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(FT_ApproxTime());
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;
461 rx_SetConnDeadTime(ahost->callback_rxcon, 4);
462 rx_SetConnHardDeadTime(ahost->callback_rxcon, AFS_HARDDEADTIME);
469 for (i = 0; i < nfids && i < AFSCBMAX; i++) {
470 tcbs[i].CallBackVersion = CALLBACK_VERSION;
471 tcbs[i].ExpirationTime = 0;
472 tcbs[i].CallBackType = CB_DROPPED;
474 tf.AFSCBFids_len = i;
475 tf.AFSCBFids_val = &(fids[j]);
479 tc.AFSCBs_val = tcbs;
481 cb_conn = ahost->callback_rxcon;
482 rx_GetConnection(cb_conn);
484 code |= RXAFSCB_CallBack(cb_conn, &tf, &tc);
485 rx_PutConnection(cb_conn);
493 /* the locked flag tells us if the host entry has already been locked
494 * by our parent. I don't think anybody actually calls us with the
495 * host locked, but here's how to make that work: GetSomeSpace has to
496 * change so that it doesn't attempt to lock any hosts < "host". That
497 * means that it might be unable to free any objects, so it has to
498 * return an exit status. If it fails, then AddCallBack1 might fail,
499 * as well. If so, the host->ResetDone should probably be set to 0,
500 * and we probably don't want to return a callback promise to the
501 * cache manager, either. */
503 AddCallBack1(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
511 if (!(host->hostFlags & HOSTDELETED))
512 retVal = AddCallBack1_r(host, fid, thead, type, 1);
522 AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
525 struct FileEntry *fe;
526 struct CallBack *cb = 0, *lastcb = 0;
527 struct FileEntry *newfe = 0;
528 afs_uint32 time_out = 0;
529 afs_uint32 *Thead = thead;
530 struct CallBack *newcb = 0;
533 cbstuff.AddCallBacks++;
537 /* allocate these guys first, since we can't call the allocator with
538 * the host structure locked -- or we might deadlock. However, we have
539 * to avoid races with FindFE... */
540 while (!(newcb = GetCB())) {
541 GetSomeSpace_r(host, locked);
543 while (!(newfe = GetFE())) { /* Get it now, so we don't have to call */
544 /* GetSomeSpace with the host locked, later. This might turn out to */
545 /* have been unneccessary, but that's actually kind of unlikely, since */
546 /* most files are not shared. */
547 GetSomeSpace_r(host, locked);
551 h_Lock_r(host); /* this can yield, so do it before we get any */
553 if (host->hostFlags & HOSTDELETED) {
561 if (type == CB_NORMAL) {
563 TimeCeiling(FT_ApproxTime() + TimeOut(fe ? fe->ncbs : 0) +
565 Thead = THead(CBtime(time_out));
566 } else if (type == CB_VOLUME) {
567 time_out = TimeCeiling((60 * 120 + FT_ApproxTime()) + ServerBias);
568 Thead = THead(CBtime(time_out));
569 } else if (type == CB_BULK) {
570 /* bulk status can get so many callbacks all at once, and most of them
571 * are probably not for things that will be used for long.
574 TimeCeiling(FT_ApproxTime() + ServerBias +
575 TimeOut(22 + (fe ? fe->ncbs : 0)));
576 Thead = THead(CBtime(time_out));
587 fe->volid = fid->Volume;
588 fe->vnode = fid->Vnode;
589 fe->unique = fid->Unique;
592 hash = FEHash(fid->Volume, fid->Unique);
593 fe->fnext = HashTable[hash];
594 HashTable[hash] = fetoi(fe);
596 for (safety = 0, lastcb = cb = itocb(fe->firstcb); cb;
597 lastcb = cb, cb = itocb(cb->cnext), safety++) {
598 if (safety > cbstuff.nblks) {
599 ViceLog(0, ("AddCallBack1: Internal Error -- shutting down.\n"));
600 DumpCallBackState_r();
601 ShutDownAndCore(PANIC);
603 if (cb->hhead == h_htoi(host))
606 if (cb) { /* Already have call back: move to new timeout list */
607 /* don't change delayed callbacks back to normal ones */
608 if (cb->status != CB_DELAYED)
610 /* Only move if new timeout is longer */
611 if (TNorm(ttoi(Thead)) > TNorm(cb->thead)) {
615 if (newfe == NULL) { /* we are using the new FE */
616 fe->firstcb = cbtoi(cb);
618 cb->fhead = fetoi(fe);
623 *(lastcb ? &lastcb->cnext : &fe->firstcb) = cbtoi(cb);
626 cb->fhead = fetoi(fe);
632 /* now free any still-unused callback or host entries */
638 if (!locked) /* freecb and freefe might(?) yield */
641 if (type == CB_NORMAL || type == CB_VOLUME || type == CB_BULK)
642 return time_out - ServerBias; /* Expires sooner at workstation */
648 CompareCBA(const void *e1, const void *e2)
650 const struct cbstruct *cba1 = (const struct cbstruct *)e1;
651 const struct cbstruct *cba2 = (const struct cbstruct *)e2;
652 return ((cba1->hp)->index - (cba2->hp)->index);
655 /* Take an array full of hosts, all held. Break callbacks to them, and
656 * release the holds once you're done.
657 * Currently only works for a single Fid in afidp array.
658 * If you want to make this work with multiple fids, you need to fix
659 * the error handling. One approach would be to force a reset if a
660 * multi-fid call fails, or you could add delayed callbacks for each
661 * fid. You probably also need to sort and remove duplicate hosts.
662 * When this is called from the BreakVolumeCallBacks path, it does NOT
663 * force a reset if the RPC fails, it just marks the host down and tries
664 * to create a delayed callback. */
665 /* N.B. be sure that code works when ncbas == 0 */
666 /* N.B. requires all the cba[*].hp pointers to be valid... */
667 /* This routine does not hold a lock on the host for the duration of
668 * the BreakCallBack RPC, which is a significant deviation from tradition.
669 * It _does_ get a lock on the host before setting VenusDown = 1,
670 * which is sufficient only if VenusDown = 0 only happens when the
671 * lock is held over the RPC and the subsequent VenusDown == 0
672 * wherever that is done. */
674 MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
675 struct AFSCBFids *afidp)
678 struct rx_connection *conns[MAX_CB_HOSTS];
679 static struct AFSCBs tc = { 0, 0 };
680 int multi_to_cba_map[MAX_CB_HOSTS];
682 osi_Assert(ncbas <= MAX_CB_HOSTS);
684 /* sort cba list to avoid makecall issues */
685 qsort(cba, ncbas, sizeof(struct cbstruct), CompareCBA);
687 /* set up conns for multi-call */
688 for (i = 0, j = 0; i < ncbas; i++) {
689 struct host *thishost = cba[i].hp;
690 if (!thishost || (thishost->hostFlags & HOSTDELETED)) {
693 rx_GetConnection(thishost->callback_rxcon);
694 multi_to_cba_map[j] = i;
695 conns[j++] = thishost->callback_rxcon;
698 rx_SetConnDeadTime(thishost->callback_rxcon, 4);
699 rx_SetConnHardDeadTime(thishost->callback_rxcon, AFS_HARDDEADTIME);
703 if (j) { /* who knows what multi would do with 0 conns? */
707 multi_RXAFSCB_CallBack(afidp, &tc);
713 i = multi_to_cba_map[multi_i];
719 ("BCB: INTERNAL ERROR: hp=%p, cba=%p, thead=%u\n",
723 ** try breaking callbacks on alternate interface addresses
725 if (MultiBreakCallBackAlternateAddress(hp, afidp)) {
728 ("BCB: Failed on file %u.%u.%u, "
729 "Host %p (%s:%d) is down\n",
730 afidp->AFSCBFids_val->Volume,
731 afidp->AFSCBFids_val->Vnode,
732 afidp->AFSCBFids_val->Unique,
734 afs_inet_ntoa_r(hp->host, hoststr),
740 if (!(hp->hostFlags & HOSTDELETED)) {
741 hp->hostFlags |= VENUSDOWN;
743 * We always go into AddCallBack1_r with the host locked
745 AddCallBack1_r(hp, afidp->AFSCBFids_val, itot(idx),
759 for (i = 0; i < ncbas; i++) {
767 /* H_UNLOCK around this so h_FreeConnection does not deadlock.
768 h_FreeConnection should *never* be called on a callback connection,
769 but on 10/27/04 a deadlock occurred where it was, when we know why,
770 this should be reverted. -- shadow */
772 for (i = 0; i < j; i++) {
773 rx_PutConnection(conns[i]);
781 * Break all call backs for fid, except for the specified host (unless flag
782 * is true, in which case all get a callback message. Assumption: the specified
783 * host is h_Held, by the caller; the others aren't.
784 * Specified host may be bogus, that's ok. This used to check to see if the
785 * host was down in two places, once right after the host was h_held, and
786 * again after it was locked. That race condition is incredibly rare and
787 * relatively harmless even when it does occur, so we don't check for it now.
789 /* if flag is true, send a break callback msg to "host", too */
791 BreakCallBack(struct host *xhost, AFSFid * fid, int flag)
793 struct FileEntry *fe;
794 struct CallBack *cb, *nextcb;
795 struct cbstruct cbaDef[MAX_CB_HOSTS], *cba = cbaDef;
796 unsigned int ncbas, cbaAlloc = MAX_CB_HOSTS;
803 ("BCB: BreakCallBack(Host %p all but %s:%d, (%u,%u,%u))\n",
804 xhost, afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
805 fid->Volume, fid->Vnode, fid->Unique));
808 ("BCB: BreakCallBack(No Host, (%u,%u,%u))\n",
809 fid->Volume, fid->Vnode, fid->Unique));
812 cbstuff.BreakCallBacks++;
817 hostindex = xhost ? h_htoi(xhost) : 0;
818 cb = itocb(fe->firstcb);
819 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
820 /* the most common case is what follows the || */
823 tf.AFSCBFids_len = 1;
824 tf.AFSCBFids_val = fid;
826 for (ncbas = 0; cb ; cb = nextcb) {
827 nextcb = itocb(cb->cnext);
828 if ((cb->hhead != hostindex || flag)
829 && (cb->status == CB_BULK || cb->status == CB_NORMAL
830 || cb->status == CB_VOLUME)) {
831 struct host *thishost = h_itoh(cb->hhead);
833 ViceLog(0, ("BCB: BOGUS! cb->hhead is NULL!\n"));
834 } else if (thishost->hostFlags & VENUSDOWN) {
836 ("BCB: %p (%s:%d) is down; delaying break call back\n",
837 thishost, afs_inet_ntoa_r(thishost->host, hoststr),
838 ntohs(thishost->port)));
839 cb->status = CB_DELAYED;
841 if (!(thishost->hostFlags & HOSTDELETED)) {
843 if (ncbas == cbaAlloc) { /* Need more space */
844 int curLen = cbaAlloc*sizeof(cba[0]);
845 struct cbstruct *cbaOld = (cba == cbaDef) ? NULL : cba;
847 /* There are logical contraints elsewhere that the number of hosts
848 (i.e. h_HTSPERBLOCK*h_MAXHOSTTABLES) remains in the realm of a signed "int".
849 cbaAlloc is defined unsigned int hence doubling below cannot overflow
851 cbaAlloc = cbaAlloc<<1; /* double */
852 cba = realloc(cbaOld, cbaAlloc * sizeof(cba[0]));
854 if (cbaOld == NULL) { /* realloc wouldn't have copied from cbaDef */
855 memcpy(cba, cbaDef, curLen);
858 cba[ncbas].hp = thishost;
859 cba[ncbas].thead = cb->thead;
864 CDel(cb, 1); /* Usually first; so this delete
865 * is reasonably inexpensive */
871 struct cbstruct *cba2;
874 for (cba2 = cba, num = ncbas; ncbas > 0; cba2 += num, ncbas -= num) {
875 num = (ncbas > MAX_CB_HOSTS) ? MAX_CB_HOSTS : ncbas;
876 MultiBreakCallBack_r(cba2, num, &tf);
880 if (cba != cbaDef) free(cba);
887 /* Delete (do not break) single call back for fid */
889 DeleteCallBack(struct host *host, AFSFid * fid)
891 struct FileEntry *fe;
896 cbstuff.DeleteCallBacks++;
899 /* do not care if the host has been HOSTDELETED */
905 ("DCB: No call backs for fid (%u, %u, %u)\n", fid->Volume,
906 fid->Vnode, fid->Unique));
909 pcb = FindCBPtr(fe, host);
912 ("DCB: No call back for host %p (%s:%d), (%u, %u, %u)\n",
913 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
914 fid->Volume, fid->Vnode, fid->Unique));
928 * Delete (do not break) all call backs for fid. This call doesn't
929 * set all of the various host locks, but it shouldn't really matter
930 * since we're not adding callbacks, but deleting them. I'm not sure
931 * why it doesn't set the lock, however; perhaps it should.
934 DeleteFileCallBacks(AFSFid * fid)
936 struct FileEntry *fe;
942 cbstuff.DeleteFiles++;
947 ("DF: No fid (%u,%u,%u) to delete\n", fid->Volume, fid->Vnode,
951 for (n = 0, cbi = fe->firstcb; cbi; n++) {
964 /* Delete (do not break) all call backs for host. The host should be
967 DeleteAllCallBacks_r(struct host *host, int deletefe)
972 cbstuff.DeleteAllCallBacks++;
973 cbi = first = host->cblist;
975 ViceLog(8, ("DV: no call backs\n"));
983 } while (cbi != first);
989 * Break all delayed call backs for host. Returns 1 if all call backs
990 * successfully broken; 0 otherwise. Assumes host is h_Held and h_Locked.
991 * Must be called with VenusDown set for this host
994 BreakDelayedCallBacks(struct host *host)
998 retVal = BreakDelayedCallBacks_r(host);
1004 BreakDelayedCallBacks_r(struct host *host)
1006 struct AFSFid fids[AFSCBMAX];
1007 int cbi, first, nfids;
1008 struct CallBack *cb;
1011 struct rx_connection *cb_conn;
1013 cbstuff.nbreakers++;
1014 if (!(host->hostFlags & RESETDONE) && !(host->hostFlags & HOSTDELETED)) {
1015 host->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1016 cb_conn = host->callback_rxcon;
1017 rx_GetConnection(cb_conn);
1018 if (host->interface) {
1021 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1024 code = RXAFSCB_InitCallBackState(cb_conn);
1026 rx_PutConnection(cb_conn);
1029 host->hostFlags |= ALTADDR; /* alternate addresses are valid */
1033 ("CB: Call back connect back failed (in break delayed) "
1034 "for Host %p (%s:%d)\n",
1035 host, afs_inet_ntoa_r(host->host, hoststr),
1036 ntohs(host->port)));
1038 host->hostFlags |= VENUSDOWN;
1041 ("InitCallBackState success on %p (%s:%d)\n",
1042 host, afs_inet_ntoa_r(host->host, hoststr),
1043 ntohs(host->port)));
1044 /* reset was done successfully */
1045 host->hostFlags |= RESETDONE;
1046 host->hostFlags &= ~VENUSDOWN;
1049 while (!(host->hostFlags & HOSTDELETED)) {
1051 host->hostFlags &= ~VENUSDOWN; /* presume up */
1052 cbi = first = host->cblist;
1056 first = host->cblist;
1059 if (cb->status == CB_DELAYED) {
1060 struct FileEntry *fe = itofe(cb->fhead);
1061 fids[nfids].Volume = fe->volid;
1062 fids[nfids].Vnode = fe->vnode;
1063 fids[nfids].Unique = fe->unique;
1069 } while (cbi && cbi != first && nfids < AFSCBMAX);
1075 if (XCallBackBulk_r(host, fids, nfids)) {
1076 /* Failed, again: put them back, probably with old
1081 ("CB: XCallBackBulk failed, Host %p (%s:%d); "
1082 "callback list follows:\n",
1083 host, afs_inet_ntoa_r(host->host, hoststr),
1084 ntohs(host->port)));
1086 for (i = 0; i < nfids; i++) {
1089 ("CB: Host %p (%s:%d), file %u.%u.%u "
1090 "(part of bulk callback)\n",
1091 host, afs_inet_ntoa_r(host->host, hoststr),
1092 ntohs(host->port), fids[i].Volume,
1093 fids[i].Vnode, fids[i].Unique));
1096 * AddCallBack1_r(host, &fids[i], itot(thead[i]), CB_DELAYED, 1);
1097 * * but it turns out to cause too many tricky locking problems.
1098 * * now, if break delayed fails, screw it. */
1100 host->hostFlags |= VENUSDOWN; /* Failed */
1101 ClearHostCallbacks_r(host, 1 /* locked */ );
1105 if (nfids < AFSCBMAX)
1109 cbstuff.nbreakers--;
1110 /* If we succeeded it's always ok to unset HFE_LATER */
1111 if (!(host->hostFlags & VENUSDOWN))
1112 host->hostFlags &= ~HFE_LATER;
1113 return (host->hostFlags & VENUSDOWN);
1117 MultiBreakVolumeCallBack_r(struct host *host,
1118 struct VCBParams *parms, int deletefe)
1122 if (host->hostFlags & HOSTDELETED)
1125 if (!(host->hostFlags & HCBREAK))
1126 return 0; /* host is not flagged to notify */
1128 if (host->hostFlags & VENUSDOWN) {
1130 /* Do not care if the host is now HOSTDELETED */
1133 ("BVCB: volume callback for Host %p (%s:%d) failed\n",
1134 host, afs_inet_ntoa_r(host->host, hoststr),
1135 ntohs(host->port)));
1137 DeleteAllCallBacks_r(host, deletefe); /* Delete all callback state
1138 * rather than attempting to
1139 * selectively remember to
1140 * delete the volume callbacks
1142 host->hostFlags &= ~(RESETDONE|HCBREAK); /* Do InitCallBackState when host returns */
1146 osi_Assert(parms->ncbas <= MAX_CB_HOSTS);
1148 /* Do not call MultiBreakCallBack on the current host structure
1149 ** because it would prematurely release the hold on the host
1151 if (parms->ncbas == MAX_CB_HOSTS) {
1152 struct AFSCBFids tf;
1154 tf.AFSCBFids_len = 1;
1155 tf.AFSCBFids_val = parms->fid;
1157 /* this releases all the hosts */
1158 MultiBreakCallBack_r(parms->cba, parms->ncbas, &tf);
1162 parms->cba[parms->ncbas].hp = host;
1163 parms->cba[(parms->ncbas)++].thead = parms->thead;
1164 host->hostFlags &= ~HCBREAK;
1166 /* we have more work to do on this host, so make sure we keep a reference
1174 MultiBreakVolumeLaterCallBack(struct host *host, void *rock)
1176 struct VCBParams *parms = (struct VCBParams *)rock;
1179 retval = MultiBreakVolumeCallBack_r(host, parms, 0);
1185 * Break all call backs on a single volume. Don't call this with any
1186 * hosts h_held. Note that this routine clears the callbacks before
1187 * actually breaking them, and that the vnode isn't locked during this
1188 * operation, so that people might see temporary callback loss while
1189 * this function is executing. It is just a temporary state, however,
1190 * since the callback will be broken later by this same function.
1192 * Now uses multi-RX for CallBack RPC in a different thread,
1193 * only marking them here.
1195 extern pthread_cond_t fsync_cond;
1198 BreakVolumeCallBacksLater(afs_uint32 volume)
1202 struct FileEntry *fe;
1203 struct CallBack *cb;
1207 ViceLog(25, ("Setting later on volume %u\n", volume));
1209 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1210 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1211 if (fe->volid == volume) {
1212 struct CallBack *cbnext;
1213 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1214 host = h_itoh(cb->hhead);
1215 host->hostFlags |= HFE_LATER;
1216 cb->status = CB_DELAYED;
1217 cbnext = itocb(cb->cnext);
1220 fe->status |= FE_LATER;
1229 /* didn't find any callbacks, so return right away. */
1233 ViceLog(25, ("Fsync thread wakeup\n"));
1235 CV_BROADCAST(&fsync_cond);
1241 BreakLaterCallBacks(void)
1246 struct CallBack *cb;
1247 struct FileEntry *fe = NULL;
1248 struct FileEntry *myfe = NULL;
1250 struct VCBParams henumParms;
1251 unsigned short tthead = 0; /* zero is illegal value */
1255 ViceLog(25, ("Looking for FileEntries to unchain\n"));
1258 /* Pick the first volume we see to clean up */
1259 fid.Volume = fid.Vnode = fid.Unique = 0;
1261 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1262 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1263 if (fe && (fe->status & FE_LATER)
1264 && (fid.Volume == 0 || fid.Volume == fe->volid)) {
1265 /* Ugly, but used to avoid left side casting */
1266 struct object *tmpfe;
1268 ("Unchaining for %u:%u:%u\n", fe->vnode, fe->unique,
1270 fid.Volume = fe->volid;
1272 fe->status &= ~FE_LATER; /* not strictly needed */
1273 /* Works since volid is deeper than the largest pointer */
1274 tmpfe = (struct object *)fe;
1275 tmpfe->next = (struct object *)myfe;
1288 /* loop over FEs from myfe and free/break */
1290 for (fe = myfe; fe;) {
1291 struct CallBack *cbnext;
1292 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1293 cbnext = itocb(cb->cnext);
1294 host = h_itoh(cb->hhead);
1295 if (cb->status == CB_DELAYED) {
1296 if (!(host->hostFlags & HOSTDELETED)) {
1297 /* mark this host for notification */
1298 host->hostFlags |= HCBREAK;
1299 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1305 CDel(cb, 0); /* Don't let CDel clean up the fe */
1306 /* leave flag for MultiBreakVolumeCallBack to clear */
1309 ("Found host %p (%s:%d) non-DELAYED cb for %u:%u:%u\n",
1310 host, afs_inet_ntoa_r(host->host, hoststr),
1311 ntohs(host->port), fe->vnode, fe->unique, fe->volid));
1315 fe = (struct FileEntry *)((struct object *)fe)->next;
1320 ViceLog(125, ("Breaking volume %u\n", fid.Volume));
1321 henumParms.ncbas = 0;
1322 henumParms.fid = &fid;
1323 henumParms.thead = tthead;
1325 h_Enumerate(MultiBreakVolumeLaterCallBack, (char *)&henumParms);
1327 if (henumParms.ncbas) { /* do left-overs */
1328 struct AFSCBFids tf;
1329 tf.AFSCBFids_len = 1;
1330 tf.AFSCBFids_val = &fid;
1332 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf);
1333 henumParms.ncbas = 0;
1338 /* Arrange to be called again */
1343 * Delete all timed-out call back entries (to be called periodically by file
1347 CleanupTimedOutCallBacks(void)
1350 CleanupTimedOutCallBacks_r();
1356 CleanupTimedOutCallBacks_r(void)
1358 afs_uint32 now = CBtime(FT_ApproxTime());
1360 struct CallBack *cb;
1364 while (tfirst <= now) {
1366 cbi = *(thead = THead(tfirst));
1372 ("CCB: deleting timed out call back %x (%s:%d), (%u,%u,%u)\n",
1373 h_itoh(cb->hhead)->host,
1374 afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
1375 h_itoh(cb->hhead)->port, itofe(cb->fhead)->volid,
1376 itofe(cb->fhead)->vnode, itofe(cb->fhead)->unique));
1380 if (ntimedout > cbstuff.nblks) {
1381 ViceLog(0, ("CCB: Internal Error -- shutting down...\n"));
1382 DumpCallBackState_r();
1383 ShutDownAndCore(PANIC);
1385 } while (cbi != *thead);
1390 cbstuff.CBsTimedOut += ntimedout;
1391 ViceLog(7, ("CCB: deleted %d timed out callbacks\n", ntimedout));
1392 return (ntimedout > 0);
1396 * parameters to pass to lih*_r from h_Enumerate_r when trying to find a host
1397 * from which to clear callbacks.
1401 * Points to the least interesting host found; try to clear callbacks on
1402 * this host after h_Enumerate_r(lih*_r)'ing.
1407 * The last host we got from lih*_r, but we couldn't clear its callbacks
1408 * for some reason. Choose the next-best host after this one (with the
1409 * current lih*_r, this means to only select hosts that have an ActiveCall
1410 * newer than lastlih).
1412 struct host *lastlih;
1415 /* Value of host->refCount that allows us to reliably infer that
1416 * host may be held by some other thread */
1417 #define OTHER_MUSTHOLD_LIH 2
1419 /* This version does not allow 'host' to be selected unless its ActiveCall
1420 * is newer than 'params->lastlih' which is the host with the oldest
1421 * ActiveCall from the last pass (if it is provided). We filter out any hosts
1422 * that are are held by other threads.
1424 * There is a small problem here, but it may not be easily fixable. Say we
1425 * select some host A, and give it back to GetSomeSpace_r. GSS_r for some
1426 * reason cannot clear the callbacks on A, and so calls us again with
1427 * lastlih = A. Suppose there is another host B that has the same ActiveCall
1428 * time as A. We will now skip over host B, since
1429 * 'hostB->ActiveCall > hostA->ActiveCall' is not true. This could result in
1430 * us prematurely going to the GSS_r 2nd or 3rd pass, and making us a little
1431 * inefficient. This should be pretty rare, though, except perhaps in cases
1432 * with very small numbers of hosts.
1434 * Also filter out any hosts with HOSTDELETED set. h_Enumerate_r should in
1435 * theory not give these to us anyway, but be paranoid.
1438 lih0_r(struct host *host, void *rock)
1440 struct lih_params *params = (struct lih_params *)rock;
1442 /* OTHER_MUSTHOLD_LIH is because the h_Enum loop holds us once */
1444 && (!(host->hostFlags & HOSTDELETED))
1445 && (host->refCount < OTHER_MUSTHOLD_LIH)
1446 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1447 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1450 h_Release_r(params->lih); /* release prev host */
1459 /* same as lih0_r, except we do not prevent held hosts from being selected. */
1461 lih1_r(struct host *host, void *rock)
1463 struct lih_params *params = (struct lih_params *)rock;
1466 && (!(host->hostFlags & HOSTDELETED))
1467 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1468 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1471 h_Release_r(params->lih); /* release prev host */
1480 /* This could be upgraded to get more space each time */
1481 /* first pass: sequentially find the oldest host which isn't held by
1482 anyone for which we can clear callbacks;
1484 /* second pass: sequentially find the oldest host regardless of
1485 whether or not the host is held; skipping 'hostp' */
1486 /* third pass: attempt to clear callbacks from 'hostp' */
1487 /* always called with hostp unlocked */
1489 /* Note: hostlist is ordered most recently created host first and
1490 * its order has no relationship to the most recently used. */
1491 extern struct host *hostList;
1493 GetSomeSpace_r(struct host *hostp, int locked)
1496 struct lih_params params;
1499 if (cbstuff.GotSomeSpaces == 0) {
1500 /* only log this once; if GSS is getting called constantly, that's not
1501 * good but don't make things worse by spamming the log. */
1502 ViceLog(0, ("We have run out of callback space; forcing callback revocation. "
1503 "This suggests the fileserver is configured with insufficient "
1504 "callbacks; you probably want to increase the -cb fileserver "
1505 "parameter (current setting: %u). The fileserver will continue "
1506 "to operate, but this may indicate a severe performance problem\n",
1508 ViceLog(0, ("This message is logged at most once; for more information "
1509 "see the OpenAFS documentation and fileserver xstat collection 3\n"));
1512 cbstuff.GotSomeSpaces++;
1514 ("GSS: First looking for timed out call backs via CleanupCallBacks\n"));
1515 if (CleanupTimedOutCallBacks_r()) {
1521 params.lastlih = NULL;
1526 h_Enumerate_r(i == 0 ? lih0_r : lih1_r, hostList, ¶ms);
1529 if (params.lastlih) {
1530 h_Release_r(params.lastlih);
1531 params.lastlih = NULL;
1535 /* note that 'hp' was held by lih*_r; we will need to release it */
1537 if ((hp != hostp) && !ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
1542 params.lastlih = hp;
1543 /* params.lastlih will be released on the next iteration, after
1548 * Next time try getting callbacks from any host even if
1549 * it's held, since the only other option is starvation for
1550 * the file server (i.e. until the callback timeout arrives).
1553 params.lastlih = NULL;
1556 ("GSS: Try harder for longest inactive host cnt= %d\n",
1561 /* Could not obtain space from other hosts, clear hostp's callback state */
1566 ClearHostCallbacks_r(hostp, 1 /*already locked */ );
1573 /* locked - set if caller has already locked the host */
1575 ClearHostCallbacks_r(struct host *hp, int locked)
1579 struct rx_connection *cb_conn = NULL;
1582 ("GSS: Delete longest inactive host %p (%s:%d)\n",
1583 hp, afs_inet_ntoa_r(hp->host, hoststr), ntohs(hp->port)));
1585 if ((hp->hostFlags & HOSTDELETED)) {
1586 /* hp could go away after reacquiring H_LOCK in h_NBLock_r, so we can't
1587 * really use it; its callbacks will get cleared anyway when
1588 * h_TossStuff_r gets its hands on it */
1594 /** Try a non-blocking lock. If the lock is already held return
1595 * after releasing hold on hp
1598 if (h_NBLock_r(hp)) {
1603 if (hp->Console & 2) {
1605 * If the special console field is set it means that a thread
1606 * is waiting in AddCallBack1 after it set pointers to the
1607 * file entry and/or callback entry. Because of the bogus
1608 * usage of h_hold it won't prevent from another thread, this
1609 * one, to remove all the callbacks so just to be safe we keep
1610 * a reference. NOTE, on the last phase we'll free the calling
1611 * host's callbacks but that's ok...
1615 DeleteAllCallBacks_r(hp, 1);
1616 if (hp->hostFlags & VENUSDOWN) {
1617 hp->hostFlags &= ~RESETDONE; /* remember that we must do a reset */
1618 } else if (!(hp->hostFlags & HOSTDELETED)) {
1619 /* host is up, try a call */
1620 hp->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1621 cb_conn = hp->callback_rxcon;
1622 rx_GetConnection(hp->callback_rxcon);
1623 if (hp->interface) {
1626 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1629 code = RXAFSCB_InitCallBackState(cb_conn);
1631 rx_PutConnection(cb_conn);
1634 hp->hostFlags |= ALTADDR; /* alternate addresses are valid */
1636 /* failed, mark host down and need reset */
1637 hp->hostFlags |= VENUSDOWN;
1638 hp->hostFlags &= ~RESETDONE;
1640 /* reset succeeded, we're done */
1641 hp->hostFlags |= RESETDONE;
1650 #endif /* INTERPRET_DUMP */
1654 PrintCallBackStats(void)
1657 "%d add CB, %d break CB, %d del CB, %d del FE, %d CB's timed out, %d space reclaim, %d del host\n",
1658 cbstuff.AddCallBacks, cbstuff.BreakCallBacks,
1659 cbstuff.DeleteCallBacks, cbstuff.DeleteFiles, cbstuff.CBsTimedOut,
1660 cbstuff.GotSomeSpaces, cbstuff.DeleteAllCallBacks);
1661 fprintf(stderr, "%d CBs, %d FEs, (%d of total of %d 16-byte blocks)\n",
1662 cbstuff.nCBs, cbstuff.nFEs, cbstuff.nCBs + cbstuff.nFEs,
1664 fprintf(stderr, "%d GSS1, %d GSS2, %d GSS3, %d GSS4, %d GSS5 (internal counters)\n",
1665 cbstuff.GSS1, cbstuff.GSS2, cbstuff.GSS3, cbstuff.GSS4, cbstuff.GSS5);
1670 #define MAGIC 0x12345678 /* To check byte ordering of dump when it is read in */
1671 #define MAGICV2 0x12345679 /* To check byte ordering & version of dump when it is read in */
1674 #ifndef INTERPRET_DUMP
1676 #ifdef AFS_DEMAND_ATTACH_FS
1679 * callback state serialization
1681 static int cb_stateSaveTimeouts(struct fs_dump_state * state);
1682 static int cb_stateSaveFEHash(struct fs_dump_state * state);
1683 static int cb_stateSaveFEs(struct fs_dump_state * state);
1684 static int cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe);
1685 static int cb_stateRestoreTimeouts(struct fs_dump_state * state);
1686 static int cb_stateRestoreFEHash(struct fs_dump_state * state);
1687 static int cb_stateRestoreFEs(struct fs_dump_state * state);
1688 static int cb_stateRestoreFE(struct fs_dump_state * state);
1689 static int cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
1690 struct iovec * iov, int niovecs);
1692 static int cb_stateVerifyFEHash(struct fs_dump_state * state);
1693 static int cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe);
1694 static int cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe);
1695 static int cb_stateVerifyTimeoutQueues(struct fs_dump_state * state);
1697 static int cb_stateFEToDiskEntry(struct FileEntry *, struct FEDiskEntry *);
1698 static int cb_stateDiskEntryToFE(struct fs_dump_state * state,
1699 struct FEDiskEntry *, struct FileEntry *);
1701 static int cb_stateCBToDiskEntry(struct CallBack *, struct CBDiskEntry *);
1702 static int cb_stateDiskEntryToCB(struct fs_dump_state * state,
1703 struct CBDiskEntry *, struct CallBack *);
1705 static int cb_stateFillHeader(struct callback_state_header * hdr);
1706 static int cb_stateCheckHeader(struct callback_state_header * hdr);
1708 static int cb_stateAllocMap(struct fs_dump_state * state);
1711 cb_stateSave(struct fs_dump_state * state)
1715 AssignInt64(state->eof_offset, &state->hdr->cb_offset);
1717 /* invalidate callback state header */
1718 memset(state->cb_hdr, 0, sizeof(struct callback_state_header));
1719 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1720 sizeof(struct callback_state_header))) {
1725 fs_stateIncEOF(state, sizeof(struct callback_state_header));
1727 /* dump timeout state */
1728 if (cb_stateSaveTimeouts(state)) {
1733 /* dump fe hashtable state */
1734 if (cb_stateSaveFEHash(state)) {
1739 /* dump callback state */
1740 if (cb_stateSaveFEs(state)) {
1745 /* write the callback state header to disk */
1746 cb_stateFillHeader(state->cb_hdr);
1747 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1748 sizeof(struct callback_state_header))) {
1758 cb_stateRestore(struct fs_dump_state * state)
1762 if (fs_stateReadHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1763 sizeof(struct callback_state_header))) {
1768 if (cb_stateCheckHeader(state->cb_hdr)) {
1773 if (cb_stateAllocMap(state)) {
1778 if (cb_stateRestoreTimeouts(state)) {
1783 if (cb_stateRestoreFEHash(state)) {
1788 /* restore FEs and CBs from disk */
1789 if (cb_stateRestoreFEs(state)) {
1794 /* restore the timeout queue heads */
1795 tfirst = state->cb_hdr->tfirst;
1802 cb_stateRestoreIndices(struct fs_dump_state * state)
1805 struct FileEntry * fe;
1806 struct CallBack * cb;
1808 /* restore indices in the FileEntry structures */
1809 for (i = 1; i < state->fe_map.len; i++) {
1810 if (state->fe_map.entries[i].new_idx) {
1811 fe = itofe(state->fe_map.entries[i].new_idx);
1813 /* restore the fe->fnext entry */
1814 if (fe_OldToNew(state, fe->fnext, &fe->fnext)) {
1819 /* restore the fe->firstcb entry */
1820 if (cb_OldToNew(state, fe->firstcb, &fe->firstcb)) {
1827 /* restore indices in the CallBack structures */
1828 for (i = 1; i < state->cb_map.len; i++) {
1829 if (state->cb_map.entries[i].new_idx) {
1830 cb = itocb(state->cb_map.entries[i].new_idx);
1832 /* restore the cb->cnext entry */
1833 if (cb_OldToNew(state, cb->cnext, &cb->cnext)) {
1838 /* restore the cb->fhead entry */
1839 if (fe_OldToNew(state, cb->fhead, &cb->fhead)) {
1844 /* restore the cb->hhead entry */
1845 if (h_OldToNew(state, cb->hhead, &cb->hhead)) {
1850 /* restore the cb->tprev entry */
1851 if (cb_OldToNew(state, cb->tprev, &cb->tprev)) {
1856 /* restore the cb->tnext entry */
1857 if (cb_OldToNew(state, cb->tnext, &cb->tnext)) {
1862 /* restore the cb->hprev entry */
1863 if (cb_OldToNew(state, cb->hprev, &cb->hprev)) {
1868 /* restore the cb->hnext entry */
1869 if (cb_OldToNew(state, cb->hnext, &cb->hnext)) {
1876 /* restore the timeout queue head indices */
1877 for (i = 0; i < state->cb_timeout_hdr->records; i++) {
1878 if (cb_OldToNew(state, timeout[i], &timeout[i])) {
1884 /* restore the FE hash table queue heads */
1885 for (i = 0; i < state->cb_fehash_hdr->records; i++) {
1886 if (fe_OldToNew(state, HashTable[i], &HashTable[i])) {
1897 cb_stateVerify(struct fs_dump_state * state)
1901 if (cb_stateVerifyFEHash(state)) {
1905 if (cb_stateVerifyTimeoutQueues(state)) {
1913 cb_stateVerifyFEHash(struct fs_dump_state * state)
1916 struct FileEntry * fe;
1917 afs_uint32 fei, chain_len;
1919 for (i = 0; i < FEHASH_SIZE; i++) {
1921 for (fei = HashTable[i], fe = itofe(fei);
1923 fei = fe->fnext, fe = itofe(fei)) {
1924 if (fei > cbstuff.nblks) {
1925 ViceLog(0, ("cb_stateVerifyFEHash: error: index out of range (fei=%d)\n", fei));
1929 if (cb_stateVerifyFE(state, fe)) {
1932 if (chain_len > FS_STATE_FE_MAX_HASH_CHAIN_LEN) {
1933 ViceLog(0, ("cb_stateVerifyFEHash: error: hash chain %d length exceeds %d; assuming there's a loop\n",
1934 i, FS_STATE_FE_MAX_HASH_CHAIN_LEN));
1946 cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe)
1950 if ((fe->firstcb && !fe->ncbs) ||
1951 (!fe->firstcb && fe->ncbs)) {
1952 ViceLog(0, ("cb_stateVerifyFE: error: fe->firstcb does not agree with fe->ncbs (fei=%lu, fe->firstcb=%lu, fe->ncbs=%lu)\n",
1953 afs_printable_uint32_lu(fetoi(fe)),
1954 afs_printable_uint32_lu(fe->firstcb),
1955 afs_printable_uint32_lu(fe->ncbs)));
1958 if (cb_stateVerifyFCBList(state, fe)) {
1959 ViceLog(0, ("cb_stateVerifyFE: error: FCBList failed verification (fei=%lu)\n",
1960 afs_printable_uint32_lu(fetoi(fe))));
1968 cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe)
1971 afs_uint32 cbi, fei, chain_len = 0;
1972 struct CallBack * cb;
1976 for (cbi = fe->firstcb, cb = itocb(cbi);
1978 cbi = cb->cnext, cb = itocb(cbi)) {
1979 if (cbi > cbstuff.nblks) {
1980 ViceLog(0, ("cb_stateVerifyFCBList: error: list index out of range (cbi=%d, ncbs=%d)\n",
1981 cbi, cbstuff.nblks));
1985 if (cb->fhead != fei) {
1986 ViceLog(0, ("cb_stateVerifyFCBList: error: cb->fhead != fei (fei=%d, cb->fhead=%d)\n",
1990 if (chain_len > FS_STATE_FCB_MAX_LIST_LEN) {
1991 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (fei=%d); assuming there's a loop\n",
1992 FS_STATE_FCB_MAX_LIST_LEN, fei));
1999 if (fe->ncbs != chain_len) {
2000 ViceLog(0, ("cb_stateVerifyFCBList: error: list length mismatch (len=%d, fe->ncbs=%d)\n",
2001 chain_len, fe->ncbs));
2010 cb_stateVerifyHCBList(struct fs_dump_state * state, struct host * host)
2013 afs_uint32 hi, chain_len, cbi;
2014 struct CallBack *cb, *ncb;
2019 for (cbi = host->cblist, cb = itocb(cbi);
2021 cbi = cb->hnext, cb = ncb) {
2022 if (chain_len && (host->cblist == cbi)) {
2023 /* we've wrapped around the circular list, and everything looks ok */
2026 if (cb->hhead != hi) {
2027 ViceLog(0, ("cb_stateVerifyHCBList: error: incorrect cb->hhead (cbi=%d, h->index=%d, cb->hhead=%d)\n",
2028 cbi, hi, cb->hhead));
2031 if (!cb->hprev || !cb->hnext) {
2032 ViceLog(0, ("cb_stateVerifyHCBList: error: null index in circular list (cbi=%d, h->index=%d)\n",
2037 if ((cb->hprev > cbstuff.nblks) ||
2038 (cb->hnext > cbstuff.nblks)) {
2039 ViceLog(0, ("cb_stateVerifyHCBList: error: list index out of range (cbi=%d, h->index=%d, cb->hprev=%d, cb->hnext=%d, nCBs=%d)\n",
2040 cbi, hi, cb->hprev, cb->hnext, cbstuff.nblks));
2044 ncb = itocb(cb->hnext);
2045 if (cbi != ncb->hprev) {
2046 ViceLog(0, ("cb_stateVerifyHCBList: error: corrupt linked list (cbi=%d, h->index=%d)\n",
2051 if (chain_len > FS_STATE_HCB_MAX_LIST_LEN) {
2052 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (h->index=%d); assuming there's a loop\n",
2053 FS_STATE_HCB_MAX_LIST_LEN, hi));
2065 cb_stateVerifyTimeoutQueues(struct fs_dump_state * state)
2068 afs_uint32 cbi, chain_len;
2069 struct CallBack *cb, *ncb;
2071 for (i = 0; i < CB_NUM_TIMEOUT_QUEUES; i++) {
2073 for (cbi = timeout[i], cb = itocb(cbi);
2075 cbi = cb->tnext, cb = ncb) {
2076 if (chain_len && (cbi == timeout[i])) {
2077 /* we've wrapped around the circular list, and everything looks ok */
2080 if (cbi > cbstuff.nblks) {
2081 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: list index out of range (cbi=%d, tindex=%d)\n",
2086 if (itot(cb->thead) != &timeout[i]) {
2087 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: cb->thead points to wrong timeout queue (tindex=%d, cbi=%d, cb->thead=%d)\n",
2088 i, cbi, cb->thead));
2091 if (!cb->tprev || !cb->tnext) {
2092 ViceLog(0, ("cb_stateVerifyTimeoutQueues: null index in circular list (cbi=%d, tindex=%d)\n",
2097 if ((cb->tprev > cbstuff.nblks) ||
2098 (cb->tnext > cbstuff.nblks)) {
2099 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list index out of range (cbi=%d, tindex=%d, cb->tprev=%d, cb->tnext=%d, nCBs=%d)\n",
2100 cbi, i, cb->tprev, cb->tnext, cbstuff.nblks));
2104 ncb = itocb(cb->tnext);
2105 if (cbi != ncb->tprev) {
2106 ViceLog(0, ("cb_stateVerifyTimeoutQueues: corrupt linked list (cbi=%d, tindex=%d)\n",
2111 if (chain_len > FS_STATE_TCB_MAX_LIST_LEN) {
2112 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list length exceeds %d (tindex=%d); assuming there's a loop\n",
2113 FS_STATE_TCB_MAX_LIST_LEN, i));
2125 cb_stateSaveTimeouts(struct fs_dump_state * state)
2128 struct iovec iov[2];
2130 AssignInt64(state->eof_offset, &state->cb_hdr->timeout_offset);
2132 memset(state->cb_timeout_hdr, 0, sizeof(struct callback_state_fehash_header));
2133 state->cb_timeout_hdr->magic = CALLBACK_STATE_TIMEOUT_MAGIC;
2134 state->cb_timeout_hdr->records = CB_NUM_TIMEOUT_QUEUES;
2135 state->cb_timeout_hdr->len = sizeof(struct callback_state_timeout_header) +
2136 (state->cb_timeout_hdr->records * sizeof(afs_uint32));
2138 iov[0].iov_base = (char *)state->cb_timeout_hdr;
2139 iov[0].iov_len = sizeof(struct callback_state_timeout_header);
2140 iov[1].iov_base = (char *)timeout;
2141 iov[1].iov_len = sizeof(timeout);
2143 if (fs_stateSeek(state, &state->cb_hdr->timeout_offset)) {
2148 if (fs_stateWriteV(state, iov, 2)) {
2153 fs_stateIncEOF(state, state->cb_timeout_hdr->len);
2160 cb_stateRestoreTimeouts(struct fs_dump_state * state)
2164 if (fs_stateReadHeader(state, &state->cb_hdr->timeout_offset,
2165 state->cb_timeout_hdr,
2166 sizeof(struct callback_state_timeout_header))) {
2171 if (state->cb_timeout_hdr->magic != CALLBACK_STATE_TIMEOUT_MAGIC) {
2175 if (state->cb_timeout_hdr->records != CB_NUM_TIMEOUT_QUEUES) {
2180 len = state->cb_timeout_hdr->records * sizeof(afs_uint32);
2182 if (state->cb_timeout_hdr->len !=
2183 (sizeof(struct callback_state_timeout_header) + len)) {
2188 if (fs_stateRead(state, timeout, len)) {
2198 cb_stateSaveFEHash(struct fs_dump_state * state)
2201 struct iovec iov[2];
2203 AssignInt64(state->eof_offset, &state->cb_hdr->fehash_offset);
2205 memset(state->cb_fehash_hdr, 0, sizeof(struct callback_state_fehash_header));
2206 state->cb_fehash_hdr->magic = CALLBACK_STATE_FEHASH_MAGIC;
2207 state->cb_fehash_hdr->records = FEHASH_SIZE;
2208 state->cb_fehash_hdr->len = sizeof(struct callback_state_fehash_header) +
2209 (state->cb_fehash_hdr->records * sizeof(afs_uint32));
2211 iov[0].iov_base = (char *)state->cb_fehash_hdr;
2212 iov[0].iov_len = sizeof(struct callback_state_fehash_header);
2213 iov[1].iov_base = (char *)HashTable;
2214 iov[1].iov_len = sizeof(HashTable);
2216 if (fs_stateSeek(state, &state->cb_hdr->fehash_offset)) {
2221 if (fs_stateWriteV(state, iov, 2)) {
2226 fs_stateIncEOF(state, state->cb_fehash_hdr->len);
2233 cb_stateRestoreFEHash(struct fs_dump_state * state)
2237 if (fs_stateReadHeader(state, &state->cb_hdr->fehash_offset,
2238 state->cb_fehash_hdr,
2239 sizeof(struct callback_state_fehash_header))) {
2244 if (state->cb_fehash_hdr->magic != CALLBACK_STATE_FEHASH_MAGIC) {
2248 if (state->cb_fehash_hdr->records != FEHASH_SIZE) {
2253 len = state->cb_fehash_hdr->records * sizeof(afs_uint32);
2255 if (state->cb_fehash_hdr->len !=
2256 (sizeof(struct callback_state_fehash_header) + len)) {
2261 if (fs_stateRead(state, HashTable, len)) {
2271 cb_stateSaveFEs(struct fs_dump_state * state)
2275 struct FileEntry *fe;
2277 AssignInt64(state->eof_offset, &state->cb_hdr->fe_offset);
2279 for (hash = 0; hash < FEHASH_SIZE ; hash++) {
2280 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
2282 if (cb_stateSaveFE(state, fe)) {
2294 cb_stateRestoreFEs(struct fs_dump_state * state)
2296 int count, nFEs, ret = 0;
2298 nFEs = state->cb_hdr->nFEs;
2300 for (count = 0; count < nFEs; count++) {
2301 if (cb_stateRestoreFE(state)) {
2312 cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe)
2314 int ret = 0, iovcnt, cbi, written = 0;
2316 struct callback_state_entry_header hdr;
2317 struct FEDiskEntry fedsk;
2318 struct CBDiskEntry cbdsk[16];
2319 struct iovec iov[16];
2320 struct CallBack *cb;
2323 if (fei > state->cb_hdr->fe_max) {
2324 state->cb_hdr->fe_max = fei;
2327 memset(&hdr, 0, sizeof(struct callback_state_entry_header));
2329 if (cb_stateFEToDiskEntry(fe, &fedsk)) {
2334 iov[0].iov_base = (char *)&hdr;
2335 iov[0].iov_len = sizeof(hdr);
2336 iov[1].iov_base = (char *)&fedsk;
2337 iov[1].iov_len = sizeof(struct FEDiskEntry);
2340 for (cbi = fe->firstcb, cb = itocb(cbi);
2342 cbi = cb->cnext, cb = itocb(cbi), hdr.nCBs++) {
2343 if (cbi > state->cb_hdr->cb_max) {
2344 state->cb_hdr->cb_max = cbi;
2346 if (cb_stateCBToDiskEntry(cb, &cbdsk[iovcnt])) {
2350 cbdsk[iovcnt].index = cbi;
2351 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2352 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2354 if ((iovcnt == 16) || (!cb->cnext)) {
2355 if (fs_stateWriteV(state, iov, iovcnt)) {
2364 hdr.magic = CALLBACK_STATE_ENTRY_MAGIC;
2365 hdr.len = sizeof(hdr) + sizeof(struct FEDiskEntry) +
2366 (hdr.nCBs * sizeof(struct CBDiskEntry));
2369 if (fs_stateWriteV(state, iov, iovcnt)) {
2374 if (fs_stateWriteHeader(state, &state->eof_offset, &hdr, sizeof(hdr))) {
2380 fs_stateIncEOF(state, hdr.len);
2383 if (fs_stateSeek(state, &state->eof_offset)) {
2389 state->cb_hdr->nFEs++;
2390 state->cb_hdr->nCBs += hdr.nCBs;
2397 cb_stateRestoreFE(struct fs_dump_state * state)
2399 int ret = 0, iovcnt, nCBs;
2400 struct callback_state_entry_header hdr;
2401 struct FEDiskEntry fedsk;
2402 struct CBDiskEntry cbdsk[16];
2403 struct iovec iov[16];
2404 struct FileEntry * fe;
2406 iov[0].iov_base = (char *)&hdr;
2407 iov[0].iov_len = sizeof(hdr);
2408 iov[1].iov_base = (char *)&fedsk;
2409 iov[1].iov_len = sizeof(fedsk);
2412 if (fs_stateReadV(state, iov, iovcnt)) {
2417 if (hdr.magic != CALLBACK_STATE_ENTRY_MAGIC) {
2424 ViceLog(0, ("cb_stateRestoreFE: ran out of free FileEntry structures\n"));
2429 if (cb_stateDiskEntryToFE(state, &fedsk, fe)) {
2435 for (iovcnt = 0, nCBs = 0;
2438 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2439 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2441 if ((iovcnt == 16) || (nCBs == hdr.nCBs - 1)) {
2442 if (fs_stateReadV(state, iov, iovcnt)) {
2446 if (cb_stateRestoreCBs(state, fe, iov, iovcnt)) {
2460 cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
2461 struct iovec * iov, int niovecs)
2464 struct CallBack * cb;
2465 struct CBDiskEntry * cbdsk;
2467 for (idx = 0; idx < niovecs; idx++) {
2468 cbdsk = (struct CBDiskEntry *) iov[idx].iov_base;
2470 if (cbdsk->cb.hhead < state->h_map.len &&
2471 state->h_map.entries[cbdsk->cb.hhead].valid == FS_STATE_IDX_SKIPPED) {
2475 if ((cb = GetCB()) == NULL) {
2476 ViceLog(0, ("cb_stateRestoreCBs: ran out of free CallBack structures\n"));
2480 if (cb_stateDiskEntryToCB(state, cbdsk, cb)) {
2481 ViceLog(0, ("cb_stateRestoreCBs: corrupt CallBack disk entry\n"));
2493 cb_stateFillHeader(struct callback_state_header * hdr)
2495 hdr->stamp.magic = CALLBACK_STATE_MAGIC;
2496 hdr->stamp.version = CALLBACK_STATE_VERSION;
2497 hdr->tfirst = tfirst;
2502 cb_stateCheckHeader(struct callback_state_header * hdr)
2506 if (hdr->stamp.magic != CALLBACK_STATE_MAGIC) {
2508 } else if (hdr->stamp.version != CALLBACK_STATE_VERSION) {
2510 } else if ((hdr->nFEs > cbstuff.nblks) || (hdr->nCBs > cbstuff.nblks)) {
2511 ViceLog(0, ("cb_stateCheckHeader: saved callback state larger than callback memory allocation\n"));
2517 /* disk entry conversion routines */
2519 cb_stateFEToDiskEntry(struct FileEntry * in, struct FEDiskEntry * out)
2521 memcpy(&out->fe, in, sizeof(struct FileEntry));
2522 out->index = fetoi(in);
2527 cb_stateDiskEntryToFE(struct fs_dump_state * state,
2528 struct FEDiskEntry * in, struct FileEntry * out)
2532 memcpy(out, &in->fe, sizeof(struct FileEntry));
2534 /* setup FE map entry */
2535 if (!in->index || (in->index >= state->fe_map.len)) {
2536 ViceLog(0, ("cb_stateDiskEntryToFE: index (%d) out of range",
2541 state->fe_map.entries[in->index].valid = FS_STATE_IDX_VALID;
2542 state->fe_map.entries[in->index].old_idx = in->index;
2543 state->fe_map.entries[in->index].new_idx = fetoi(out);
2550 cb_stateCBToDiskEntry(struct CallBack * in, struct CBDiskEntry * out)
2552 memcpy(&out->cb, in, sizeof(struct CallBack));
2553 out->index = cbtoi(in);
2558 cb_stateDiskEntryToCB(struct fs_dump_state * state,
2559 struct CBDiskEntry * in, struct CallBack * out)
2563 memcpy(out, &in->cb, sizeof(struct CallBack));
2565 /* setup CB map entry */
2566 if (!in->index || (in->index >= state->cb_map.len)) {
2567 ViceLog(0, ("cb_stateDiskEntryToCB: index (%d) out of range\n",
2572 state->cb_map.entries[in->index].valid = FS_STATE_IDX_VALID;
2573 state->cb_map.entries[in->index].old_idx = in->index;
2574 state->cb_map.entries[in->index].new_idx = cbtoi(out);
2580 /* index map routines */
2582 cb_stateAllocMap(struct fs_dump_state * state)
2584 state->fe_map.len = state->cb_hdr->fe_max + 1;
2585 state->cb_map.len = state->cb_hdr->cb_max + 1;
2586 state->fe_map.entries = (struct idx_map_entry_t *)
2587 calloc(state->fe_map.len, sizeof(struct idx_map_entry_t));
2588 state->cb_map.entries = (struct idx_map_entry_t *)
2589 calloc(state->cb_map.len, sizeof(struct idx_map_entry_t));
2590 return ((state->fe_map.entries != NULL) && (state->cb_map.entries != NULL)) ? 0 : 1;
2594 fe_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2598 /* FEs use a one-based indexing system, so old==0 implies no mapping */
2604 if (old >= state->fe_map.len) {
2605 ViceLog(0, ("fe_OldToNew: index %d is out of range\n", old));
2607 } else if (state->fe_map.entries[old].valid != FS_STATE_IDX_VALID ||
2608 state->fe_map.entries[old].old_idx != old) { /* sanity check */
2609 ViceLog(0, ("fe_OldToNew: index %d points to an invalid FileEntry record\n", old));
2612 *new = state->fe_map.entries[old].new_idx;
2620 cb_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2624 /* CBs use a one-based indexing system, so old==0 implies no mapping */
2630 if (old >= state->cb_map.len) {
2631 ViceLog(0, ("cb_OldToNew: index %d is out of range\n", old));
2633 } else if (state->cb_map.entries[old].valid != FS_STATE_IDX_VALID ||
2634 state->cb_map.entries[old].old_idx != old) { /* sanity check */
2635 ViceLog(0, ("cb_OldToNew: index %d points to an invalid CallBack record\n", old));
2638 *new = state->cb_map.entries[old].new_idx;
2644 #endif /* AFS_DEMAND_ATTACH_FS */
2647 DumpCallBackState_r(void)
2650 afs_uint32 magic = MAGICV2, now = (afs_int32) FT_ApproxTime(), freelisthead;
2652 oflag = O_WRONLY | O_CREAT | O_TRUNC;
2656 fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, oflag, 0666);
2659 ("Couldn't create callback dump file %s\n",
2660 AFSDIR_SERVER_CBKDUMP_FILEPATH));
2663 (void)write(fd, &magic, sizeof(magic));
2664 (void)write(fd, &now, sizeof(now));
2665 (void)write(fd, &cbstuff, sizeof(cbstuff));
2666 (void)write(fd, TimeOuts, sizeof(TimeOuts));
2667 (void)write(fd, timeout, sizeof(timeout));
2668 (void)write(fd, &tfirst, sizeof(tfirst));
2669 freelisthead = cbtoi((struct CallBack *)CBfree);
2670 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2671 freelisthead = fetoi((struct FileEntry *)FEfree);
2672 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2673 (void)write(fd, HashTable, sizeof(HashTable));
2674 (void)write(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2675 (void)write(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2682 DumpCallBackState(void) {
2686 rc = DumpCallBackState_r();
2692 #endif /* !INTERPRET_DUMP */
2694 #ifdef INTERPRET_DUMP
2696 /* This is only compiled in for the callback analyzer program */
2697 /* Returns the time of the dump */
2699 ReadDump(char *file, int timebits)
2702 afs_uint32 magic, freelisthead;
2710 fd = open(file, oflag);
2712 fprintf(stderr, "Couldn't read dump file %s\n", file);
2715 read(fd, &magic, sizeof(magic));
2716 if (magic == MAGICV2) {
2719 if (magic != MAGIC) {
2721 "Magic number of %s is invalid. You might be trying to\n",
2724 "run this program on a machine type with a different byte ordering.\n");
2728 if (timebits == 64) {
2729 read(fd, &now64, sizeof(afs_int64));
2730 now = (afs_int32) now64;
2732 read(fd, &now, sizeof(afs_int32));
2734 read(fd, &cbstuff, sizeof(cbstuff));
2735 read(fd, TimeOuts, sizeof(TimeOuts));
2736 read(fd, timeout, sizeof(timeout));
2737 read(fd, &tfirst, sizeof(tfirst));
2738 read(fd, &freelisthead, sizeof(freelisthead));
2739 CB = ((struct CallBack
2740 *)(calloc(cbstuff.nblks, sizeof(struct CallBack)))) - 1;
2741 FE = ((struct FileEntry
2742 *)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
2743 CBfree = (struct CallBack *)itocb(freelisthead);
2744 read(fd, &freelisthead, sizeof(freelisthead));
2745 FEfree = (struct FileEntry *)itofe(freelisthead);
2746 read(fd, HashTable, sizeof(HashTable));
2747 read(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2748 read(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2750 perror("Error reading dumpfile");
2757 #include "AFS_component_version_number.h"
2759 #include "AFS_component_version_number.c"
2762 static afs_uint32 *cbTrack;
2765 main(int argc, char **argv)
2767 int err = 0, cbi = 0, stats = 0, noptions = 0, all = 0, vol = 0, raw = 0;
2769 struct FileEntry *fe;
2770 struct CallBack *cb;
2774 memset(&fid, 0, sizeof(fid));
2777 while (argc && **argv == '-') {
2780 if (!strcmp(*argv, "-host")) {
2786 cbi = atoi(*++argv);
2787 } else if (!strcmp(*argv, "-fid")) {
2793 fid.Volume = atoi(*++argv);
2794 fid.Vnode = atoi(*++argv);
2795 fid.Unique = atoi(*++argv);
2796 } else if (!strcmp(*argv, "-time")) {
2797 fprintf(stderr, "-time not supported\n");
2799 } else if (!strcmp(*argv, "-stats")) {
2801 } else if (!strcmp(*argv, "-all")) {
2803 } else if (!strcmp(*argv, "-raw")) {
2805 } else if (!strcmp(*argv, "-timebits")) {
2811 timebits = atoi(*++argv);
2812 if ((timebits != 32)
2816 } else if (!strcmp(*argv, "-volume")) {
2822 vol = atoi(*++argv);
2827 if (err || argc != 1) {
2829 "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] [-timebits 32"
2831 "] callbackdumpfile\n");
2833 "[cbid is shown for each host in the hosts.dump file]\n");
2836 now = ReadDump(*argv, timebits);
2837 if (stats || noptions == 0) {
2838 time_t uxtfirst = UXtime(tfirst), tnow = now;
2839 printf("The time of the dump was %u %s", (unsigned int) now, ctime(&tnow));
2840 printf("The last time cleanup ran was %u %s", (unsigned int) uxtfirst,
2842 PrintCallBackStats();
2845 cbTrack = calloc(cbstuff.nblks, sizeof(cbTrack[0]));
2850 struct CallBack *cb;
2851 struct FileEntry *fe;
2853 for (hash = 0; hash < FEHASH_SIZE; hash++) {
2854 for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
2855 if (!vol || (fe->volid == vol)) {
2856 afs_uint32 fe_i = fetoi(fe);
2858 for (cb = itocb(fe->firstcb); cb; cb = itocb(cb->cnext)) {
2859 afs_uint32 cb_i = cbtoi(cb);
2861 if (cb_i > cbstuff.nblks) {
2862 printf("CB index out of range (%u > %d), stopped for this FE\n",
2863 cb_i, cbstuff.nblks);
2867 if (cbTrack[cb_i]) {
2868 printf("CB entry already claimed for FE[%u] (this is FE[%u]), stopped\n",
2869 cbTrack[cb_i], fe_i);
2872 cbTrack[cb_i] = fe_i;
2884 afs_uint32 cfirst = cbi;
2889 } while (cbi != cfirst);
2894 printf("No callback entries for %u.%u\n", fid.Volume, fid.Vnode);
2897 cb = itocb(fe->firstcb);
2900 cb = itocb(cb->cnext);
2905 for (i = 1; i < cbstuff.nblks; i++) {
2906 p = (afs_int32 *) & FE[i];
2907 printf("%d:%12x%12x%12x%12x\n", i, p[0], p[1], p[2], p[3]);
2916 PrintCB(struct CallBack *cb, afs_uint32 now)
2918 struct FileEntry *fe = itofe(cb->fhead);
2919 time_t expires = TIndexToTime(cb->thead);
2924 printf("vol=%u vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %lu secs at %s",
2925 fe->volid, fe->vnode, fe->ncbs, cb->hhead, cb->status, fe->status,
2926 expires - now, ctime(&expires));
2931 #if !defined(INTERPRET_DUMP)
2933 ** try breaking calbacks on afidp from host. Use multi_rx.
2934 ** return 0 on success, non-zero on failure
2937 MultiBreakCallBackAlternateAddress(struct host *host, struct AFSCBFids *afidp)
2941 retVal = MultiBreakCallBackAlternateAddress_r(host, afidp);
2947 MultiBreakCallBackAlternateAddress_r(struct host *host,
2948 struct AFSCBFids *afidp)
2951 struct rx_connection **conns;
2952 struct rx_connection *connSuccess = 0;
2953 struct AddrPort *interfaces;
2954 static struct rx_securityClass *sc = 0;
2955 static struct AFSCBs tc = { 0, 0 };
2958 /* nothing more can be done */
2959 if (!host->interface)
2960 return 1; /* failure */
2962 /* the only address is the primary interface */
2963 if (host->interface->numberOfInterfaces <= 1)
2964 return 1; /* failure */
2966 /* initialise a security object only once */
2968 sc = rxnull_NewClientSecurityObject();
2970 i = host->interface->numberOfInterfaces;
2971 interfaces = calloc(i, sizeof(struct AddrPort));
2972 conns = calloc(i, sizeof(struct rx_connection *));
2973 if (!interfaces || !conns) {
2974 ViceLogThenPanic(0, ("Failed malloc in "
2975 "MultiBreakCallBackAlternateAddress_r\n"));
2978 /* initialize alternate rx connections */
2979 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
2980 /* this is the current primary address */
2981 if (host->host == host->interface->interface[i].addr &&
2982 host->port == host->interface->interface[i].port)
2985 interfaces[j] = host->interface->interface[i];
2987 rx_NewConnection(interfaces[j].addr,
2988 interfaces[j].port, 1, sc, 0);
2989 rx_SetConnDeadTime(conns[j], 2);
2990 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
2994 osi_Assert(j); /* at least one alternate address */
2996 ("Starting multibreakcall back on all addr for host %p (%s:%d)\n",
2997 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
2999 multi_Rx(conns, j) {
3000 multi_RXAFSCB_CallBack(afidp, &tc);
3004 if (host->callback_rxcon)
3005 rx_DestroyConnection(host->callback_rxcon);
3006 host->callback_rxcon = conns[multi_i];
3007 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
3008 host->host = interfaces[multi_i].addr;
3009 host->port = interfaces[multi_i].port;
3010 h_AddHostToAddrHashTable_r(host->host, host->port, host);
3011 connSuccess = conns[multi_i];
3012 rx_SetConnDeadTime(host->callback_rxcon, 50);
3013 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3015 ("multibreakcall success with addr %s:%d\n",
3016 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3017 ntohs(interfaces[multi_i].port)));
3024 /* Destroy all connections except the one on which we succeeded */
3025 for (i = 0; i < j; i++)
3026 if (conns[i] != connSuccess)
3027 rx_DestroyConnection(conns[i]);
3033 return 0; /* success */
3035 return 1; /* failure */
3040 ** try multi_RX probes to host.
3041 ** return 0 on success, non-0 on failure
3044 MultiProbeAlternateAddress_r(struct host *host)
3047 struct rx_connection **conns;
3048 struct rx_connection *connSuccess = 0;
3049 struct AddrPort *interfaces;
3050 static struct rx_securityClass *sc = 0;
3053 /* nothing more can be done */
3054 if (!host->interface)
3055 return 1; /* failure */
3057 /* the only address is the primary interface */
3058 if (host->interface->numberOfInterfaces <= 1)
3059 return 1; /* failure */
3061 /* initialise a security object only once */
3063 sc = rxnull_NewClientSecurityObject();
3065 i = host->interface->numberOfInterfaces;
3066 interfaces = calloc(i, sizeof(struct AddrPort));
3067 conns = calloc(i, sizeof(struct rx_connection *));
3068 if (!interfaces || !conns) {
3069 ViceLogThenPanic(0, ("Failed malloc in "
3070 "MultiProbeAlternateAddress_r\n"));
3073 /* initialize alternate rx connections */
3074 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3075 /* this is the current primary address */
3076 if (host->host == host->interface->interface[i].addr &&
3077 host->port == host->interface->interface[i].port)
3080 interfaces[j] = host->interface->interface[i];
3082 rx_NewConnection(interfaces[j].addr,
3083 interfaces[j].port, 1, sc, 0);
3084 rx_SetConnDeadTime(conns[j], 2);
3085 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3089 osi_Assert(j); /* at least one alternate address */
3091 ("Starting multiprobe on all addr for host %p (%s:%d)\n",
3092 host, afs_inet_ntoa_r(host->host, hoststr),
3093 ntohs(host->port)));
3095 multi_Rx(conns, j) {
3096 multi_RXAFSCB_ProbeUuid(&host->interface->uuid);
3100 if (host->callback_rxcon)
3101 rx_DestroyConnection(host->callback_rxcon);
3102 host->callback_rxcon = conns[multi_i];
3103 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
3104 host->host = interfaces[multi_i].addr;
3105 host->port = interfaces[multi_i].port;
3106 h_AddHostToAddrHashTable_r(host->host, host->port, host);
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) */