHashTable_r(host->host, host->port, host);
2961 connSuccess = conns[multi_i];
2962 rx_SetConnDeadTime(host->callback_rxcon, 50);
2963 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
2965 ("multibreakcall success with addr %s:%d\n",
2966 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
2967 ntohs(interfaces[multi_i].port)));
2974 /* Destroy all connections except the one on which we succeeded */
2975 for (i = 0; i < j; i++)
2976 if (conns[i] != connSuccess)
2977 rx_DestroyConnection(conns[i]);
2983 return 0; /* success */
2985 return 1; /* failure */
2990 ** try multi_RX probes to host.
2991 ** return 0 on success, non-0 on failure
2994 MultiProbeAlternateAddress_r(struct host *host)
2997 struct rx_connection **conns;
2998 struct rx_connection *connSuccess = 0;
2999 struct AddrPort *interfaces;
3000 static struct rx_securityClass *sc = 0;
3003 /* nothing more can be done */
3004 if (!host->interface)
3005 return 1; /* failure */
3007 /* the only address is the primary interface */
3008 if (host->interface->numberOfInterfaces <= 1)
3009 return 1; /* failure */
3011 /* initialise a security object only once */
3013 sc = rxnull_NewClientSecurityObject();
3015 i = host->interface->numberOfInterfaces;
3016 interfaces = calloc(i, sizeof(struct AddrPort));
3017 conns = calloc(i, sizeof(struct rx_connection *));
3018 if (!interfaces || !conns) {
3019 ViceLog(0, ("Failed malloc in MultiProbeAlternateAddress_r\n"));
3023 /* initialize alternate rx connections */
3024 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3025 /* this is the current primary address */
3026 if (host->host == host->interface->interface[i].addr &&
3027 host->port == host->interface->interface[i].port)
3030 interfaces[j] = host->interface->interface[i];
3032 rx_NewConnection(interfaces[j].addr,
3033 interfaces[j].port, 1, sc, 0);
3034 rx_SetConnDeadTime(conns[j], 2);
3035 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3039 assert(j); /* at least one alternate address */
3041 ("Starting multiprobe on all addr for host %x (%s:%d)\n",
3042 host, afs_inet_ntoa_r(host->host, hoststr),
3043 ntohs(host->port)));
3045 multi_Rx(conns, j) {
3046 multi_RXAFSCB_ProbeUuid(&host->interface->uuid);
3050 if (host->callback_rxcon)
3051 rx_DestroyConnection(host->callback_rxcon);
3052 host->callback_rxcon = conns[multi_i];
3053 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
3054 host->host = interfaces[multi_i].addr;
3055 host->port = interfaces[multi_i].port;
3056 h_AddHostToAddrHashTable_r(host->host, host->port, host);
3057 connSuccess = conns[multi_i];
3058 rx_SetConnDeadTime(host->callback_rxcon, 50);
3059 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3061 ("multiprobe success with addr %s:%d\n",
3062 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3063 ntohs(interfaces[multi_i].port)));
3068 ("multiprobe failure with addr %s:%d\n",
3069 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3070 ntohs(interfaces[multi_i].port)));
3072 /* This is less than desirable but its the best we can do.
3073 * The AFS Cache Manager will return either 0 for a Uuid
3074 * match and a 1 for a non-match. If the error is 1 we
3075 * therefore know that our mapping of IP address to Uuid
3076 * is wrong. We should attempt to find the correct
3077 * Uuid and fix the host tables.
3079 if (multi_error == 1) {
3080 /* remove the current alternate address from this host */
3082 removeInterfaceAddr_r(host, interfaces[multi_i].addr, interfaces[multi_i].port);
3086 #ifdef AFS_DEMAND_ATTACH_FS
3087 /* try to bail ASAP if the fileserver is shutting down */
3089 if (fs_state.mode == FS_MODE_SHUTDOWN) {
3098 /* Destroy all connections except the one on which we succeeded */
3099 for (i = 0; i < j; i++)
3100 if (conns[i] != connSuccess)
3101 rx_DestroyConnection(conns[i]);
3107 return 0; /* success */
3109 return 1; /* failure */
3112 #endif /* !defined(INTERPRET_DUMP) */