2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
9 * Portions Copyright (c) 2006 Sine Nomine Associates
13 * NEW callback package callback.c (replaces vicecb.c)
14 * Updated call back routines, NOW with:
16 * Faster DeleteVenus (Now called DeleteAllCallBacks)
17 * Call back breaking for volumes
18 * Adaptive timeouts on call backs
19 * Architected for Multi RPC
20 * No locks (currently implicit vnode locks--these will go, to)
21 * Delayed call back when rpc connection down.
22 * Bulk break of delayed call backs when rpc connection
24 * Strict limit on number of call backs.
26 * InitCallBack(nblocks)
27 * Initialize: nblocks is max number # of file entries + # of callback entries
28 * nblocks must be < 65536
29 * Space used is nblocks*16 bytes
30 * Note that space will be reclaimed by breaking callbacks of old hosts
32 * time = AddCallBack(host, fid)
34 * Returns the expiration time at the workstation.
36 * BreakCallBack(host, fid)
37 * Break all call backs for fid, except for the specified host.
40 * BreakVolumeCallBacksLater(volume)
41 * Break all call backs on volume, using single call to each host
42 * Delete all the call backs.
44 * DeleteCallBack(host,fid)
45 * Delete (do not break) single call back for fid.
47 * DeleteFileCallBacks(fid)
48 * Delete (do not break) all call backs for fid.
50 * DeleteAllCallBacks(host)
51 * Delete (do not break) all call backs for host.
53 * CleanupTimedOutCallBacks()
54 * Delete all timed out call back entries
55 * Must be called periodically by file server.
57 * BreakDelayedCallBacks(host)
58 * Break all delayed call backs for host.
59 * Returns 1: one or more failed, 0: success.
61 * PrintCallBackStats()
62 * Print statistics about call backs to stdout.
64 * DumpCallBacks() ---wishful thinking---
65 * Dump call back state to /tmp/callback.state.
66 * This is separately interpretable by the program pcb.
68 * Notes: In general, if a call back to a host doesn't get through,
69 * then HostDown, supplied elsewhere, is called. BreakDelayedCallBacks,
70 * however, does not call HostDown, but instead returns an indication of
71 * success if all delayed call backs were finally broken.
73 * BreakDelayedCallBacks MUST be called at the first sign of activity
74 * from the host after HostDown has been called (or a previous
75 * BreakDelayedCallBacks failed). The BreakDelayedCallBacks must be
76 * allowed to complete before any requests from that host are handled.
77 * If BreakDelayedCallBacks fails, then the host should remain
78 * down (and the request should be failed).
80 * CleanupCallBacks MUST be called periodically by the file server for
81 * this package to work correctly. Every 5 minutes is suggested.
84 #include <afsconfig.h>
85 #include <afs/param.h>
90 #ifdef HAVE_SYS_FILE_H
94 #include <afs/nfs.h> /* yuck. This is an abomination. */
97 #include <afs/afscbint.h>
98 #include <afs/afsutil.h>
100 #include <afs/ihandle.h>
101 #include <afs/vnode.h>
102 #include <afs/volume.h>
103 #include "viced_prototypes.h"
106 #include <afs/ptclient.h> /* need definition of prlist for host.h */
108 #include "callback.h"
109 #ifdef AFS_DEMAND_ATTACH_FS
110 #include "serialize_state.h"
111 #endif /* AFS_DEMAND_ATTACH_FS */
114 extern afsUUID FS_HostUUID;
115 extern int hostCount;
117 #ifndef INTERPRET_DUMP
118 static int ShowProblems = 1;
121 struct cbcounters cbstuff;
123 static struct FileEntry * FE = NULL; /* don't use FE[0] */
124 static struct CallBack * CB = NULL; /* don't use CB[0] */
126 static struct CallBack * CBfree = NULL;
127 static struct FileEntry * FEfree = NULL;
130 /* Time to live for call backs depends upon number of users of the file.
131 * TimeOuts is indexed by this number/8 (using TimeOut macro). Times
132 * in this table are for the workstation; server timeouts, add
135 static int TimeOuts[] = {
136 /* Note: don't make the first entry larger than 4 hours (see above) */
137 4 * 60 * 60, /* 0-7 users */
138 1 * 60 * 60, /* 8-15 users */
139 30 * 60, /* 16-23 users */
140 15 * 60, /* 24-31 users */
141 15 * 60, /* 32-39 users */
142 10 * 60, /* 40-47 users */
143 10 * 60, /* 48-55 users */
144 10 * 60, /* 56-63 users */
145 }; /* Anything more: MinTimeOut */
147 /* minimum time given for a call back */
148 #ifndef INTERPRET_DUMP
149 static int MinTimeOut = (7 * 60);
152 /* Heads of CB queues; a timeout index is 1+index into this array */
153 static afs_uint32 timeout[CB_NUM_TIMEOUT_QUEUES];
155 static afs_int32 tfirst; /* cbtime of oldest unexpired call back time queue */
158 /* 16 byte object get/free routines */
163 /* Prototypes for static routines */
164 static struct FileEntry *FindFE(AFSFid * fid);
166 #ifndef INTERPRET_DUMP
167 static struct CallBack *iGetCB(int *nused);
168 static int iFreeCB(struct CallBack *cb, int *nused);
169 static struct FileEntry *iGetFE(int *nused);
170 static int iFreeFE(struct FileEntry *fe, int *nused);
171 static int TAdd(struct CallBack *cb, afs_uint32 * thead);
172 static int TDel(struct CallBack *cb);
173 static int HAdd(struct CallBack *cb, struct host *host);
174 static int HDel(struct CallBack *cb);
175 static int CDel(struct CallBack *cb, int deletefe);
176 static int CDelPtr(struct FileEntry *fe, afs_uint32 * cbp,
178 static afs_uint32 *FindCBPtr(struct FileEntry *fe, struct host *host);
179 static int FDel(struct FileEntry *fe);
180 static int AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead,
181 int type, int locked);
182 static void MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
183 struct AFSCBFids *afidp);
184 static int MultiBreakVolumeCallBack_r(struct host *host,
185 struct VCBParams *parms, int deletefe);
186 static int MultiBreakVolumeLaterCallBack(struct host *host, void *rock);
187 static int GetSomeSpace_r(struct host *hostp, int locked);
188 static int ClearHostCallbacks_r(struct host *hp, int locked);
189 static int DumpCallBackState_r(void);
192 #define GetCB() ((struct CallBack *)iGetCB(&cbstuff.nCBs))
193 #define GetFE() ((struct FileEntry *)iGetFE(&cbstuff.nFEs))
194 #define FreeCB(cb) iFreeCB((struct CallBack *)cb, &cbstuff.nCBs)
195 #define FreeFE(fe) iFreeFE((struct FileEntry *)fe, &cbstuff.nFEs)
198 /* Other protos - move out sometime */
199 void PrintCB(struct CallBack *cb, afs_uint32 now);
201 static afs_uint32 HashTable[FEHASH_SIZE]; /* File entry hash table */
203 static struct FileEntry *
208 struct FileEntry *fe;
210 hash = FEHash(fid->Volume, fid->Unique);
211 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
213 if (fe->volid == fid->Volume && fe->unique == fid->Unique
214 && fe->vnode == fid->Vnode && (fe->status & FE_LATER) != FE_LATER)
220 #ifndef INTERPRET_DUMP
222 static struct CallBack *
225 struct CallBack *ret;
227 if ((ret = CBfree)) {
228 CBfree = (struct CallBack *)(((struct object *)ret)->next);
235 iFreeCB(struct CallBack *cb, int *nused)
237 ((struct object *)cb)->next = (struct object *)CBfree;
243 static struct FileEntry *
246 struct FileEntry *ret;
248 if ((ret = FEfree)) {
249 FEfree = (struct FileEntry *)(((struct object *)ret)->next);
256 iFreeFE(struct FileEntry *fe, int *nused)
258 ((struct object *)fe)->next = (struct object *)FEfree;
264 /* Add cb to end of specified timeout list */
266 TAdd(struct CallBack *cb, afs_uint32 * thead)
269 (*thead) = cb->tnext = cb->tprev = cbtoi(cb);
271 struct CallBack *thp = itocb(*thead);
273 cb->tprev = thp->tprev;
277 thp->tprev = (itocb(thp->tprev)->tnext = cbtoi(cb));
279 thp->tprev = cbtoi(cb);
282 cb->thead = ttoi(thead);
286 /* Delete call back entry from timeout list */
288 TDel(struct CallBack *cb)
290 afs_uint32 *thead = itot(cb->thead);
292 if (*thead == cbtoi(cb))
293 *thead = (*thead == cb->tnext ? 0 : cb->tnext);
294 if (itocb(cb->tprev))
295 itocb(cb->tprev)->tnext = cb->tnext;
296 if (itocb(cb->tnext))
297 itocb(cb->tnext)->tprev = cb->tprev;
301 /* Add cb to end of specified host list */
303 HAdd(struct CallBack *cb, struct host *host)
305 cb->hhead = h_htoi(host);
307 host->cblist = cb->hnext = cb->hprev = cbtoi(cb);
309 struct CallBack *fcb = itocb(host->cblist);
311 cb->hprev = fcb->hprev;
312 cb->hnext = cbtoi(fcb);
313 fcb->hprev = (itocb(fcb->hprev)->hnext = cbtoi(cb));
318 /* Delete call back entry from host list */
320 HDel(struct CallBack *cb)
322 afs_uint32 *hhead = &h_itoh(cb->hhead)->cblist;
324 if (*hhead == cbtoi(cb))
325 *hhead = (*hhead == cb->hnext ? 0 : cb->hnext);
326 itocb(cb->hprev)->hnext = cb->hnext;
327 itocb(cb->hnext)->hprev = cb->hprev;
331 /* Delete call back entry from fid's chain of cb's */
332 /* N.B. This one also deletes the CB, and also possibly parent FE, so
333 * make sure that it is not on any other list before calling this
336 CDel(struct CallBack *cb, int deletefe)
339 struct FileEntry *fe = itofe(cb->fhead);
343 for (safety = 0, cbp = &fe->firstcb; *cbp && *cbp != cbi;
344 cbp = &itocb(*cbp)->cnext, safety++) {
345 if (safety > cbstuff.nblks + 10) {
346 ViceLogThenPanic(0, ("CDel: Internal Error -- shutting down: "
347 "wanted %d from %d, now at %d\n",
348 cbi, fe->firstcb, *cbp));
349 DumpCallBackState_r();
350 ShutDownAndCore(PANIC);
353 CDelPtr(fe, cbp, deletefe);
357 /* Same as CDel, but pointer to parent pointer to CB entry is passed,
358 * as well as file entry */
359 /* N.B. This one also deletes the CB, and also possibly parent FE, so
360 * make sure that it is not on any other list before calling this
362 static int Ccdelpt = 0, CcdelB = 0;
365 CDelPtr(struct FileEntry *fe, afs_uint32 * cbp,
378 if ((--fe->ncbs == 0) && deletefe)
384 FindCBPtr(struct FileEntry *fe, struct host *host)
386 afs_uint32 hostindex = h_htoi(host);
391 for (safety = 0, cbp = &fe->firstcb; *cbp; cbp = &cb->cnext, safety++) {
392 if (safety > cbstuff.nblks) {
393 ViceLog(0, ("FindCBPtr: Internal Error -- shutting down.\n"));
394 DumpCallBackState_r();
395 ShutDownAndCore(PANIC);
398 if (cb->hhead == hostindex)
404 /* Delete file entry from hash table */
406 FDel(struct FileEntry *fe)
409 afs_uint32 *p = &HashTable[FEHash(fe->volid, fe->unique)];
411 while (*p && *p != fei)
412 p = &itofe(*p)->fnext;
419 /* initialize the callback package */
421 InitCallBack(int nblks)
424 tfirst = CBtime(time(NULL));
425 /* N.B. The "-1", below, is because
426 * FE[0] and CB[0] are not used--and not allocated */
427 FE = ((struct FileEntry *)(calloc(nblks, sizeof(struct FileEntry))));
429 ViceLogThenPanic(0, ("Failed malloc in InitCallBack\n"));
431 FE--; /* FE[0] is supposed to point to junk */
432 cbstuff.nFEs = nblks;
434 FreeFE(&FE[cbstuff.nFEs]); /* This is correct */
435 CB = ((struct CallBack *)(calloc(nblks, sizeof(struct CallBack))));
437 ViceLogThenPanic(0, ("Failed malloc in InitCallBack\n"));
439 CB--; /* CB[0] is supposed to point to junk */
440 cbstuff.nCBs = nblks;
442 FreeCB(&CB[cbstuff.nCBs]); /* This is correct */
443 cbstuff.nblks = nblks;
444 cbstuff.nbreakers = 0;
450 XCallBackBulk_r(struct host * ahost, struct AFSFid * fids, afs_int32 nfids)
452 struct AFSCallBack tcbs[AFSCBMAX];
458 struct rx_connection *cb_conn = NULL;
460 rx_SetConnDeadTime(ahost->callback_rxcon, 4);
461 rx_SetConnHardDeadTime(ahost->callback_rxcon, AFS_HARDDEADTIME);
467 for (i = 0; i < nfids && i < AFSCBMAX; i++) {
468 tcbs[i].CallBackVersion = CALLBACK_VERSION;
469 tcbs[i].ExpirationTime = 0;
470 tcbs[i].CallBackType = CB_DROPPED;
472 tf.AFSCBFids_len = i;
473 tf.AFSCBFids_val = &(fids[j]);
477 tc.AFSCBs_val = tcbs;
479 cb_conn = ahost->callback_rxcon;
480 rx_GetConnection(cb_conn);
482 code |= RXAFSCB_CallBack(cb_conn, &tf, &tc);
483 rx_PutConnection(cb_conn);
491 /* the locked flag tells us if the host entry has already been locked
492 * by our parent. I don't think anybody actually calls us with the
493 * host locked, but here's how to make that work: GetSomeSpace has to
494 * change so that it doesn't attempt to lock any hosts < "host". That
495 * means that it might be unable to free any objects, so it has to
496 * return an exit status. If it fails, then AddCallBack1 might fail,
497 * as well. If so, the host->ResetDone should probably be set to 0,
498 * and we probably don't want to return a callback promise to the
499 * cache manager, either. */
501 AddCallBack1(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
509 if (!(host->hostFlags & HOSTDELETED))
510 retVal = AddCallBack1_r(host, fid, thead, type, 1);
520 AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
523 struct FileEntry *fe;
524 struct CallBack *cb = 0, *lastcb = 0;
525 struct FileEntry *newfe = 0;
526 afs_uint32 time_out = 0;
527 afs_uint32 *Thead = thead;
528 struct CallBack *newcb = 0;
531 cbstuff.AddCallBacks++;
535 /* allocate these guys first, since we can't call the allocator with
536 * the host structure locked -- or we might deadlock. However, we have
537 * to avoid races with FindFE... */
538 while (!(newcb = GetCB())) {
539 GetSomeSpace_r(host, locked);
541 while (!(newfe = GetFE())) { /* Get it now, so we don't have to call */
542 /* GetSomeSpace with the host locked, later. This might turn out to */
543 /* have been unneccessary, but that's actually kind of unlikely, since */
544 /* most files are not shared. */
545 GetSomeSpace_r(host, locked);
549 h_Lock_r(host); /* this can yield, so do it before we get any */
551 if (host->hostFlags & HOSTDELETED) {
559 if (type == CB_NORMAL) {
561 TimeCeiling(time(NULL) + TimeOut(fe ? fe->ncbs : 0) +
563 Thead = THead(CBtime(time_out));
564 } else if (type == CB_VOLUME) {
565 time_out = TimeCeiling((60 * 120 + time(NULL)) + ServerBias);
566 Thead = THead(CBtime(time_out));
567 } else if (type == CB_BULK) {
568 /* bulk status can get so many callbacks all at once, and most of them
569 * are probably not for things that will be used for long.
572 TimeCeiling(time(NULL) + ServerBias +
573 TimeOut(22 + (fe ? fe->ncbs : 0)));
574 Thead = THead(CBtime(time_out));
585 fe->volid = fid->Volume;
586 fe->vnode = fid->Vnode;
587 fe->unique = fid->Unique;
590 hash = FEHash(fid->Volume, fid->Unique);
591 fe->fnext = HashTable[hash];
592 HashTable[hash] = fetoi(fe);
594 for (safety = 0, lastcb = cb = itocb(fe->firstcb); cb;
595 lastcb = cb, cb = itocb(cb->cnext), safety++) {
596 if (safety > cbstuff.nblks) {
597 ViceLog(0, ("AddCallBack1: Internal Error -- shutting down.\n"));
598 DumpCallBackState_r();
599 ShutDownAndCore(PANIC);
601 if (cb->hhead == h_htoi(host))
604 if (cb) { /* Already have call back: move to new timeout list */
605 /* don't change delayed callbacks back to normal ones */
606 if (cb->status != CB_DELAYED)
608 /* Only move if new timeout is longer */
609 if (TNorm(ttoi(Thead)) > TNorm(cb->thead)) {
613 if (newfe == NULL) { /* we are using the new FE */
614 fe->firstcb = cbtoi(cb);
616 cb->fhead = fetoi(fe);
621 *(lastcb ? &lastcb->cnext : &fe->firstcb) = cbtoi(cb);
624 cb->fhead = fetoi(fe);
630 /* now free any still-unused callback or host entries */
636 if (!locked) /* freecb and freefe might(?) yield */
639 if (type == CB_NORMAL || type == CB_VOLUME || type == CB_BULK)
640 return time_out - ServerBias; /* Expires sooner at workstation */
646 CompareCBA(const void *e1, const void *e2)
648 const struct cbstruct *cba1 = (const struct cbstruct *)e1;
649 const struct cbstruct *cba2 = (const struct cbstruct *)e2;
650 return ((cba1->hp)->index - (cba2->hp)->index);
653 /* Take an array full of hosts, all held. Break callbacks to them, and
654 * release the holds once you're done.
655 * Currently only works for a single Fid in afidp array.
656 * If you want to make this work with multiple fids, you need to fix
657 * the error handling. One approach would be to force a reset if a
658 * multi-fid call fails, or you could add delayed callbacks for each
659 * fid. You probably also need to sort and remove duplicate hosts.
660 * When this is called from the BreakVolumeCallBacks path, it does NOT
661 * force a reset if the RPC fails, it just marks the host down and tries
662 * to create a delayed callback. */
663 /* N.B. be sure that code works when ncbas == 0 */
664 /* N.B. requires all the cba[*].hp pointers to be valid... */
665 /* This routine does not hold a lock on the host for the duration of
666 * the BreakCallBack RPC, which is a significant deviation from tradition.
667 * It _does_ get a lock on the host before setting VenusDown = 1,
668 * which is sufficient only if VenusDown = 0 only happens when the
669 * lock is held over the RPC and the subsequent VenusDown == 0
670 * wherever that is done. */
672 MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
673 struct AFSCBFids *afidp)
676 struct rx_connection *conns[MAX_CB_HOSTS];
677 static struct AFSCBs tc = { 0, 0 };
678 int multi_to_cba_map[MAX_CB_HOSTS];
680 osi_Assert(ncbas <= MAX_CB_HOSTS);
682 /* sort cba list to avoid makecall issues */
683 qsort(cba, ncbas, sizeof(struct cbstruct), CompareCBA);
685 /* set up conns for multi-call */
686 for (i = 0, j = 0; i < ncbas; i++) {
687 struct host *thishost = cba[i].hp;
688 if (!thishost || (thishost->hostFlags & HOSTDELETED)) {
691 rx_GetConnection(thishost->callback_rxcon);
692 multi_to_cba_map[j] = i;
693 conns[j++] = thishost->callback_rxcon;
695 rx_SetConnDeadTime(thishost->callback_rxcon, 4);
696 rx_SetConnHardDeadTime(thishost->callback_rxcon, AFS_HARDDEADTIME);
699 if (j) { /* who knows what multi would do with 0 conns? */
703 multi_RXAFSCB_CallBack(afidp, &tc);
709 i = multi_to_cba_map[multi_i];
715 ("BCB: INTERNAL ERROR: hp=%p, cba=%p, thead=%u\n",
719 ** try breaking callbacks on alternate interface addresses
721 if (MultiBreakCallBackAlternateAddress(hp, afidp)) {
724 ("BCB: Failed on file %u.%u.%u, "
725 "Host %p (%s:%d) is down\n",
726 afidp->AFSCBFids_val->Volume,
727 afidp->AFSCBFids_val->Vnode,
728 afidp->AFSCBFids_val->Unique,
730 afs_inet_ntoa_r(hp->host, hoststr),
736 if (!(hp->hostFlags & HOSTDELETED)) {
737 hp->hostFlags |= VENUSDOWN;
739 * We always go into AddCallBack1_r with the host locked
741 AddCallBack1_r(hp, afidp->AFSCBFids_val, itot(idx),
755 for (i = 0; i < ncbas; i++) {
763 /* H_UNLOCK around this so h_FreeConnection does not deadlock.
764 h_FreeConnection should *never* be called on a callback connection,
765 but on 10/27/04 a deadlock occurred where it was, when we know why,
766 this should be reverted. -- shadow */
768 for (i = 0; i < j; i++) {
769 rx_PutConnection(conns[i]);
777 * Break all call backs for fid, except for the specified host (unless flag
778 * is true, in which case all get a callback message. Assumption: the specified
779 * host is h_Held, by the caller; the others aren't.
780 * Specified host may be bogus, that's ok. This used to check to see if the
781 * host was down in two places, once right after the host was h_held, and
782 * again after it was locked. That race condition is incredibly rare and
783 * relatively harmless even when it does occur, so we don't check for it now.
785 /* if flag is true, send a break callback msg to "host", too */
787 BreakCallBack(struct host *xhost, AFSFid * fid, int flag)
789 struct FileEntry *fe;
790 struct CallBack *cb, *nextcb;
791 struct cbstruct cbaDef[MAX_CB_HOSTS], *cba = cbaDef;
792 unsigned int ncbas, cbaAlloc = MAX_CB_HOSTS;
799 ("BCB: BreakCallBack(Host %p all but %s:%d, (%u,%u,%u))\n",
800 xhost, afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
801 fid->Volume, fid->Vnode, fid->Unique));
804 ("BCB: BreakCallBack(No Host, (%u,%u,%u))\n",
805 fid->Volume, fid->Vnode, fid->Unique));
808 cbstuff.BreakCallBacks++;
813 hostindex = xhost ? h_htoi(xhost) : 0;
814 cb = itocb(fe->firstcb);
815 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
816 /* the most common case is what follows the || */
819 tf.AFSCBFids_len = 1;
820 tf.AFSCBFids_val = fid;
822 for (ncbas = 0; cb ; cb = nextcb) {
823 nextcb = itocb(cb->cnext);
824 if ((cb->hhead != hostindex || flag)
825 && (cb->status == CB_BULK || cb->status == CB_NORMAL
826 || cb->status == CB_VOLUME)) {
827 struct host *thishost = h_itoh(cb->hhead);
829 ViceLog(0, ("BCB: BOGUS! cb->hhead is NULL!\n"));
830 } else if (thishost->hostFlags & VENUSDOWN) {
832 ("BCB: %p (%s:%d) is down; delaying break call back\n",
833 thishost, afs_inet_ntoa_r(thishost->host, hoststr),
834 ntohs(thishost->port)));
835 cb->status = CB_DELAYED;
837 if (!(thishost->hostFlags & HOSTDELETED)) {
839 if (ncbas == cbaAlloc) { /* Need more space */
840 int curLen = cbaAlloc*sizeof(cba[0]);
841 struct cbstruct *cbaOld = (cba == cbaDef) ? NULL : cba;
843 /* There are logical contraints elsewhere that the number of hosts
844 (i.e. h_HTSPERBLOCK*h_MAXHOSTTABLES) remains in the realm of a signed "int".
845 cbaAlloc is defined unsigned int hence doubling below cannot overflow
847 cbaAlloc = cbaAlloc<<1; /* double */
848 cba = realloc(cbaOld, cbaAlloc * sizeof(cba[0]));
850 if (cbaOld == NULL) { /* realloc wouldn't have copied from cbaDef */
851 memcpy(cba, cbaDef, curLen);
854 cba[ncbas].hp = thishost;
855 cba[ncbas].thead = cb->thead;
860 CDel(cb, 1); /* Usually first; so this delete
861 * is reasonably inexpensive */
867 struct cbstruct *cba2;
870 for (cba2 = cba, num = ncbas; ncbas > 0; cba2 += num, ncbas -= num) {
871 num = (ncbas > MAX_CB_HOSTS) ? MAX_CB_HOSTS : ncbas;
872 MultiBreakCallBack_r(cba2, num, &tf);
876 if (cba != cbaDef) free(cba);
883 /* Delete (do not break) single call back for fid */
885 DeleteCallBack(struct host *host, AFSFid * fid)
887 struct FileEntry *fe;
892 cbstuff.DeleteCallBacks++;
895 /* do not care if the host has been HOSTDELETED */
901 ("DCB: No call backs for fid (%u, %u, %u)\n", fid->Volume,
902 fid->Vnode, fid->Unique));
905 pcb = FindCBPtr(fe, host);
908 ("DCB: No call back for host %p (%s:%d), (%u, %u, %u)\n",
909 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
910 fid->Volume, fid->Vnode, fid->Unique));
924 * Delete (do not break) all call backs for fid. This call doesn't
925 * set all of the various host locks, but it shouldn't really matter
926 * since we're not adding callbacks, but deleting them. I'm not sure
927 * why it doesn't set the lock, however; perhaps it should.
930 DeleteFileCallBacks(AFSFid * fid)
932 struct FileEntry *fe;
938 cbstuff.DeleteFiles++;
943 ("DF: No fid (%u,%u,%u) to delete\n", fid->Volume, fid->Vnode,
947 for (n = 0, cbi = fe->firstcb; cbi; n++) {
960 /* Delete (do not break) all call backs for host. The host should be
963 DeleteAllCallBacks_r(struct host *host, int deletefe)
968 cbstuff.DeleteAllCallBacks++;
969 cbi = first = host->cblist;
971 ViceLog(8, ("DV: no call backs\n"));
979 } while (cbi != first);
985 * Break all delayed call backs for host. Returns 1 if all call backs
986 * successfully broken; 0 otherwise. Assumes host is h_Held and h_Locked.
987 * Must be called with VenusDown set for this host
990 BreakDelayedCallBacks(struct host *host)
994 retVal = BreakDelayedCallBacks_r(host);
1000 BreakDelayedCallBacks_r(struct host *host)
1002 struct AFSFid fids[AFSCBMAX];
1003 int cbi, first, nfids;
1004 struct CallBack *cb;
1007 struct rx_connection *cb_conn;
1009 cbstuff.nbreakers++;
1010 if (!(host->hostFlags & RESETDONE) && !(host->hostFlags & HOSTDELETED)) {
1011 host->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1012 cb_conn = host->callback_rxcon;
1013 rx_GetConnection(cb_conn);
1014 if (host->interface) {
1017 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1020 code = RXAFSCB_InitCallBackState(cb_conn);
1022 rx_PutConnection(cb_conn);
1025 host->hostFlags |= ALTADDR; /* alternate addresses are valid */
1029 ("CB: Call back connect back failed (in break delayed) "
1030 "for Host %p (%s:%d)\n",
1031 host, afs_inet_ntoa_r(host->host, hoststr),
1032 ntohs(host->port)));
1034 host->hostFlags |= VENUSDOWN;
1037 ("InitCallBackState success on %p (%s:%d)\n",
1038 host, afs_inet_ntoa_r(host->host, hoststr),
1039 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 struct FileEntry *fe = itofe(cb->fhead);
1057 fids[nfids].Volume = fe->volid;
1058 fids[nfids].Vnode = fe->vnode;
1059 fids[nfids].Unique = fe->unique;
1065 } while (cbi && cbi != first && nfids < AFSCBMAX);
1071 if (XCallBackBulk_r(host, fids, nfids)) {
1072 /* Failed, again: put them back, probably with old
1077 ("CB: XCallBackBulk failed, Host %p (%s:%d); "
1078 "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 %p (%s:%d), file %u.%u.%u "
1086 "(part of bulk callback)\n",
1087 host, afs_inet_ntoa_r(host->host, hoststr),
1088 ntohs(host->port), fids[i].Volume,
1089 fids[i].Vnode, fids[i].Unique));
1092 * AddCallBack1_r(host, &fids[i], itot(thead[i]), CB_DELAYED, 1);
1093 * * but it turns out to cause too many tricky locking problems.
1094 * * now, if break delayed fails, screw it. */
1096 host->hostFlags |= VENUSDOWN; /* Failed */
1097 ClearHostCallbacks_r(host, 1 /* locked */ );
1101 if (nfids < AFSCBMAX)
1105 cbstuff.nbreakers--;
1106 /* If we succeeded it's always ok to unset HFE_LATER */
1107 if (!(host->hostFlags & VENUSDOWN))
1108 host->hostFlags &= ~HFE_LATER;
1109 return (host->hostFlags & VENUSDOWN);
1113 MultiBreakVolumeCallBack_r(struct host *host,
1114 struct VCBParams *parms, int deletefe)
1118 if (host->hostFlags & HOSTDELETED)
1121 if (!(host->hostFlags & HCBREAK))
1122 return 0; /* host is not flagged to notify */
1124 if (host->hostFlags & VENUSDOWN) {
1126 /* Do not care if the host is now HOSTDELETED */
1129 ("BVCB: volume callback for Host %p (%s:%d) failed\n",
1130 host, afs_inet_ntoa_r(host->host, hoststr),
1131 ntohs(host->port)));
1133 DeleteAllCallBacks_r(host, deletefe); /* Delete all callback state
1134 * rather than attempting to
1135 * selectively remember to
1136 * delete the volume callbacks
1138 host->hostFlags &= ~(RESETDONE|HCBREAK); /* Do InitCallBackState when host returns */
1142 osi_Assert(parms->ncbas <= MAX_CB_HOSTS);
1144 /* Do not call MultiBreakCallBack on the current host structure
1145 ** because it would prematurely release the hold on the host
1147 if (parms->ncbas == MAX_CB_HOSTS) {
1148 struct AFSCBFids tf;
1150 tf.AFSCBFids_len = 1;
1151 tf.AFSCBFids_val = parms->fid;
1153 /* this releases all the hosts */
1154 MultiBreakCallBack_r(parms->cba, parms->ncbas, &tf);
1158 parms->cba[parms->ncbas].hp = host;
1159 parms->cba[(parms->ncbas)++].thead = parms->thead;
1160 host->hostFlags &= ~HCBREAK;
1162 /* we have more work to do on this host, so make sure we keep a reference
1170 MultiBreakVolumeLaterCallBack(struct host *host, void *rock)
1172 struct VCBParams *parms = (struct VCBParams *)rock;
1175 retval = MultiBreakVolumeCallBack_r(host, parms, 0);
1181 * Break all call backs on a single volume. Don't call this with any
1182 * hosts h_held. Note that this routine clears the callbacks before
1183 * actually breaking them, and that the vnode isn't locked during this
1184 * operation, so that people might see temporary callback loss while
1185 * this function is executing. It is just a temporary state, however,
1186 * since the callback will be broken later by this same function.
1188 * Now uses multi-RX for CallBack RPC in a different thread,
1189 * only marking them here.
1191 extern pthread_cond_t fsync_cond;
1194 BreakVolumeCallBacksLater(afs_uint32 volume)
1198 struct FileEntry *fe;
1199 struct CallBack *cb;
1203 ViceLog(25, ("Setting later on volume %u\n", volume));
1205 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1206 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1207 if (fe->volid == volume) {
1208 struct CallBack *cbnext;
1209 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1210 host = h_itoh(cb->hhead);
1211 host->hostFlags |= HFE_LATER;
1212 cb->status = CB_DELAYED;
1213 cbnext = itocb(cb->cnext);
1216 fe->status |= FE_LATER;
1225 /* didn't find any callbacks, so return right away. */
1229 ViceLog(25, ("Fsync thread wakeup\n"));
1231 CV_BROADCAST(&fsync_cond);
1237 BreakLaterCallBacks(void)
1242 struct CallBack *cb;
1243 struct FileEntry *fe = NULL;
1244 struct FileEntry *myfe = NULL;
1246 struct VCBParams henumParms;
1247 unsigned short tthead = 0; /* zero is illegal value */
1251 ViceLog(25, ("Looking for FileEntries to unchain\n"));
1254 /* Pick the first volume we see to clean up */
1255 fid.Volume = fid.Vnode = fid.Unique = 0;
1257 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1258 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1259 if (fe && (fe->status & FE_LATER)
1260 && (fid.Volume == 0 || fid.Volume == fe->volid)) {
1261 /* Ugly, but used to avoid left side casting */
1262 struct object *tmpfe;
1264 ("Unchaining for %u:%u:%u\n", fe->vnode, fe->unique,
1266 fid.Volume = fe->volid;
1268 fe->status &= ~FE_LATER; /* not strictly needed */
1269 /* Works since volid is deeper than the largest pointer */
1270 tmpfe = (struct object *)fe;
1271 tmpfe->next = (struct object *)myfe;
1284 /* loop over FEs from myfe and free/break */
1286 for (fe = myfe; fe;) {
1287 struct CallBack *cbnext;
1288 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1289 cbnext = itocb(cb->cnext);
1290 host = h_itoh(cb->hhead);
1291 if (cb->status == CB_DELAYED) {
1292 if (!(host->hostFlags & HOSTDELETED)) {
1293 /* mark this host for notification */
1294 host->hostFlags |= HCBREAK;
1295 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1301 CDel(cb, 0); /* Don't let CDel clean up the fe */
1302 /* leave flag for MultiBreakVolumeCallBack to clear */
1305 ("Found host %p (%s:%d) non-DELAYED cb for %u:%u:%u\n",
1306 host, afs_inet_ntoa_r(host->host, hoststr),
1307 ntohs(host->port), fe->vnode, fe->unique, fe->volid));
1311 fe = (struct FileEntry *)((struct object *)fe)->next;
1316 ViceLog(125, ("Breaking volume %u\n", fid.Volume));
1317 henumParms.ncbas = 0;
1318 henumParms.fid = &fid;
1319 henumParms.thead = tthead;
1321 h_Enumerate(MultiBreakVolumeLaterCallBack, (char *)&henumParms);
1323 if (henumParms.ncbas) { /* do left-overs */
1324 struct AFSCBFids tf;
1325 tf.AFSCBFids_len = 1;
1326 tf.AFSCBFids_val = &fid;
1328 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf);
1329 henumParms.ncbas = 0;
1334 /* Arrange to be called again */
1339 * Delete all timed-out call back entries (to be called periodically by file
1343 CleanupTimedOutCallBacks(void)
1346 CleanupTimedOutCallBacks_r();
1352 CleanupTimedOutCallBacks_r(void)
1354 afs_uint32 now = CBtime(time(NULL));
1356 struct CallBack *cb;
1360 while (tfirst <= now) {
1362 cbi = *(thead = THead(tfirst));
1368 ("CCB: deleting timed out call back %x (%s:%d), (%u,%u,%u)\n",
1369 h_itoh(cb->hhead)->host,
1370 afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
1371 h_itoh(cb->hhead)->port, itofe(cb->fhead)->volid,
1372 itofe(cb->fhead)->vnode, itofe(cb->fhead)->unique));
1376 if (ntimedout > cbstuff.nblks) {
1377 ViceLog(0, ("CCB: Internal Error -- shutting down...\n"));
1378 DumpCallBackState_r();
1379 ShutDownAndCore(PANIC);
1381 } while (cbi != *thead);
1386 cbstuff.CBsTimedOut += ntimedout;
1387 ViceLog(7, ("CCB: deleted %d timed out callbacks\n", ntimedout));
1388 return (ntimedout > 0);
1392 * parameters to pass to lih*_r from h_Enumerate_r when trying to find a host
1393 * from which to clear callbacks.
1397 * Points to the least interesting host found; try to clear callbacks on
1398 * this host after h_Enumerate_r(lih*_r)'ing.
1403 * The last host we got from lih*_r, but we couldn't clear its callbacks
1404 * for some reason. Choose the next-best host after this one (with the
1405 * current lih*_r, this means to only select hosts that have an ActiveCall
1406 * newer than lastlih).
1408 struct host *lastlih;
1411 /* Value of host->refCount that allows us to reliably infer that
1412 * host may be held by some other thread */
1413 #define OTHER_MUSTHOLD_LIH 2
1415 /* This version does not allow 'host' to be selected unless its ActiveCall
1416 * is newer than 'params->lastlih' which is the host with the oldest
1417 * ActiveCall from the last pass (if it is provided). We filter out any hosts
1418 * that are are held by other threads.
1420 * There is a small problem here, but it may not be easily fixable. Say we
1421 * select some host A, and give it back to GetSomeSpace_r. GSS_r for some
1422 * reason cannot clear the callbacks on A, and so calls us again with
1423 * lastlih = A. Suppose there is another host B that has the same ActiveCall
1424 * time as A. We will now skip over host B, since
1425 * 'hostB->ActiveCall > hostA->ActiveCall' is not true. This could result in
1426 * us prematurely going to the GSS_r 2nd or 3rd pass, and making us a little
1427 * inefficient. This should be pretty rare, though, except perhaps in cases
1428 * with very small numbers of hosts.
1430 * Also filter out any hosts with HOSTDELETED set. h_Enumerate_r should in
1431 * theory not give these to us anyway, but be paranoid.
1434 lih0_r(struct host *host, void *rock)
1436 struct lih_params *params = (struct lih_params *)rock;
1438 /* OTHER_MUSTHOLD_LIH is because the h_Enum loop holds us once */
1440 && (!(host->hostFlags & HOSTDELETED))
1441 && (host->refCount < OTHER_MUSTHOLD_LIH)
1442 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1443 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1446 h_Release_r(params->lih); /* release prev host */
1455 /* same as lih0_r, except we do not prevent held hosts from being selected. */
1457 lih1_r(struct host *host, void *rock)
1459 struct lih_params *params = (struct lih_params *)rock;
1462 && (!(host->hostFlags & HOSTDELETED))
1463 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1464 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1467 h_Release_r(params->lih); /* release prev host */
1476 /* This could be upgraded to get more space each time */
1477 /* first pass: sequentially find the oldest host which isn't held by
1478 anyone for which we can clear callbacks;
1480 /* second pass: sequentially find the oldest host regardless of
1481 whether or not the host is held; skipping 'hostp' */
1482 /* third pass: attempt to clear callbacks from 'hostp' */
1483 /* always called with hostp unlocked */
1485 /* Note: hostlist is ordered most recently created host first and
1486 * its order has no relationship to the most recently used. */
1487 extern struct host *hostList;
1489 GetSomeSpace_r(struct host *hostp, int locked)
1492 struct lih_params params;
1495 if (cbstuff.GotSomeSpaces == 0) {
1496 /* only log this once; if GSS is getting called constantly, that's not
1497 * good but don't make things worse by spamming the log. */
1498 ViceLog(0, ("We have run out of callback space; forcing callback revocation. "
1499 "This suggests the fileserver is configured with insufficient "
1500 "callbacks; you probably want to increase the -cb fileserver "
1501 "parameter (current setting: %u). The fileserver will continue "
1502 "to operate, but this may indicate a severe performance problem\n",
1504 ViceLog(0, ("This message is logged at most once; for more information "
1505 "see the OpenAFS documentation and fileserver xstat collection 3\n"));
1508 cbstuff.GotSomeSpaces++;
1510 ("GSS: First looking for timed out call backs via CleanupCallBacks\n"));
1511 if (CleanupTimedOutCallBacks_r()) {
1517 params.lastlih = NULL;
1522 h_Enumerate_r(i == 0 ? lih0_r : lih1_r, hostList, ¶ms);
1525 if (params.lastlih) {
1526 h_Release_r(params.lastlih);
1527 params.lastlih = NULL;
1531 /* note that 'hp' was held by lih*_r; we will need to release it */
1533 if ((hp != hostp) && !ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
1538 params.lastlih = hp;
1539 /* params.lastlih will be released on the next iteration, after
1544 * Next time try getting callbacks from any host even if
1545 * it's held, since the only other option is starvation for
1546 * the file server (i.e. until the callback timeout arrives).
1549 params.lastlih = NULL;
1552 ("GSS: Try harder for longest inactive host cnt= %d\n",
1557 /* Could not obtain space from other hosts, clear hostp's callback state */
1562 ClearHostCallbacks_r(hostp, 1 /*already locked */ );
1569 /* locked - set if caller has already locked the host */
1571 ClearHostCallbacks_r(struct host *hp, int locked)
1575 struct rx_connection *cb_conn = NULL;
1578 ("GSS: Delete longest inactive host %p (%s:%d)\n",
1579 hp, afs_inet_ntoa_r(hp->host, hoststr), ntohs(hp->port)));
1581 if ((hp->hostFlags & HOSTDELETED)) {
1582 /* hp could go away after reacquiring H_LOCK in h_NBLock_r, so we can't
1583 * really use it; its callbacks will get cleared anyway when
1584 * h_TossStuff_r gets its hands on it */
1590 /** Try a non-blocking lock. If the lock is already held return
1591 * after releasing hold on hp
1594 if (h_NBLock_r(hp)) {
1599 if (hp->Console & 2) {
1601 * If the special console field is set it means that a thread
1602 * is waiting in AddCallBack1 after it set pointers to the
1603 * file entry and/or callback entry. Because of the bogus
1604 * usage of h_hold it won't prevent from another thread, this
1605 * one, to remove all the callbacks so just to be safe we keep
1606 * a reference. NOTE, on the last phase we'll free the calling
1607 * host's callbacks but that's ok...
1611 DeleteAllCallBacks_r(hp, 1);
1612 if (hp->hostFlags & VENUSDOWN) {
1613 hp->hostFlags &= ~RESETDONE; /* remember that we must do a reset */
1614 } else if (!(hp->hostFlags & HOSTDELETED)) {
1615 /* host is up, try a call */
1616 hp->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1617 cb_conn = hp->callback_rxcon;
1618 rx_GetConnection(hp->callback_rxcon);
1619 if (hp->interface) {
1622 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1625 code = RXAFSCB_InitCallBackState(cb_conn);
1627 rx_PutConnection(cb_conn);
1630 hp->hostFlags |= ALTADDR; /* alternate addresses are valid */
1632 /* failed, mark host down and need reset */
1633 hp->hostFlags |= VENUSDOWN;
1634 hp->hostFlags &= ~RESETDONE;
1636 /* reset succeeded, we're done */
1637 hp->hostFlags |= RESETDONE;
1646 #endif /* INTERPRET_DUMP */
1650 PrintCallBackStats(void)
1653 "%d add CB, %d break CB, %d del CB, %d del FE, %d CB's timed out, %d space reclaim, %d del host\n",
1654 cbstuff.AddCallBacks, cbstuff.BreakCallBacks,
1655 cbstuff.DeleteCallBacks, cbstuff.DeleteFiles, cbstuff.CBsTimedOut,
1656 cbstuff.GotSomeSpaces, cbstuff.DeleteAllCallBacks);
1657 fprintf(stderr, "%d CBs, %d FEs, (%d of total of %d 16-byte blocks)\n",
1658 cbstuff.nCBs, cbstuff.nFEs, cbstuff.nCBs + cbstuff.nFEs,
1660 fprintf(stderr, "%d GSS1, %d GSS2, %d GSS3, %d GSS4, %d GSS5 (internal counters)\n",
1661 cbstuff.GSS1, cbstuff.GSS2, cbstuff.GSS3, cbstuff.GSS4, cbstuff.GSS5);
1666 #define MAGIC 0x12345678 /* To check byte ordering of dump when it is read in */
1667 #define MAGICV2 0x12345679 /* To check byte ordering & version of dump when it is read in */
1670 #ifndef INTERPRET_DUMP
1672 #ifdef AFS_DEMAND_ATTACH_FS
1675 * callback state serialization
1677 static int cb_stateSaveTimeouts(struct fs_dump_state * state);
1678 static int cb_stateSaveFEHash(struct fs_dump_state * state);
1679 static int cb_stateSaveFEs(struct fs_dump_state * state);
1680 static int cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe);
1681 static int cb_stateRestoreTimeouts(struct fs_dump_state * state);
1682 static int cb_stateRestoreFEHash(struct fs_dump_state * state);
1683 static int cb_stateRestoreFEs(struct fs_dump_state * state);
1684 static int cb_stateRestoreFE(struct fs_dump_state * state);
1685 static int cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
1686 struct iovec * iov, int niovecs);
1688 static int cb_stateVerifyFEHash(struct fs_dump_state * state);
1689 static int cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe);
1690 static int cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe);
1691 static int cb_stateVerifyTimeoutQueues(struct fs_dump_state * state);
1693 static int cb_stateFEToDiskEntry(struct FileEntry *, struct FEDiskEntry *);
1694 static int cb_stateDiskEntryToFE(struct fs_dump_state * state,
1695 struct FEDiskEntry *, struct FileEntry *);
1697 static int cb_stateCBToDiskEntry(struct CallBack *, struct CBDiskEntry *);
1698 static int cb_stateDiskEntryToCB(struct fs_dump_state * state,
1699 struct CBDiskEntry *, struct CallBack *);
1701 static int cb_stateFillHeader(struct callback_state_header * hdr);
1702 static int cb_stateCheckHeader(struct callback_state_header * hdr);
1704 static int cb_stateAllocMap(struct fs_dump_state * state);
1707 cb_stateSave(struct fs_dump_state * state)
1711 AssignInt64(state->eof_offset, &state->hdr->cb_offset);
1713 /* invalidate callback state header */
1714 memset(state->cb_hdr, 0, sizeof(struct callback_state_header));
1715 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1716 sizeof(struct callback_state_header))) {
1721 fs_stateIncEOF(state, sizeof(struct callback_state_header));
1723 /* dump timeout state */
1724 if (cb_stateSaveTimeouts(state)) {
1729 /* dump fe hashtable state */
1730 if (cb_stateSaveFEHash(state)) {
1735 /* dump callback state */
1736 if (cb_stateSaveFEs(state)) {
1741 /* write the callback state header to disk */
1742 cb_stateFillHeader(state->cb_hdr);
1743 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1744 sizeof(struct callback_state_header))) {
1754 cb_stateRestore(struct fs_dump_state * state)
1758 if (fs_stateReadHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1759 sizeof(struct callback_state_header))) {
1764 if (cb_stateCheckHeader(state->cb_hdr)) {
1769 if (cb_stateAllocMap(state)) {
1774 if (cb_stateRestoreTimeouts(state)) {
1779 if (cb_stateRestoreFEHash(state)) {
1784 /* restore FEs and CBs from disk */
1785 if (cb_stateRestoreFEs(state)) {
1790 /* restore the timeout queue heads */
1791 tfirst = state->cb_hdr->tfirst;
1798 cb_stateRestoreIndices(struct fs_dump_state * state)
1801 struct FileEntry * fe;
1802 struct CallBack * cb;
1804 /* restore indices in the FileEntry structures */
1805 for (i = 1; i < state->fe_map.len; i++) {
1806 if (state->fe_map.entries[i].new_idx) {
1807 fe = itofe(state->fe_map.entries[i].new_idx);
1809 /* restore the fe->fnext entry */
1810 if (fe_OldToNew(state, fe->fnext, &fe->fnext)) {
1815 /* restore the fe->firstcb entry */
1816 if (cb_OldToNew(state, fe->firstcb, &fe->firstcb)) {
1823 /* restore indices in the CallBack structures */
1824 for (i = 1; i < state->cb_map.len; i++) {
1825 if (state->cb_map.entries[i].new_idx) {
1826 cb = itocb(state->cb_map.entries[i].new_idx);
1828 /* restore the cb->cnext entry */
1829 if (cb_OldToNew(state, cb->cnext, &cb->cnext)) {
1834 /* restore the cb->fhead entry */
1835 if (fe_OldToNew(state, cb->fhead, &cb->fhead)) {
1840 /* restore the cb->hhead entry */
1841 if (h_OldToNew(state, cb->hhead, &cb->hhead)) {
1846 /* restore the cb->tprev entry */
1847 if (cb_OldToNew(state, cb->tprev, &cb->tprev)) {
1852 /* restore the cb->tnext entry */
1853 if (cb_OldToNew(state, cb->tnext, &cb->tnext)) {
1858 /* restore the cb->hprev entry */
1859 if (cb_OldToNew(state, cb->hprev, &cb->hprev)) {
1864 /* restore the cb->hnext entry */
1865 if (cb_OldToNew(state, cb->hnext, &cb->hnext)) {
1872 /* restore the timeout queue head indices */
1873 for (i = 0; i < state->cb_timeout_hdr->records; i++) {
1874 if (cb_OldToNew(state, timeout[i], &timeout[i])) {
1880 /* restore the FE hash table queue heads */
1881 for (i = 0; i < state->cb_fehash_hdr->records; i++) {
1882 if (fe_OldToNew(state, HashTable[i], &HashTable[i])) {
1893 cb_stateVerify(struct fs_dump_state * state)
1897 if (cb_stateVerifyFEHash(state)) {
1901 if (cb_stateVerifyTimeoutQueues(state)) {
1909 cb_stateVerifyFEHash(struct fs_dump_state * state)
1912 struct FileEntry * fe;
1913 afs_uint32 fei, chain_len;
1915 for (i = 0; i < FEHASH_SIZE; i++) {
1917 for (fei = HashTable[i], fe = itofe(fei);
1919 fei = fe->fnext, fe = itofe(fei)) {
1920 if (fei > cbstuff.nblks) {
1921 ViceLog(0, ("cb_stateVerifyFEHash: error: index out of range (fei=%d)\n", fei));
1925 if (cb_stateVerifyFE(state, fe)) {
1928 if (chain_len > FS_STATE_FE_MAX_HASH_CHAIN_LEN) {
1929 ViceLog(0, ("cb_stateVerifyFEHash: error: hash chain %d length exceeds %d; assuming there's a loop\n",
1930 i, FS_STATE_FE_MAX_HASH_CHAIN_LEN));
1942 cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe)
1946 if ((fe->firstcb && !fe->ncbs) ||
1947 (!fe->firstcb && fe->ncbs)) {
1948 ViceLog(0, ("cb_stateVerifyFE: error: fe->firstcb does not agree with fe->ncbs (fei=%lu, fe->firstcb=%lu, fe->ncbs=%lu)\n",
1949 afs_printable_uint32_lu(fetoi(fe)),
1950 afs_printable_uint32_lu(fe->firstcb),
1951 afs_printable_uint32_lu(fe->ncbs)));
1954 if (cb_stateVerifyFCBList(state, fe)) {
1955 ViceLog(0, ("cb_stateVerifyFE: error: FCBList failed verification (fei=%lu)\n",
1956 afs_printable_uint32_lu(fetoi(fe))));
1964 cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe)
1967 afs_uint32 cbi, fei, chain_len = 0;
1968 struct CallBack * cb;
1972 for (cbi = fe->firstcb, cb = itocb(cbi);
1974 cbi = cb->cnext, cb = itocb(cbi)) {
1975 if (cbi > cbstuff.nblks) {
1976 ViceLog(0, ("cb_stateVerifyFCBList: error: list index out of range (cbi=%d, ncbs=%d)\n",
1977 cbi, cbstuff.nblks));
1981 if (cb->fhead != fei) {
1982 ViceLog(0, ("cb_stateVerifyFCBList: error: cb->fhead != fei (fei=%d, cb->fhead=%d)\n",
1986 if (chain_len > FS_STATE_FCB_MAX_LIST_LEN) {
1987 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (fei=%d); assuming there's a loop\n",
1988 FS_STATE_FCB_MAX_LIST_LEN, fei));
1995 if (fe->ncbs != chain_len) {
1996 ViceLog(0, ("cb_stateVerifyFCBList: error: list length mismatch (len=%d, fe->ncbs=%d)\n",
1997 chain_len, fe->ncbs));
2006 cb_stateVerifyHCBList(struct fs_dump_state * state, struct host * host)
2009 afs_uint32 hi, chain_len, cbi;
2010 struct CallBack *cb, *ncb;
2015 for (cbi = host->cblist, cb = itocb(cbi);
2017 cbi = cb->hnext, cb = ncb) {
2018 if (chain_len && (host->cblist == cbi)) {
2019 /* we've wrapped around the circular list, and everything looks ok */
2022 if (cb->hhead != hi) {
2023 ViceLog(0, ("cb_stateVerifyHCBList: error: incorrect cb->hhead (cbi=%d, h->index=%d, cb->hhead=%d)\n",
2024 cbi, hi, cb->hhead));
2027 if (!cb->hprev || !cb->hnext) {
2028 ViceLog(0, ("cb_stateVerifyHCBList: error: null index in circular list (cbi=%d, h->index=%d)\n",
2033 if ((cb->hprev > cbstuff.nblks) ||
2034 (cb->hnext > cbstuff.nblks)) {
2035 ViceLog(0, ("cb_stateVerifyHCBList: error: list index out of range (cbi=%d, h->index=%d, cb->hprev=%d, cb->hnext=%d, nCBs=%d)\n",
2036 cbi, hi, cb->hprev, cb->hnext, cbstuff.nblks));
2040 ncb = itocb(cb->hnext);
2041 if (cbi != ncb->hprev) {
2042 ViceLog(0, ("cb_stateVerifyHCBList: error: corrupt linked list (cbi=%d, h->index=%d)\n",
2047 if (chain_len > FS_STATE_HCB_MAX_LIST_LEN) {
2048 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (h->index=%d); assuming there's a loop\n",
2049 FS_STATE_HCB_MAX_LIST_LEN, hi));
2061 cb_stateVerifyTimeoutQueues(struct fs_dump_state * state)
2064 afs_uint32 cbi, chain_len;
2065 struct CallBack *cb, *ncb;
2067 for (i = 0; i < CB_NUM_TIMEOUT_QUEUES; i++) {
2069 for (cbi = timeout[i], cb = itocb(cbi);
2071 cbi = cb->tnext, cb = ncb) {
2072 if (chain_len && (cbi == timeout[i])) {
2073 /* we've wrapped around the circular list, and everything looks ok */
2076 if (cbi > cbstuff.nblks) {
2077 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: list index out of range (cbi=%d, tindex=%d)\n",
2082 if (itot(cb->thead) != &timeout[i]) {
2083 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: cb->thead points to wrong timeout queue (tindex=%d, cbi=%d, cb->thead=%d)\n",
2084 i, cbi, cb->thead));
2087 if (!cb->tprev || !cb->tnext) {
2088 ViceLog(0, ("cb_stateVerifyTimeoutQueues: null index in circular list (cbi=%d, tindex=%d)\n",
2093 if ((cb->tprev > cbstuff.nblks) ||
2094 (cb->tnext > cbstuff.nblks)) {
2095 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list index out of range (cbi=%d, tindex=%d, cb->tprev=%d, cb->tnext=%d, nCBs=%d)\n",
2096 cbi, i, cb->tprev, cb->tnext, cbstuff.nblks));
2100 ncb = itocb(cb->tnext);
2101 if (cbi != ncb->tprev) {
2102 ViceLog(0, ("cb_stateVerifyTimeoutQueues: corrupt linked list (cbi=%d, tindex=%d)\n",
2107 if (chain_len > FS_STATE_TCB_MAX_LIST_LEN) {
2108 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list length exceeds %d (tindex=%d); assuming there's a loop\n",
2109 FS_STATE_TCB_MAX_LIST_LEN, i));
2121 cb_stateSaveTimeouts(struct fs_dump_state * state)
2124 struct iovec iov[2];
2126 AssignInt64(state->eof_offset, &state->cb_hdr->timeout_offset);
2128 memset(state->cb_timeout_hdr, 0, sizeof(struct callback_state_fehash_header));
2129 state->cb_timeout_hdr->magic = CALLBACK_STATE_TIMEOUT_MAGIC;
2130 state->cb_timeout_hdr->records = CB_NUM_TIMEOUT_QUEUES;
2131 state->cb_timeout_hdr->len = sizeof(struct callback_state_timeout_header) +
2132 (state->cb_timeout_hdr->records * sizeof(afs_uint32));
2134 iov[0].iov_base = (char *)state->cb_timeout_hdr;
2135 iov[0].iov_len = sizeof(struct callback_state_timeout_header);
2136 iov[1].iov_base = (char *)timeout;
2137 iov[1].iov_len = sizeof(timeout);
2139 if (fs_stateSeek(state, &state->cb_hdr->timeout_offset)) {
2144 if (fs_stateWriteV(state, iov, 2)) {
2149 fs_stateIncEOF(state, state->cb_timeout_hdr->len);
2156 cb_stateRestoreTimeouts(struct fs_dump_state * state)
2160 if (fs_stateReadHeader(state, &state->cb_hdr->timeout_offset,
2161 state->cb_timeout_hdr,
2162 sizeof(struct callback_state_timeout_header))) {
2167 if (state->cb_timeout_hdr->magic != CALLBACK_STATE_TIMEOUT_MAGIC) {
2171 if (state->cb_timeout_hdr->records != CB_NUM_TIMEOUT_QUEUES) {
2176 len = state->cb_timeout_hdr->records * sizeof(afs_uint32);
2178 if (state->cb_timeout_hdr->len !=
2179 (sizeof(struct callback_state_timeout_header) + len)) {
2184 if (fs_stateRead(state, timeout, len)) {
2194 cb_stateSaveFEHash(struct fs_dump_state * state)
2197 struct iovec iov[2];
2199 AssignInt64(state->eof_offset, &state->cb_hdr->fehash_offset);
2201 memset(state->cb_fehash_hdr, 0, sizeof(struct callback_state_fehash_header));
2202 state->cb_fehash_hdr->magic = CALLBACK_STATE_FEHASH_MAGIC;
2203 state->cb_fehash_hdr->records = FEHASH_SIZE;
2204 state->cb_fehash_hdr->len = sizeof(struct callback_state_fehash_header) +
2205 (state->cb_fehash_hdr->records * sizeof(afs_uint32));
2207 iov[0].iov_base = (char *)state->cb_fehash_hdr;
2208 iov[0].iov_len = sizeof(struct callback_state_fehash_header);
2209 iov[1].iov_base = (char *)HashTable;
2210 iov[1].iov_len = sizeof(HashTable);
2212 if (fs_stateSeek(state, &state->cb_hdr->fehash_offset)) {
2217 if (fs_stateWriteV(state, iov, 2)) {
2222 fs_stateIncEOF(state, state->cb_fehash_hdr->len);
2229 cb_stateRestoreFEHash(struct fs_dump_state * state)
2233 if (fs_stateReadHeader(state, &state->cb_hdr->fehash_offset,
2234 state->cb_fehash_hdr,
2235 sizeof(struct callback_state_fehash_header))) {
2240 if (state->cb_fehash_hdr->magic != CALLBACK_STATE_FEHASH_MAGIC) {
2244 if (state->cb_fehash_hdr->records != FEHASH_SIZE) {
2249 len = state->cb_fehash_hdr->records * sizeof(afs_uint32);
2251 if (state->cb_fehash_hdr->len !=
2252 (sizeof(struct callback_state_fehash_header) + len)) {
2257 if (fs_stateRead(state, HashTable, len)) {
2267 cb_stateSaveFEs(struct fs_dump_state * state)
2271 struct FileEntry *fe;
2273 AssignInt64(state->eof_offset, &state->cb_hdr->fe_offset);
2275 for (hash = 0; hash < FEHASH_SIZE ; hash++) {
2276 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
2278 if (cb_stateSaveFE(state, fe)) {
2290 cb_stateRestoreFEs(struct fs_dump_state * state)
2292 int count, nFEs, ret = 0;
2294 nFEs = state->cb_hdr->nFEs;
2296 for (count = 0; count < nFEs; count++) {
2297 if (cb_stateRestoreFE(state)) {
2308 cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe)
2310 int ret = 0, iovcnt, cbi, written = 0;
2312 struct callback_state_entry_header hdr;
2313 struct FEDiskEntry fedsk;
2314 struct CBDiskEntry cbdsk[16];
2315 struct iovec iov[16];
2316 struct CallBack *cb;
2319 if (fei > state->cb_hdr->fe_max) {
2320 state->cb_hdr->fe_max = fei;
2323 memset(&hdr, 0, sizeof(struct callback_state_entry_header));
2325 if (cb_stateFEToDiskEntry(fe, &fedsk)) {
2330 iov[0].iov_base = (char *)&hdr;
2331 iov[0].iov_len = sizeof(hdr);
2332 iov[1].iov_base = (char *)&fedsk;
2333 iov[1].iov_len = sizeof(struct FEDiskEntry);
2336 for (cbi = fe->firstcb, cb = itocb(cbi);
2338 cbi = cb->cnext, cb = itocb(cbi), hdr.nCBs++) {
2339 if (cbi > state->cb_hdr->cb_max) {
2340 state->cb_hdr->cb_max = cbi;
2342 if (cb_stateCBToDiskEntry(cb, &cbdsk[iovcnt])) {
2346 cbdsk[iovcnt].index = cbi;
2347 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2348 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2350 if ((iovcnt == 16) || (!cb->cnext)) {
2351 if (fs_stateWriteV(state, iov, iovcnt)) {
2360 hdr.magic = CALLBACK_STATE_ENTRY_MAGIC;
2361 hdr.len = sizeof(hdr) + sizeof(struct FEDiskEntry) +
2362 (hdr.nCBs * sizeof(struct CBDiskEntry));
2365 if (fs_stateWriteV(state, iov, iovcnt)) {
2370 if (fs_stateWriteHeader(state, &state->eof_offset, &hdr, sizeof(hdr))) {
2376 fs_stateIncEOF(state, hdr.len);
2379 if (fs_stateSeek(state, &state->eof_offset)) {
2385 state->cb_hdr->nFEs++;
2386 state->cb_hdr->nCBs += hdr.nCBs;
2393 cb_stateRestoreFE(struct fs_dump_state * state)
2395 int ret = 0, iovcnt, nCBs;
2396 struct callback_state_entry_header hdr;
2397 struct FEDiskEntry fedsk;
2398 struct CBDiskEntry cbdsk[16];
2399 struct iovec iov[16];
2400 struct FileEntry * fe;
2402 iov[0].iov_base = (char *)&hdr;
2403 iov[0].iov_len = sizeof(hdr);
2404 iov[1].iov_base = (char *)&fedsk;
2405 iov[1].iov_len = sizeof(fedsk);
2408 if (fs_stateReadV(state, iov, iovcnt)) {
2413 if (hdr.magic != CALLBACK_STATE_ENTRY_MAGIC) {
2420 ViceLog(0, ("cb_stateRestoreFE: ran out of free FileEntry structures\n"));
2425 if (cb_stateDiskEntryToFE(state, &fedsk, fe)) {
2431 for (iovcnt = 0, nCBs = 0;
2434 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2435 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2437 if ((iovcnt == 16) || (nCBs == hdr.nCBs - 1)) {
2438 if (fs_stateReadV(state, iov, iovcnt)) {
2442 if (cb_stateRestoreCBs(state, fe, iov, iovcnt)) {
2456 cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
2457 struct iovec * iov, int niovecs)
2460 struct CallBack * cb;
2461 struct CBDiskEntry * cbdsk;
2463 for (idx = 0; idx < niovecs; idx++) {
2464 cbdsk = (struct CBDiskEntry *) iov[idx].iov_base;
2466 if (cbdsk->cb.hhead < state->h_map.len &&
2467 state->h_map.entries[cbdsk->cb.hhead].valid == FS_STATE_IDX_SKIPPED) {
2471 if ((cb = GetCB()) == NULL) {
2472 ViceLog(0, ("cb_stateRestoreCBs: ran out of free CallBack structures\n"));
2476 if (cb_stateDiskEntryToCB(state, cbdsk, cb)) {
2477 ViceLog(0, ("cb_stateRestoreCBs: corrupt CallBack disk entry\n"));
2489 cb_stateFillHeader(struct callback_state_header * hdr)
2491 hdr->stamp.magic = CALLBACK_STATE_MAGIC;
2492 hdr->stamp.version = CALLBACK_STATE_VERSION;
2493 hdr->tfirst = tfirst;
2498 cb_stateCheckHeader(struct callback_state_header * hdr)
2502 if (hdr->stamp.magic != CALLBACK_STATE_MAGIC) {
2504 } else if (hdr->stamp.version != CALLBACK_STATE_VERSION) {
2506 } else if ((hdr->nFEs > cbstuff.nblks) || (hdr->nCBs > cbstuff.nblks)) {
2507 ViceLog(0, ("cb_stateCheckHeader: saved callback state larger than callback memory allocation\n"));
2513 /* disk entry conversion routines */
2515 cb_stateFEToDiskEntry(struct FileEntry * in, struct FEDiskEntry * out)
2517 memcpy(&out->fe, in, sizeof(struct FileEntry));
2518 out->index = fetoi(in);
2523 cb_stateDiskEntryToFE(struct fs_dump_state * state,
2524 struct FEDiskEntry * in, struct FileEntry * out)
2528 memcpy(out, &in->fe, sizeof(struct FileEntry));
2530 /* setup FE map entry */
2531 if (!in->index || (in->index >= state->fe_map.len)) {
2532 ViceLog(0, ("cb_stateDiskEntryToFE: index (%d) out of range",
2537 state->fe_map.entries[in->index].valid = FS_STATE_IDX_VALID;
2538 state->fe_map.entries[in->index].old_idx = in->index;
2539 state->fe_map.entries[in->index].new_idx = fetoi(out);
2546 cb_stateCBToDiskEntry(struct CallBack * in, struct CBDiskEntry * out)
2548 memcpy(&out->cb, in, sizeof(struct CallBack));
2549 out->index = cbtoi(in);
2554 cb_stateDiskEntryToCB(struct fs_dump_state * state,
2555 struct CBDiskEntry * in, struct CallBack * out)
2559 memcpy(out, &in->cb, sizeof(struct CallBack));
2561 /* setup CB map entry */
2562 if (!in->index || (in->index >= state->cb_map.len)) {
2563 ViceLog(0, ("cb_stateDiskEntryToCB: index (%d) out of range\n",
2568 state->cb_map.entries[in->index].valid = FS_STATE_IDX_VALID;
2569 state->cb_map.entries[in->index].old_idx = in->index;
2570 state->cb_map.entries[in->index].new_idx = cbtoi(out);
2576 /* index map routines */
2578 cb_stateAllocMap(struct fs_dump_state * state)
2580 state->fe_map.len = state->cb_hdr->fe_max + 1;
2581 state->cb_map.len = state->cb_hdr->cb_max + 1;
2582 state->fe_map.entries = (struct idx_map_entry_t *)
2583 calloc(state->fe_map.len, sizeof(struct idx_map_entry_t));
2584 state->cb_map.entries = (struct idx_map_entry_t *)
2585 calloc(state->cb_map.len, sizeof(struct idx_map_entry_t));
2586 return ((state->fe_map.entries != NULL) && (state->cb_map.entries != NULL)) ? 0 : 1;
2590 fe_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2594 /* FEs use a one-based indexing system, so old==0 implies no mapping */
2600 if (old >= state->fe_map.len) {
2601 ViceLog(0, ("fe_OldToNew: index %d is out of range\n", old));
2603 } else if (state->fe_map.entries[old].valid != FS_STATE_IDX_VALID ||
2604 state->fe_map.entries[old].old_idx != old) { /* sanity check */
2605 ViceLog(0, ("fe_OldToNew: index %d points to an invalid FileEntry record\n", old));
2608 *new = state->fe_map.entries[old].new_idx;
2616 cb_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2620 /* CBs use a one-based indexing system, so old==0 implies no mapping */
2626 if (old >= state->cb_map.len) {
2627 ViceLog(0, ("cb_OldToNew: index %d is out of range\n", old));
2629 } else if (state->cb_map.entries[old].valid != FS_STATE_IDX_VALID ||
2630 state->cb_map.entries[old].old_idx != old) { /* sanity check */
2631 ViceLog(0, ("cb_OldToNew: index %d points to an invalid CallBack record\n", old));
2634 *new = state->cb_map.entries[old].new_idx;
2640 #endif /* AFS_DEMAND_ATTACH_FS */
2643 DumpCallBackState_r(void)
2646 afs_uint32 magic = MAGICV2, now = (afs_int32) time(NULL), freelisthead;
2648 oflag = O_WRONLY | O_CREAT | O_TRUNC;
2652 fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, oflag, 0666);
2655 ("Couldn't create callback dump file %s\n",
2656 AFSDIR_SERVER_CBKDUMP_FILEPATH));
2659 (void)write(fd, &magic, sizeof(magic));
2660 (void)write(fd, &now, sizeof(now));
2661 (void)write(fd, &cbstuff, sizeof(cbstuff));
2662 (void)write(fd, TimeOuts, sizeof(TimeOuts));
2663 (void)write(fd, timeout, sizeof(timeout));
2664 (void)write(fd, &tfirst, sizeof(tfirst));
2665 freelisthead = cbtoi((struct CallBack *)CBfree);
2666 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2667 freelisthead = fetoi((struct FileEntry *)FEfree);
2668 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2669 (void)write(fd, HashTable, sizeof(HashTable));
2670 (void)write(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2671 (void)write(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2678 DumpCallBackState(void) {
2682 rc = DumpCallBackState_r();
2688 #endif /* !INTERPRET_DUMP */
2690 #ifdef INTERPRET_DUMP
2692 /* This is only compiled in for the callback analyzer program */
2693 /* Returns the time of the dump */
2695 ReadDump(char *file, int timebits)
2698 afs_uint32 magic, freelisthead;
2706 fd = open(file, oflag);
2708 fprintf(stderr, "Couldn't read dump file %s\n", file);
2711 read(fd, &magic, sizeof(magic));
2712 if (magic == MAGICV2) {
2715 if (magic != MAGIC) {
2717 "Magic number of %s is invalid. You might be trying to\n",
2720 "run this program on a machine type with a different byte ordering.\n");
2724 if (timebits == 64) {
2725 read(fd, &now64, sizeof(afs_int64));
2726 now = (afs_int32) now64;
2728 read(fd, &now, sizeof(afs_int32));
2730 read(fd, &cbstuff, sizeof(cbstuff));
2731 read(fd, TimeOuts, sizeof(TimeOuts));
2732 read(fd, timeout, sizeof(timeout));
2733 read(fd, &tfirst, sizeof(tfirst));
2734 read(fd, &freelisthead, sizeof(freelisthead));
2735 CB = ((struct CallBack
2736 *)(calloc(cbstuff.nblks, sizeof(struct CallBack)))) - 1;
2737 FE = ((struct FileEntry
2738 *)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
2739 CBfree = (struct CallBack *)itocb(freelisthead);
2740 read(fd, &freelisthead, sizeof(freelisthead));
2741 FEfree = (struct FileEntry *)itofe(freelisthead);
2742 read(fd, HashTable, sizeof(HashTable));
2743 read(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2744 read(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2746 perror("Error reading dumpfile");
2753 #include "AFS_component_version_number.h"
2755 #include "AFS_component_version_number.c"
2758 static afs_uint32 *cbTrack;
2761 main(int argc, char **argv)
2763 int err = 0, cbi = 0, stats = 0, noptions = 0, all = 0, vol = 0, raw = 0;
2765 struct FileEntry *fe;
2766 struct CallBack *cb;
2770 memset(&fid, 0, sizeof(fid));
2773 while (argc && **argv == '-') {
2776 if (!strcmp(*argv, "-host")) {
2782 cbi = atoi(*++argv);
2783 } else if (!strcmp(*argv, "-fid")) {
2789 fid.Volume = atoi(*++argv);
2790 fid.Vnode = atoi(*++argv);
2791 fid.Unique = atoi(*++argv);
2792 } else if (!strcmp(*argv, "-time")) {
2793 fprintf(stderr, "-time not supported\n");
2795 } else if (!strcmp(*argv, "-stats")) {
2797 } else if (!strcmp(*argv, "-all")) {
2799 } else if (!strcmp(*argv, "-raw")) {
2801 } else if (!strcmp(*argv, "-timebits")) {
2807 timebits = atoi(*++argv);
2808 if ((timebits != 32)
2812 } else if (!strcmp(*argv, "-volume")) {
2818 vol = atoi(*++argv);
2823 if (err || argc != 1) {
2825 "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] [-timebits 32"
2827 "] callbackdumpfile\n");
2829 "[cbid is shown for each host in the hosts.dump file]\n");
2832 now = ReadDump(*argv, timebits);
2833 if (stats || noptions == 0) {
2834 time_t uxtfirst = UXtime(tfirst), tnow = now;
2835 printf("The time of the dump was %u %s", (unsigned int) now, ctime(&tnow));
2836 printf("The last time cleanup ran was %u %s", (unsigned int) uxtfirst,
2838 PrintCallBackStats();
2841 cbTrack = calloc(cbstuff.nblks, sizeof(cbTrack[0]));
2846 struct CallBack *cb;
2847 struct FileEntry *fe;
2849 for (hash = 0; hash < FEHASH_SIZE; hash++) {
2850 for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
2851 if (!vol || (fe->volid == vol)) {
2852 afs_uint32 fe_i = fetoi(fe);
2854 for (cb = itocb(fe->firstcb); cb; cb = itocb(cb->cnext)) {
2855 afs_uint32 cb_i = cbtoi(cb);
2857 if (cb_i > cbstuff.nblks) {
2858 printf("CB index out of range (%u > %d), stopped for this FE\n",
2859 cb_i, cbstuff.nblks);
2863 if (cbTrack[cb_i]) {
2864 printf("CB entry already claimed for FE[%u] (this is FE[%u]), stopped\n",
2865 cbTrack[cb_i], fe_i);
2868 cbTrack[cb_i] = fe_i;
2880 afs_uint32 cfirst = cbi;
2885 } while (cbi != cfirst);
2890 printf("No callback entries for %u.%u\n", fid.Volume, fid.Vnode);
2893 cb = itocb(fe->firstcb);
2896 cb = itocb(cb->cnext);
2901 for (i = 1; i < cbstuff.nblks; i++) {
2902 p = (afs_int32 *) & FE[i];
2903 printf("%d:%12x%12x%12x%12x\n", i, p[0], p[1], p[2], p[3]);
2912 PrintCB(struct CallBack *cb, afs_uint32 now)
2914 struct FileEntry *fe = itofe(cb->fhead);
2915 time_t expires = TIndexToTime(cb->thead);
2920 printf("vol=%u vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %lu secs at %s",
2921 fe->volid, fe->vnode, fe->ncbs, cb->hhead, cb->status, fe->status,
2922 expires - now, ctime(&expires));
2927 #if !defined(INTERPRET_DUMP)
2929 ** try breaking calbacks on afidp from host. Use multi_rx.
2930 ** return 0 on success, non-zero on failure
2933 MultiBreakCallBackAlternateAddress(struct host *host, struct AFSCBFids *afidp)
2937 retVal = MultiBreakCallBackAlternateAddress_r(host, afidp);
2943 MultiBreakCallBackAlternateAddress_r(struct host *host,
2944 struct AFSCBFids *afidp)
2947 struct rx_connection **conns;
2948 struct rx_connection *connSuccess = 0;
2949 struct AddrPort *interfaces;
2950 static struct rx_securityClass *sc = 0;
2951 static struct AFSCBs tc = { 0, 0 };
2954 /* nothing more can be done */
2955 if (!host->interface)
2956 return 1; /* failure */
2958 /* the only address is the primary interface */
2959 if (host->interface->numberOfInterfaces <= 1)
2960 return 1; /* failure */
2962 /* initialise a security object only once */
2964 sc = rxnull_NewClientSecurityObject();
2966 i = host->interface->numberOfInterfaces;
2967 interfaces = calloc(i, sizeof(struct AddrPort));
2968 conns = calloc(i, sizeof(struct rx_connection *));
2969 if (!interfaces || !conns) {
2970 ViceLogThenPanic(0, ("Failed malloc in "
2971 "MultiBreakCallBackAlternateAddress_r\n"));
2974 /* initialize alternate rx connections */
2975 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
2976 /* this is the current primary address */
2977 if (host->host == host->interface->interface[i].addr &&
2978 host->port == host->interface->interface[i].port)
2981 interfaces[j] = host->interface->interface[i];
2983 rx_NewConnection(interfaces[j].addr,
2984 interfaces[j].port, 1, sc, 0);
2985 rx_SetConnDeadTime(conns[j], 2);
2986 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
2990 osi_Assert(j); /* at least one alternate address */
2992 ("Starting multibreakcall back on all addr for host %p (%s:%d)\n",
2993 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
2995 multi_Rx(conns, j) {
2996 multi_RXAFSCB_CallBack(afidp, &tc);
3000 if (host->callback_rxcon)
3001 rx_DestroyConnection(host->callback_rxcon);
3002 host->callback_rxcon = conns[multi_i];
3003 /* add then remove */
3004 addInterfaceAddr_r(host, interfaces[multi_i].addr,
3005 interfaces[multi_i].port);
3006 removeInterfaceAddr_r(host, host->host, host->port);
3007 host->host = interfaces[multi_i].addr;
3008 host->port = interfaces[multi_i].port;
3009 connSuccess = conns[multi_i];
3010 rx_SetConnDeadTime(host->callback_rxcon, 50);
3011 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3013 ("multibreakcall success with addr %s:%d\n",
3014 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3015 ntohs(interfaces[multi_i].port)));
3022 /* Destroy all connections except the one on which we succeeded */
3023 for (i = 0; i < j; i++)
3024 if (conns[i] != connSuccess)
3025 rx_DestroyConnection(conns[i]);
3031 return 0; /* success */
3033 return 1; /* failure */
3038 ** try multi_RX probes to host.
3039 ** return 0 on success, non-0 on failure
3042 MultiProbeAlternateAddress_r(struct host *host)
3045 struct rx_connection **conns;
3046 struct rx_connection *connSuccess = 0;
3047 struct AddrPort *interfaces;
3048 static struct rx_securityClass *sc = 0;
3051 /* nothing more can be done */
3052 if (!host->interface)
3053 return 1; /* failure */
3055 /* the only address is the primary interface */
3056 if (host->interface->numberOfInterfaces <= 1)
3057 return 1; /* failure */
3059 /* initialise a security object only once */
3061 sc = rxnull_NewClientSecurityObject();
3063 i = host->interface->numberOfInterfaces;
3064 interfaces = calloc(i, sizeof(struct AddrPort));
3065 conns = calloc(i, sizeof(struct rx_connection *));
3066 if (!interfaces || !conns) {
3067 ViceLogThenPanic(0, ("Failed malloc in "
3068 "MultiProbeAlternateAddress_r\n"));
3071 /* initialize alternate rx connections */
3072 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3073 /* this is the current primary address */
3074 if (host->host == host->interface->interface[i].addr &&
3075 host->port == host->interface->interface[i].port)
3078 interfaces[j] = host->interface->interface[i];
3080 rx_NewConnection(interfaces[j].addr,
3081 interfaces[j].port, 1, sc, 0);
3082 rx_SetConnDeadTime(conns[j], 2);
3083 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3087 osi_Assert(j); /* at least one alternate address */
3089 ("Starting multiprobe on all addr for host %p (%s:%d)\n",
3090 host, afs_inet_ntoa_r(host->host, hoststr),
3091 ntohs(host->port)));
3093 multi_Rx(conns, j) {
3094 multi_RXAFSCB_ProbeUuid(&host->interface->uuid);
3098 if (host->callback_rxcon)
3099 rx_DestroyConnection(host->callback_rxcon);
3100 host->callback_rxcon = conns[multi_i];
3101 /* add then remove */
3102 addInterfaceAddr_r(host, interfaces[multi_i].addr,
3103 interfaces[multi_i].port);
3104 removeInterfaceAddr_r(host, host->host, host->port);
3105 host->host = interfaces[multi_i].addr;
3106 host->port = interfaces[multi_i].port;
3107 connSuccess = conns[multi_i];
3108 rx_SetConnDeadTime(host->callback_rxcon, 50);
3109 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3111 ("multiprobe success with addr %s:%d\n",
3112 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3113 ntohs(interfaces[multi_i].port)));
3118 ("multiprobe failure with addr %s:%d\n",
3119 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3120 ntohs(interfaces[multi_i].port)));
3122 /* This is less than desirable but its the best we can do.
3123 * The AFS Cache Manager will return either 0 for a Uuid
3124 * match and a 1 for a non-match. If the error is 1 we
3125 * therefore know that our mapping of IP address to Uuid
3126 * is wrong. We should attempt to find the correct
3127 * Uuid and fix the host tables.
3129 if (multi_error == 1) {
3130 /* remove the current alternate address from this host */
3132 removeInterfaceAddr_r(host, interfaces[multi_i].addr, interfaces[multi_i].port);
3136 #ifdef AFS_DEMAND_ATTACH_FS
3137 /* try to bail ASAP if the fileserver is shutting down */
3139 if (fs_state.mode == FS_MODE_SHUTDOWN) {
3148 /* Destroy all connections except the one on which we succeeded */
3149 for (i = 0; i < j; i++)
3150 if (conns[i] != connSuccess)
3151 rx_DestroyConnection(conns[i]);
3157 return 0; /* success */
3159 return 1; /* failure */
3162 #endif /* !defined(INTERPRET_DUMP) */