2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
9 * Portions Copyright (c) 2006 Sine Nomine Associates
13 * NEW callback package callback.c (replaces vicecb.c)
14 * Updated call back routines, NOW with:
16 * Faster DeleteVenus (Now called DeleteAllCallBacks)
17 * Call back breaking for volumes
18 * Adaptive timeouts on call backs
19 * Architected for Multi RPC
20 * No locks (currently implicit vnode locks--these will go, to)
21 * Delayed call back when rpc connection down.
22 * Bulk break of delayed call backs when rpc connection
24 * Strict limit on number of call backs.
26 * InitCallBack(nblocks)
27 * Initialize: nblocks is max number # of file entries + # of callback entries
28 * nblocks must be < 65536
29 * Space used is nblocks*16 bytes
30 * Note that space will be reclaimed by breaking callbacks of old hosts
32 * time = AddCallBack(host, fid)
34 * Returns the expiration time at the workstation.
36 * BreakCallBack(host, fid)
37 * Break all call backs for fid, except for the specified host.
40 * BreakVolumeCallBacks(volume)
41 * Break all call backs on volume, using single call to each host
42 * Delete all the call backs.
44 * DeleteCallBack(host,fid)
45 * Delete (do not break) single call back for fid.
47 * DeleteFileCallBacks(fid)
48 * Delete (do not break) all call backs for fid.
50 * DeleteAllCallBacks(host)
51 * Delete (do not break) all call backs for host.
53 * CleanupTimedOutCallBacks()
54 * Delete all timed out call back entries
55 * Must be called periodically by file server.
57 * BreakDelayedCallBacks(host)
58 * Break all delayed call backs for host.
59 * Returns 1: one or more failed, 0: success.
61 * PrintCallBackStats()
62 * Print statistics about call backs to stdout.
64 * DumpCallBacks() ---wishful thinking---
65 * Dump call back state to /tmp/callback.state.
66 * This is separately interpretable by the program pcb.
68 * Notes: In general, if a call back to a host doesn't get through,
69 * then HostDown, supplied elsewhere, is called. BreakDelayedCallBacks,
70 * however, does not call HostDown, but instead returns an indication of
71 * success if all delayed call backs were finally broken.
73 * BreakDelayedCallBacks MUST be called at the first sign of activity
74 * from the host after HostDown has been called (or a previous
75 * BreakDelayedCallBacks failed). The BreakDelayedCallBacks must be
76 * allowed to complete before any requests from that host are handled.
77 * If BreakDelayedCallBacks fails, then the host should remain
78 * down (and the request should be failed).
80 * CleanupCallBacks MUST be called periodically by the file server for
81 * this package to work correctly. Every 5 minutes is suggested.
84 #include <afsconfig.h>
85 #include <afs/param.h>
91 #include <stdlib.h> /* for malloc() */
92 #include <time.h> /* ANSI standard location for time stuff */
101 #include <afs/assert.h>
103 #include <afs/stds.h>
105 #include <afs/nfs.h> /* yuck. This is an abomination. */
108 #include <afscbint.h>
109 #include <afs/afsutil.h>
111 #include <afs/ihandle.h>
112 #include <afs/vnode.h>
113 #include <afs/volume.h>
114 #include "viced_prototypes.h"
117 #include <afs/ptclient.h> /* need definition of prlist for host.h */
119 #include "callback.h"
120 #ifdef AFS_DEMAND_ATTACH_FS
121 #include "../tviced/serialize_state.h"
122 #endif /* AFS_DEMAND_ATTACH_FS */
125 extern afsUUID FS_HostUUID;
126 extern int hostCount;
128 #ifndef INTERPRET_DUMP
129 static int ShowProblems = 1;
132 struct cbcounters cbstuff;
134 static struct FileEntry * FE = NULL; /* don't use FE[0] */
135 static struct CallBack * CB = NULL; /* don't use CB[0] */
137 static struct CallBack * CBfree = NULL;
138 static struct FileEntry * FEfree = NULL;
141 /* Time to live for call backs depends upon number of users of the file.
142 * TimeOuts is indexed by this number/8 (using TimeOut macro). Times
143 * in this table are for the workstation; server timeouts, add
146 static int TimeOuts[] = {
147 /* Note: don't make the first entry larger than 4 hours (see above) */
148 4 * 60 * 60, /* 0-7 users */
149 1 * 60 * 60, /* 8-15 users */
150 30 * 60, /* 16-23 users */
151 15 * 60, /* 24-31 users */
152 15 * 60, /* 32-39 users */
153 10 * 60, /* 40-47 users */
154 10 * 60, /* 48-55 users */
155 10 * 60, /* 56-63 users */
156 }; /* Anything more: MinTimeOut */
158 /* minimum time given for a call back */
159 #ifndef INTERPRET_DUMP
160 static int MinTimeOut = (7 * 60);
163 /* Heads of CB queues; a timeout index is 1+index into this array */
164 static afs_uint32 timeout[CB_NUM_TIMEOUT_QUEUES];
166 static afs_int32 tfirst; /* cbtime of oldest unexpired call back time queue */
169 /* 16 byte object get/free routines */
174 /* Prototypes for static routines */
175 static struct FileEntry *FindFE(register AFSFid * fid);
177 #ifndef INTERPRET_DUMP
178 static struct CallBack *iGetCB(register int *nused);
179 static int iFreeCB(register struct CallBack *cb, register int *nused);
180 static struct FileEntry *iGetFE(register int *nused);
181 static int iFreeFE(register struct FileEntry *fe, register int *nused);
182 static int TAdd(register struct CallBack *cb, register afs_uint32 * thead);
183 static int TDel(register struct CallBack *cb);
184 static int HAdd(register struct CallBack *cb, register struct host *host);
185 static int HDel(register struct CallBack *cb);
186 static int CDel(struct CallBack *cb, int deletefe);
187 static int CDelPtr(register struct FileEntry *fe, register afs_uint32 * cbp,
189 static afs_uint32 *FindCBPtr(struct FileEntry *fe, struct host *host);
190 static int FDel(register struct FileEntry *fe);
191 static int AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead,
192 int type, int locked);
193 static void MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
194 struct AFSCBFids *afidp, struct host *xhost);
195 static int MultiBreakVolumeCallBack_r(struct host *host, int isheld,
196 struct VCBParams *parms, int deletefe);
197 static int MultiBreakVolumeCallBack(struct host *host, int isheld,
199 static int MultiBreakVolumeLaterCallBack(struct host *host, int isheld,
201 static int GetSomeSpace_r(struct host *hostp, int locked);
202 static int ClearHostCallbacks_r(struct host *hp, int locked);
205 #define GetCB() ((struct CallBack *)iGetCB(&cbstuff.nCBs))
206 #define GetFE() ((struct FileEntry *)iGetFE(&cbstuff.nFEs))
207 #define FreeCB(cb) iFreeCB((struct CallBack *)cb, &cbstuff.nCBs)
208 #define FreeFE(fe) iFreeFE((struct FileEntry *)fe, &cbstuff.nFEs)
211 /* Other protos - move out sometime */
212 void PrintCB(register struct CallBack *cb, afs_uint32 now);
214 static afs_uint32 HashTable[FEHASH_SIZE]; /* File entry hash table */
216 static struct FileEntry *
217 FindFE(register AFSFid * fid)
221 register struct FileEntry *fe;
223 hash = FEHash(fid->Volume, fid->Unique);
224 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
226 if (fe->volid == fid->Volume && fe->unique == fid->Unique
227 && fe->vnode == fid->Vnode && (fe->status & FE_LATER) != FE_LATER)
233 #ifndef INTERPRET_DUMP
235 static struct CallBack *
236 iGetCB(register int *nused)
238 register struct CallBack *ret;
240 if ((ret = CBfree)) {
241 CBfree = (struct CallBack *)(((struct object *)ret)->next);
248 iFreeCB(register struct CallBack *cb, register int *nused)
250 ((struct object *)cb)->next = (struct object *)CBfree;
256 static struct FileEntry *
257 iGetFE(register int *nused)
259 register struct FileEntry *ret;
261 if ((ret = FEfree)) {
262 FEfree = (struct FileEntry *)(((struct object *)ret)->next);
269 iFreeFE(register struct FileEntry *fe, register int *nused)
271 ((struct object *)fe)->next = (struct object *)FEfree;
277 /* Add cb to end of specified timeout list */
279 TAdd(register struct CallBack *cb, register afs_uint32 * thead)
282 (*thead) = cb->tnext = cb->tprev = cbtoi(cb);
284 register struct CallBack *thp = itocb(*thead);
286 cb->tprev = thp->tprev;
290 thp->tprev = (itocb(thp->tprev)->tnext = cbtoi(cb));
292 thp->tprev = cbtoi(cb);
295 cb->thead = ttoi(thead);
299 /* Delete call back entry from timeout list */
301 TDel(register struct CallBack *cb)
303 register afs_uint32 *thead = itot(cb->thead);
305 if (*thead == cbtoi(cb))
306 *thead = (*thead == cb->tnext ? 0 : cb->tnext);
307 if (itocb(cb->tprev))
308 itocb(cb->tprev)->tnext = cb->tnext;
309 if (itocb(cb->tnext))
310 itocb(cb->tnext)->tprev = cb->tprev;
314 /* Add cb to end of specified host list */
316 HAdd(register struct CallBack *cb, register struct host *host)
318 cb->hhead = h_htoi(host);
320 host->cblist = cb->hnext = cb->hprev = cbtoi(cb);
322 register struct CallBack *fcb = itocb(host->cblist);
324 cb->hprev = fcb->hprev;
325 cb->hnext = cbtoi(fcb);
326 fcb->hprev = (itocb(fcb->hprev)->hnext = cbtoi(cb));
331 /* Delete call back entry from host list */
333 HDel(register struct CallBack *cb)
335 register afs_uint32 *hhead = &h_itoh(cb->hhead)->cblist;
337 if (*hhead == cbtoi(cb))
338 *hhead = (*hhead == cb->hnext ? 0 : cb->hnext);
339 itocb(cb->hprev)->hnext = cb->hnext;
340 itocb(cb->hnext)->hprev = cb->hprev;
344 /* Delete call back entry from fid's chain of cb's */
345 /* N.B. This one also deletes the CB, and also possibly parent FE, so
346 * make sure that it is not on any other list before calling this
349 CDel(struct CallBack *cb, int deletefe)
352 struct FileEntry *fe = itofe(cb->fhead);
353 register afs_uint32 *cbp;
356 for (safety = 0, cbp = &fe->firstcb; *cbp && *cbp != cbi;
357 cbp = &itocb(*cbp)->cnext, safety++) {
358 if (safety > cbstuff.nblks + 10) {
361 ("CDel: Internal Error -- shutting down: wanted %d from %d, now at %d\n",
362 cbi, fe->firstcb, *cbp));
364 ShutDownAndCore(PANIC);
367 CDelPtr(fe, cbp, deletefe);
371 /* Same as CDel, but pointer to parent pointer to CB entry is passed,
372 * as well as file entry */
373 /* N.B. This one also deletes the CB, and also possibly parent FE, so
374 * make sure that it is not on any other list before calling this
376 static int Ccdelpt = 0, CcdelB = 0;
379 CDelPtr(register struct FileEntry *fe, register afs_uint32 * cbp,
382 register struct CallBack *cb;
392 if ((--fe->ncbs == 0) && deletefe)
398 FindCBPtr(struct FileEntry *fe, struct host *host)
400 register afs_uint32 hostindex = h_htoi(host);
401 register struct CallBack *cb;
402 register afs_uint32 *cbp;
405 for (safety = 0, cbp = &fe->firstcb; *cbp; cbp = &cb->cnext, safety++) {
406 if (safety > cbstuff.nblks) {
407 ViceLog(0, ("FindCBPtr: Internal Error -- shutting down.\n"));
409 ShutDownAndCore(PANIC);
412 if (cb->hhead == hostindex)
418 /* Delete file entry from hash table */
420 FDel(register struct FileEntry *fe)
422 register int fei = fetoi(fe);
423 register afs_uint32 *p = &HashTable[FEHash(fe->volid, fe->unique)];
425 while (*p && *p != fei)
426 p = &itofe(*p)->fnext;
433 /* initialize the callback package */
435 InitCallBack(int nblks)
438 tfirst = CBtime(FT_ApproxTime());
439 /* N.B. The "-1", below, is because
440 * FE[0] and CB[0] are not used--and not allocated */
441 FE = ((struct FileEntry *)(calloc(nblks, sizeof(struct FileEntry))));
443 ViceLog(0, ("Failed malloc in InitCallBack\n"));
446 FE--; /* FE[0] is supposed to point to junk */
447 cbstuff.nFEs = nblks;
449 FreeFE(&FE[cbstuff.nFEs]); /* This is correct */
450 CB = ((struct CallBack *)(calloc(nblks, sizeof(struct CallBack))));
452 ViceLog(0, ("Failed malloc in InitCallBack\n"));
455 CB--; /* CB[0] is supposed to point to junk */
456 cbstuff.nCBs = nblks;
458 FreeCB(&CB[cbstuff.nCBs]); /* This is correct */
459 cbstuff.nblks = nblks;
460 cbstuff.nbreakers = 0;
466 XCallBackBulk_r(struct host * ahost, struct AFSFid * fids, afs_int32 nfids)
468 struct AFSCallBack tcbs[AFSCBMAX];
474 struct rx_connection *cb_conn = NULL;
477 rx_SetConnDeadTime(ahost->callback_rxcon, 4);
478 rx_SetConnHardDeadTime(ahost->callback_rxcon, AFS_HARDDEADTIME);
485 for (i = 0; i < nfids && i < AFSCBMAX; i++) {
486 tcbs[i].CallBackVersion = CALLBACK_VERSION;
487 tcbs[i].ExpirationTime = 0;
488 tcbs[i].CallBackType = CB_DROPPED;
490 tf.AFSCBFids_len = i;
491 tf.AFSCBFids_val = &(fids[j]);
495 tc.AFSCBs_val = tcbs;
497 cb_conn = ahost->callback_rxcon;
498 rx_GetConnection(cb_conn);
500 code |= RXAFSCB_CallBack(cb_conn, &tf, &tc);
501 rx_PutConnection(cb_conn);
509 /* the locked flag tells us if the host entry has already been locked
510 * by our parent. I don't think anybody actually calls us with the
511 * host locked, but here's how to make that work: GetSomeSpace has to
512 * change so that it doesn't attempt to lock any hosts < "host". That
513 * means that it might be unable to free any objects, so it has to
514 * return an exit status. If it fails, then AddCallBack1 might fail,
515 * as well. If so, the host->ResetDone should probably be set to 0,
516 * and we probably don't want to return a callback promise to the
517 * cache manager, either. */
519 AddCallBack1(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
527 retVal = AddCallBack1_r(host, fid, thead, type, 1);
537 AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
540 struct FileEntry *fe;
541 struct CallBack *cb = 0, *lastcb = 0;
542 struct FileEntry *newfe = 0;
543 afs_uint32 time_out = 0;
544 afs_uint32 *Thead = thead;
545 struct CallBack *newcb = 0;
548 cbstuff.AddCallBacks++;
552 /* allocate these guys first, since we can't call the allocator with
553 * the host structure locked -- or we might deadlock. However, we have
554 * to avoid races with FindFE... */
555 while (!(newcb = GetCB())) {
556 GetSomeSpace_r(host, locked);
558 while (!(newfe = GetFE())) { /* Get it now, so we don't have to call */
559 /* GetSomeSpace with the host locked, later. This might turn out to */
560 /* have been unneccessary, but that's actually kind of unlikely, since */
561 /* most files are not shared. */
562 GetSomeSpace_r(host, locked);
566 h_Lock_r(host); /* this can yield, so do it before we get any */
571 if (type == CB_NORMAL) {
573 TimeCeiling(FT_ApproxTime() + TimeOut(fe ? fe->ncbs : 0) +
575 Thead = THead(CBtime(time_out));
576 } else if (type == CB_VOLUME) {
577 time_out = TimeCeiling((60 * 120 + FT_ApproxTime()) + ServerBias);
578 Thead = THead(CBtime(time_out));
579 } else if (type == CB_BULK) {
580 /* bulk status can get so many callbacks all at once, and most of them
581 * are probably not for things that will be used for long.
584 TimeCeiling(FT_ApproxTime() + ServerBias +
585 TimeOut(22 + (fe ? fe->ncbs : 0)));
586 Thead = THead(CBtime(time_out));
592 register afs_uint32 hash;
597 fe->volid = fid->Volume;
598 fe->vnode = fid->Vnode;
599 fe->unique = fid->Unique;
602 hash = FEHash(fid->Volume, fid->Unique);
603 fe->fnext = HashTable[hash];
604 HashTable[hash] = fetoi(fe);
606 for (safety = 0, lastcb = cb = itocb(fe->firstcb); cb;
607 lastcb = cb, cb = itocb(cb->cnext), safety++) {
608 if (safety > cbstuff.nblks) {
609 ViceLog(0, ("AddCallBack1: Internal Error -- shutting down.\n"));
611 ShutDownAndCore(PANIC);
613 if (cb->hhead == h_htoi(host))
616 if (cb) { /* Already have call back: move to new timeout list */
617 /* don't change delayed callbacks back to normal ones */
618 if (cb->status != CB_DELAYED)
620 /* Only move if new timeout is longer */
621 if (TNorm(ttoi(Thead)) > TNorm(cb->thead)) {
625 if (newfe == NULL) { /* we are using the new FE */
626 fe->firstcb = cbtoi(cb);
628 cb->fhead = fetoi(fe);
633 *(lastcb ? &lastcb->cnext : &fe->firstcb) = cbtoi(cb);
636 cb->fhead = fetoi(fe);
642 /* now free any still-unused callback or host entries */
648 if (!locked) /* freecb and freefe might(?) yield */
651 if (type == CB_NORMAL || type == CB_VOLUME || type == CB_BULK)
652 return time_out - ServerBias; /* Expires sooner at workstation */
658 CompareCBA(const void *e1, const void *e2)
660 const struct cbstruct *cba1 = (const struct cbstruct *)e1;
661 const struct cbstruct *cba2 = (const struct cbstruct *)e2;
662 return ((cba1->hp)->index - (cba2->hp)->index);
665 /* Take an array full of hosts, all held. Break callbacks to them, and
666 * release the holds once you're done, except don't release xhost. xhost
667 * may be NULL. Currently only works for a single Fid in afidp array.
668 * If you want to make this work with multiple fids, you need to fix
669 * the error handling. One approach would be to force a reset if a
670 * multi-fid call fails, or you could add delayed callbacks for each
671 * fid. You probably also need to sort and remove duplicate hosts.
672 * When this is called from the BreakVolumeCallBacks path, it does NOT
673 * force a reset if the RPC fails, it just marks the host down and tries
674 * to create a delayed callback. */
675 /* N.B. be sure that code works when ncbas == 0 */
676 /* N.B. requires all the cba[*].hp pointers to be valid... */
677 /* This routine does not hold a lock on the host for the duration of
678 * the BreakCallBack RPC, which is a significant deviation from tradition.
679 * It _does_ get a lock on the host before setting VenusDown = 1,
680 * which is sufficient only if VenusDown = 0 only happens when the
681 * lock is held over the RPC and the subsequent VenusDown == 0
682 * wherever that is done. */
684 MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
685 struct AFSCBFids *afidp, struct host *xhost)
688 struct rx_connection *conns[MAX_CB_HOSTS];
689 static struct AFSCBs tc = { 0, 0 };
690 int multi_to_cba_map[MAX_CB_HOSTS];
692 assert(ncbas <= MAX_CB_HOSTS);
694 /* sort cba list to avoid makecall issues */
695 qsort(cba, ncbas, sizeof(struct cbstruct), CompareCBA);
697 /* set up conns for multi-call */
698 for (i = 0, j = 0; i < ncbas; i++) {
699 struct host *thishost = cba[i].hp;
700 if (!thishost || (thishost->hostFlags & HOSTDELETED)) {
703 rx_GetConnection(thishost->callback_rxcon);
704 multi_to_cba_map[j] = i;
705 conns[j++] = thishost->callback_rxcon;
708 rx_SetConnDeadTime(thishost->callback_rxcon, 4);
709 rx_SetConnHardDeadTime(thishost->callback_rxcon, AFS_HARDDEADTIME);
713 if (j) { /* who knows what multi would do with 0 conns? */
717 multi_RXAFSCB_CallBack(afidp, &tc);
723 i = multi_to_cba_map[multi_i];
729 ("BCB: INTERNAL ERROR: hp=%x, cba=%x, thead=%u\n",
733 ** try breaking callbacks on alternate interface addresses
735 if (MultiBreakCallBackAlternateAddress(hp, afidp)) {
738 ("BCB: Failed on file %u.%u.%u, Host %x (%s:%d) is down\n",
739 afidp->AFSCBFids_val->Volume,
740 afidp->AFSCBFids_val->Vnode,
741 afidp->AFSCBFids_val->Unique,
743 afs_inet_ntoa_r(hp->host, hoststr),
749 hp->hostFlags |= VENUSDOWN;
751 * We always go into AddCallBack1_r with the host locked
753 AddCallBack1_r(hp, afidp->AFSCBFids_val, itot(idx),
766 for (i = 0; i < ncbas; i++) {
769 if (hp && xhost != hp) {
774 /* H_UNLOCK around this so h_FreeConnection does not deadlock.
775 h_FreeConnection should *never* be called on a callback connection,
776 but on 10/27/04 a deadlock occurred where it was, when we know why,
777 this should be reverted. -- shadow */
779 for (i = 0; i < j; i++) {
780 rx_PutConnection(conns[i]);
788 * Break all call backs for fid, except for the specified host (unless flag
789 * is true, in which case all get a callback message. Assumption: the specified
790 * host is h_Held, by the caller; the others aren't.
791 * Specified host may be bogus, that's ok. This used to check to see if the
792 * host was down in two places, once right after the host was h_held, and
793 * again after it was locked. That race condition is incredibly rare and
794 * relatively harmless even when it does occur, so we don't check for it now.
796 /* if flag is true, send a break callback msg to "host", too */
798 BreakCallBack(struct host *xhost, AFSFid * fid, int flag)
800 struct FileEntry *fe;
801 struct CallBack *cb, *nextcb;
802 struct cbstruct cba[MAX_CB_HOSTS];
809 ("BCB: BreakCallBack(Host %x all but %s:%d, (%u,%u,%u))\n",
810 xhost, afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
811 fid->Volume, fid->Vnode, fid->Unique));
814 cbstuff.BreakCallBacks++;
819 hostindex = h_htoi(xhost);
820 cb = itocb(fe->firstcb);
821 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
822 /* the most common case is what follows the || */
825 tf.AFSCBFids_len = 1;
826 tf.AFSCBFids_val = fid;
829 for (ncbas = 0; cb && ncbas < MAX_CB_HOSTS; cb = nextcb) {
830 nextcb = itocb(cb->cnext);
831 if ((cb->hhead != hostindex || flag)
832 && (cb->status == CB_BULK || cb->status == CB_NORMAL
833 || cb->status == CB_VOLUME)) {
834 struct host *thishost = h_itoh(cb->hhead);
836 ViceLog(0, ("BCB: BOGUS! cb->hhead is NULL!\n"));
837 } else if (thishost->hostFlags & VENUSDOWN) {
839 ("BCB: %x (%s:%d) is down; delaying break call back\n",
840 thishost, afs_inet_ntoa_r(thishost->host, hoststr),
841 ntohs(thishost->port)));
842 cb->status = CB_DELAYED;
845 cba[ncbas].hp = thishost;
846 cba[ncbas].thead = cb->thead;
850 CDel(cb, 1); /* Usually first; so this delete
851 * is reasonably inexpensive */
857 MultiBreakCallBack_r(cba, ncbas, &tf, xhost);
859 /* we need to to all these initializations again because MultiBreakCallBack may block */
864 cb = itocb(fe->firstcb);
865 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
866 /* the most common case is what follows the || */
877 /* Delete (do not break) single call back for fid */
879 DeleteCallBack(struct host *host, AFSFid * fid)
881 register struct FileEntry *fe;
882 register afs_uint32 *pcb;
886 cbstuff.DeleteCallBacks++;
894 ("DCB: No call backs for fid (%u, %u, %u)\n", fid->Volume,
895 fid->Vnode, fid->Unique));
898 pcb = FindCBPtr(fe, host);
901 ("DCB: No call back for host %x (%s:%d), (%u, %u, %u)\n",
902 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
903 fid->Volume, fid->Vnode, fid->Unique));
917 * Delete (do not break) all call backs for fid. This call doesn't
918 * set all of the various host locks, but it shouldn't really matter
919 * since we're not adding callbacks, but deleting them. I'm not sure
920 * why it doesn't set the lock, however; perhaps it should.
923 DeleteFileCallBacks(AFSFid * fid)
925 register struct FileEntry *fe;
926 register struct CallBack *cb;
927 register afs_uint32 cbi;
931 cbstuff.DeleteFiles++;
936 ("DF: No fid (%u,%u,%u) to delete\n", fid->Volume, fid->Vnode,
940 for (n = 0, cbi = fe->firstcb; cbi; n++) {
953 /* Delete (do not break) all call backs for host. The host should be
956 DeleteAllCallBacks_r(struct host *host, int deletefe)
958 register struct CallBack *cb;
959 register int cbi, first;
961 cbstuff.DeleteAllCallBacks++;
962 cbi = first = host->cblist;
964 ViceLog(8, ("DV: no call backs\n"));
972 } while (cbi != first);
978 * Break all delayed call backs for host. Returns 1 if all call backs
979 * successfully broken; 0 otherwise. Assumes host is h_Held and h_Locked.
980 * Must be called with VenusDown set for this host
983 BreakDelayedCallBacks(struct host *host)
987 retVal = BreakDelayedCallBacks_r(host);
993 BreakDelayedCallBacks_r(struct host *host)
995 struct AFSFid fids[AFSCBMAX];
996 u_byte thead[AFSCBMAX]; /* This should match thead in struct Callback */
997 int cbi, first, nfids;
1001 struct rx_connection *cb_conn;
1003 cbstuff.nbreakers++;
1004 if (!(host->hostFlags & RESETDONE) && !(host->hostFlags & HOSTDELETED)) {
1005 host->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1006 cb_conn = host->callback_rxcon;
1007 rx_GetConnection(cb_conn);
1008 if (host->interface) {
1011 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1014 code = RXAFSCB_InitCallBackState(cb_conn);
1016 rx_PutConnection(cb_conn);
1019 host->hostFlags |= ALTADDR; /* alternate addresses are valid */
1023 ("CB: Call back connect back failed (in break delayed) for Host %x (%s:%d)\n",
1024 host, afs_inet_ntoa_r(host->host, hoststr),
1025 ntohs(host->port)));
1027 host->hostFlags |= VENUSDOWN;
1030 ("InitCallBackState success on %x (%s:%d)\n",
1031 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
1032 /* reset was done successfully */
1033 host->hostFlags |= RESETDONE;
1034 host->hostFlags &= ~VENUSDOWN;
1037 while (!(host->hostFlags & HOSTDELETED)) {
1039 host->hostFlags &= ~VENUSDOWN; /* presume up */
1040 cbi = first = host->cblist;
1044 first = host->cblist;
1047 if (cb->status == CB_DELAYED) {
1048 register struct FileEntry *fe = itofe(cb->fhead);
1049 thead[nfids] = cb->thead;
1050 fids[nfids].Volume = fe->volid;
1051 fids[nfids].Vnode = fe->vnode;
1052 fids[nfids].Unique = fe->unique;
1058 } while (cbi && cbi != first && nfids < AFSCBMAX);
1064 if (XCallBackBulk_r(host, fids, nfids)) {
1065 /* Failed, again: put them back, probably with old
1070 ("CB: XCallBackBulk failed, Host %x (%s:%d); callback list follows:\n",
1071 host, afs_inet_ntoa_r(host->host, hoststr),
1072 ntohs(host->port)));
1074 for (i = 0; i < nfids; i++) {
1077 ("CB: Host %x (%s:%d), file %u.%u.%u (part of bulk callback)\n",
1078 host, afs_inet_ntoa_r(host->host, hoststr),
1079 ntohs(host->port), fids[i].Volume,
1080 fids[i].Vnode, fids[i].Unique));
1083 * AddCallBack1_r(host, &fids[i], itot(thead[i]), CB_DELAYED, 1);
1084 * * but it turns out to cause too many tricky locking problems.
1085 * * now, if break delayed fails, screw it. */
1087 host->hostFlags |= VENUSDOWN; /* Failed */
1088 ClearHostCallbacks_r(host, 1 /* locked */ );
1092 if (nfids < AFSCBMAX)
1096 cbstuff.nbreakers--;
1097 /* If we succeeded it's always ok to unset HFE_LATER */
1098 if (!(host->hostFlags & VENUSDOWN))
1099 host->hostFlags &= ~HFE_LATER;
1100 return (host->hostFlags & VENUSDOWN);
1104 ** isheld is 0 if the host is held in h_Enumerate
1105 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1108 MultiBreakVolumeCallBack_r(struct host *host, int isheld,
1109 struct VCBParams *parms, int deletefe)
1114 return isheld; /* host is held only by h_Enumerate, do nothing */
1115 if (host->hostFlags & HOSTDELETED)
1116 return 0; /* host is deleted, release hold */
1118 if (host->hostFlags & VENUSDOWN) {
1120 if (host->hostFlags & HOSTDELETED) {
1122 return 0; /* Release hold */
1125 ("BVCB: volume call back for Host %x (%s:%d) failed\n",
1126 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
1129 ("CB: volume callback for Host %x (%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; /* Do InitCallBackState when host returns */
1140 return 0; /* release hold */
1142 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, 0 /* xhost */ );
1158 parms->cba[parms->ncbas].hp = host;
1159 parms->cba[(parms->ncbas)++].thead = parms->thead;
1160 return 1; /* DON'T release hold, because we still need it. */
1164 ** isheld is 0 if the host is held in h_Enumerate
1165 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1168 MultiBreakVolumeCallBack(struct host *host, int isheld, void *rock)
1170 struct VCBParams *parms = (struct VCBParams *) rock;
1174 retval = MultiBreakVolumeCallBack_r(host, isheld, parms, 1);
1180 ** isheld is 0 if the host is held in h_Enumerate
1181 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1184 MultiBreakVolumeLaterCallBack(struct host *host, int isheld, void *rock)
1186 struct VCBParams *parms = (struct VCBParams *)rock;
1189 retval = MultiBreakVolumeCallBack_r(host, isheld, parms, 0);
1195 * Break all call backs on a single volume. Don't call this with any
1196 * hosts h_held. Note that this routine clears the callbacks before
1197 * actually breaking them, and that the vnode isn't locked during this
1198 * operation, so that people might see temporary callback loss while
1199 * this function is executing. It is just a temporary state, however,
1200 * since the callback will be broken later by this same function.
1202 * Now uses multi-RX for CallBack RPC. Note that the
1203 * multiBreakCallBacks routine does not force a reset if the RPC
1204 * fails, unlike the previous version of this routine, but does create
1205 * a delayed callback. Resets will be forced if the host is
1206 * determined to be down before the RPC is executed.
1209 BreakVolumeCallBacks(afs_uint32 volume)
1214 struct CallBack *cb;
1215 struct FileEntry *fe;
1217 struct VCBParams henumParms;
1218 afs_uint32 tthead = 0; /* zero is illegal value */
1221 fid.Volume = volume, fid.Vnode = fid.Unique = 0;
1222 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1223 for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
1224 if (fe->volid == volume) {
1225 register struct CallBack *cbnext;
1226 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1227 host = h_itoh(cb->hhead);
1229 cbnext = itocb(cb->cnext);
1230 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1236 /* leave hold for MultiBreakVolumeCallBack to clear */
1247 /* didn't find any callbacks, so return right away. */
1251 henumParms.ncbas = 0;
1252 henumParms.fid = &fid;
1253 henumParms.thead = tthead;
1255 h_Enumerate(MultiBreakVolumeCallBack, &henumParms);
1257 if (henumParms.ncbas) { /* do left-overs */
1258 struct AFSCBFids tf;
1259 tf.AFSCBFids_len = 1;
1260 tf.AFSCBFids_val = &fid;
1262 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf, 0);
1264 henumParms.ncbas = 0;
1270 #ifdef AFS_PTHREAD_ENV
1271 extern pthread_cond_t fsync_cond;
1273 extern char fsync_wait[];
1277 BreakVolumeCallBacksLater(afs_uint32 volume)
1281 struct FileEntry *fe;
1282 struct CallBack *cb;
1286 ViceLog(25, ("Setting later on volume %u\n", volume));
1288 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1289 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1290 if (fe->volid == volume) {
1291 register struct CallBack *cbnext;
1292 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1293 host = h_itoh(cb->hhead);
1294 host->hostFlags |= HFE_LATER;
1295 cb->status = CB_DELAYED;
1296 cbnext = itocb(cb->cnext);
1299 fe->status |= FE_LATER;
1308 /* didn't find any callbacks, so return right away. */
1312 ViceLog(25, ("Fsync thread wakeup\n"));
1313 #ifdef AFS_PTHREAD_ENV
1315 assert(pthread_cond_broadcast(&fsync_cond) == 0);
1318 LWP_NoYieldSignal(fsync_wait);
1324 BreakLaterCallBacks(void)
1329 struct CallBack *cb;
1330 struct FileEntry *fe = NULL;
1331 struct FileEntry *myfe = NULL;
1333 struct VCBParams henumParms;
1334 unsigned short tthead = 0; /* zero is illegal value */
1338 ViceLog(25, ("Looking for FileEntries to unchain\n"));
1341 /* Pick the first volume we see to clean up */
1342 fid.Volume = fid.Vnode = fid.Unique = 0;
1344 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1345 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1346 if (fe && (fe->status & FE_LATER)
1347 && (fid.Volume == 0 || fid.Volume == fe->volid)) {
1348 /* Ugly, but used to avoid left side casting */
1349 struct object *tmpfe;
1351 ("Unchaining for %u:%u:%u\n", fe->vnode, fe->unique,
1353 fid.Volume = fe->volid;
1355 /* Works since volid is deeper than the largest pointer */
1356 tmpfe = (struct object *)fe;
1357 tmpfe->next = (struct object *)myfe;
1370 /* loop over FEs from myfe and free/break */
1372 for (fe = myfe; fe;) {
1373 register struct CallBack *cbnext;
1374 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1375 cbnext = itocb(cb->cnext);
1376 host = h_itoh(cb->hhead);
1377 if (cb->status == CB_DELAYED) {
1379 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1384 CDel(cb, 0); /* Don't let CDel clean up the fe */
1385 /* leave hold for MultiBreakVolumeCallBack to clear */
1388 ("Found host %x (%s:%d) non-DELAYED cb for %u:%u:%u\n",
1389 host, afs_inet_ntoa_r(host->host, hoststr),
1390 ntohs(host->port), fe->vnode, fe->unique, fe->volid));
1394 fe = (struct FileEntry *)((struct object *)fe)->next;
1399 ViceLog(125, ("Breaking volume %u\n", fid.Volume));
1400 henumParms.ncbas = 0;
1401 henumParms.fid = &fid;
1402 henumParms.thead = tthead;
1404 h_Enumerate(MultiBreakVolumeLaterCallBack, (char *)&henumParms);
1406 if (henumParms.ncbas) { /* do left-overs */
1407 struct AFSCBFids tf;
1408 tf.AFSCBFids_len = 1;
1409 tf.AFSCBFids_val = &fid;
1411 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf, 0);
1412 henumParms.ncbas = 0;
1417 /* Arrange to be called again */
1422 * Delete all timed-out call back entries (to be called periodically by file
1426 CleanupTimedOutCallBacks(void)
1429 CleanupTimedOutCallBacks_r();
1435 CleanupTimedOutCallBacks_r(void)
1437 afs_uint32 now = CBtime(FT_ApproxTime());
1438 register afs_uint32 *thead;
1439 register struct CallBack *cb;
1440 register int ntimedout = 0;
1443 while (tfirst <= now) {
1445 cbi = *(thead = THead(tfirst));
1451 ("CCB: deleting timed out call back %x (%s:%d), (%u,%u,%u)\n",
1452 h_itoh(cb->hhead)->host,
1453 afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
1454 h_itoh(cb->hhead)->port, itofe(cb->fhead)->volid,
1455 itofe(cb->fhead)->vnode, itofe(cb->fhead)->unique));
1459 if (ntimedout > cbstuff.nblks) {
1460 ViceLog(0, ("CCB: Internal Error -- shutting down...\n"));
1461 DumpCallBackState();
1462 ShutDownAndCore(PANIC);
1464 } while (cbi != *thead);
1469 cbstuff.CBsTimedOut += ntimedout;
1470 ViceLog(7, ("CCB: deleted %d timed out callbacks\n", ntimedout));
1471 return (ntimedout > 0);
1474 static struct host *lih_host;
1475 static int lih_host_held;
1477 /* This version does not allow 'host' to be selected unless its ActiveCall
1478 * is newer than 'hostp' which is the host with the oldest ActiveCall from
1479 * the last pass (if it is provided). We filter out any hosts that are
1480 * are held by other threads.
1483 lih0_r(register struct host *host, register int held, void *rock)
1485 struct host *hostp = (struct host *) rock;
1487 && (hostp && host != hostp)
1488 && (!held && !h_OtherHolds_r(host))
1489 && (!lih_host || host->ActiveCall < lih_host->ActiveCall)
1490 && (!hostp || host->ActiveCall > hostp->ActiveCall)) {
1491 if (lih_host != NULL && lih_host_held) {
1492 h_Release_r(lih_host);
1495 lih_host_held = !held;
1501 /* This version does not allow 'host' to be selected unless its ActiveCall
1502 * is newer than 'hostp' which is the host with the oldest ActiveCall from
1503 * the last pass (if it is provided). In this second varient, we do not
1504 * prevent held hosts from being selected.
1507 lih1_r(register struct host *host, register int held, void *rock)
1509 struct host *hostp = (struct host *) rock;
1512 && (hostp && host != hostp)
1513 && (!lih_host || host->ActiveCall < lih_host->ActiveCall)
1514 && (!hostp || host->ActiveCall > hostp->ActiveCall)) {
1515 if (lih_host != NULL && lih_host_held) {
1516 h_Release_r(lih_host);
1519 lih_host_held = !held;
1525 /* This could be upgraded to get more space each time */
1526 /* first pass: sequentially find the oldest host which isn't held by
1527 anyone for which we can clear callbacks;
1529 /* second pass: sequentially find the oldest host regardless of
1530 whether or not the host is held; skipping 'hostp' */
1531 /* third pass: attempt to clear callbacks from 'hostp' */
1532 /* always called with hostp unlocked */
1534 /* Note: hostlist is ordered most recently created host first and
1535 * its order has no relationship to the most recently used. */
1536 extern struct host *hostList;
1538 GetSomeSpace_r(struct host *hostp, int locked)
1540 register struct host *hp, *hp1, *hp2;
1543 cbstuff.GotSomeSpaces++;
1545 ("GSS: First looking for timed out call backs via CleanupCallBacks\n"));
1546 if (CleanupTimedOutCallBacks_r()) {
1556 h_Enumerate_r(i == 0 ? lih0_r : lih1_r, hp2, (char *)hp1);
1559 /* set in lih_r! private copy before giving up H_LOCK */
1560 int lih_host_held2=lih_host_held;
1562 if ((hp != hostp) && !ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
1573 * Next time try getting callbacks from any host even if
1574 * it's deleted (that's actually great since we can freely
1575 * remove its callbacks) or it's held since the only other
1576 * option is starvation for the file server (i.e. until the
1577 * callback timeout arrives).
1584 ("GSS: Try harder for longest inactive host cnt= %d\n",
1589 /* Could not obtain space from other hosts, clear hostp's callback state */
1594 ClearHostCallbacks_r(hostp, 1 /*already locked */ );
1601 /* locked - set if caller has already locked the host */
1603 ClearHostCallbacks_r(struct host *hp, int locked)
1608 struct rx_connection *cb_conn = NULL;
1611 ("GSS: Delete longest inactive host %x (%s:%d)\n",
1612 hp, afs_inet_ntoa_r(hp->host, hoststr), ntohs(hp->port)));
1613 if (!(held = h_Held_r(hp)))
1616 /** Try a non-blocking lock. If the lock is already held return
1617 * after releasing hold on hp
1620 if (h_NBLock_r(hp)) {
1626 if (hp->Console & 2) {
1628 * If the special console field is set it means that a thread
1629 * is waiting in AddCallBack1 after it set pointers to the
1630 * file entry and/or callback entry. Because of the bogus
1631 * usage of h_hold it won't prevent from another thread, this
1632 * one, to remove all the callbacks so just to be safe we keep
1633 * a reference. NOTE, on the last phase we'll free the calling
1634 * host's callbacks but that's ok...
1638 DeleteAllCallBacks_r(hp, 1);
1639 if (hp->hostFlags & VENUSDOWN) {
1640 hp->hostFlags &= ~RESETDONE; /* remember that we must do a reset */
1642 /* host is up, try a call */
1643 hp->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1644 cb_conn = hp->callback_rxcon;
1645 rx_GetConnection(hp->callback_rxcon);
1646 if (hp->interface) {
1649 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1652 code = RXAFSCB_InitCallBackState(cb_conn);
1654 rx_PutConnection(cb_conn);
1657 hp->hostFlags |= ALTADDR; /* alternate addresses are valid */
1659 /* failed, mark host down and need reset */
1660 hp->hostFlags |= VENUSDOWN;
1661 hp->hostFlags &= ~RESETDONE;
1663 /* reset succeeded, we're done */
1664 hp->hostFlags |= RESETDONE;
1675 #endif /* INTERPRET_DUMP */
1679 PrintCallBackStats(void)
1682 "%d add CB, %d break CB, %d del CB, %d del FE, %d CB's timed out, %d space reclaim, %d del host\n",
1683 cbstuff.AddCallBacks, cbstuff.BreakCallBacks,
1684 cbstuff.DeleteCallBacks, cbstuff.DeleteFiles, cbstuff.CBsTimedOut,
1685 cbstuff.GotSomeSpaces, cbstuff.DeleteAllCallBacks);
1686 fprintf(stderr, "%d CBs, %d FEs, (%d of total of %d 16-byte blocks)\n",
1687 cbstuff.nCBs, cbstuff.nFEs, cbstuff.nCBs + cbstuff.nFEs,
1693 #define MAGIC 0x12345678 /* To check byte ordering of dump when it is read in */
1695 #ifndef INTERPRET_DUMP
1697 #ifdef AFS_DEMAND_ATTACH_FS
1700 * callback state serialization
1702 static int cb_stateSaveTimeouts(struct fs_dump_state * state);
1703 static int cb_stateSaveFEHash(struct fs_dump_state * state);
1704 static int cb_stateSaveFEs(struct fs_dump_state * state);
1705 static int cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe);
1706 static int cb_stateRestoreTimeouts(struct fs_dump_state * state);
1707 static int cb_stateRestoreFEHash(struct fs_dump_state * state);
1708 static int cb_stateRestoreFEs(struct fs_dump_state * state);
1709 static int cb_stateRestoreFE(struct fs_dump_state * state);
1710 static int cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
1711 struct iovec * iov, int niovecs);
1713 static int cb_stateVerifyFEHash(struct fs_dump_state * state);
1714 static int cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe);
1715 static int cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe);
1716 static int cb_stateVerifyTimeoutQueues(struct fs_dump_state * state);
1718 static int cb_stateFEToDiskEntry(struct FileEntry *, struct FEDiskEntry *);
1719 static int cb_stateDiskEntryToFE(struct fs_dump_state * state,
1720 struct FEDiskEntry *, struct FileEntry *);
1722 static int cb_stateCBToDiskEntry(struct CallBack *, struct CBDiskEntry *);
1723 static int cb_stateDiskEntryToCB(struct fs_dump_state * state,
1724 struct CBDiskEntry *, struct CallBack *);
1726 static int cb_stateFillHeader(struct callback_state_header * hdr);
1727 static int cb_stateCheckHeader(struct callback_state_header * hdr);
1729 static int cb_stateAllocMap(struct fs_dump_state * state);
1732 cb_stateSave(struct fs_dump_state * state)
1736 AssignInt64(state->eof_offset, &state->hdr->cb_offset);
1738 /* invalidate callback state header */
1739 memset(state->cb_hdr, 0, sizeof(struct callback_state_header));
1740 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1741 sizeof(struct callback_state_header))) {
1746 fs_stateIncEOF(state, sizeof(struct callback_state_header));
1748 /* dump timeout state */
1749 if (cb_stateSaveTimeouts(state)) {
1754 /* dump fe hashtable state */
1755 if (cb_stateSaveFEHash(state)) {
1760 /* dump callback state */
1761 if (cb_stateSaveFEs(state)) {
1766 /* write the callback state header to disk */
1767 cb_stateFillHeader(state->cb_hdr);
1768 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1769 sizeof(struct callback_state_header))) {
1779 cb_stateRestore(struct fs_dump_state * state)
1783 if (fs_stateReadHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1784 sizeof(struct callback_state_header))) {
1789 if (cb_stateCheckHeader(state->cb_hdr)) {
1794 if (cb_stateAllocMap(state)) {
1799 if (cb_stateRestoreTimeouts(state)) {
1804 if (cb_stateRestoreFEHash(state)) {
1809 /* restore FEs and CBs from disk */
1810 if (cb_stateRestoreFEs(state)) {
1815 /* restore the timeout queue heads */
1816 tfirst = state->cb_hdr->tfirst;
1823 cb_stateRestoreIndices(struct fs_dump_state * state)
1826 struct FileEntry * fe;
1827 struct CallBack * cb;
1829 /* restore indices in the FileEntry structures */
1830 for (i = 1; i < state->fe_map.len; i++) {
1831 if (state->fe_map.entries[i].new_idx) {
1832 fe = itofe(state->fe_map.entries[i].new_idx);
1834 /* restore the fe->fnext entry */
1835 if (fe_OldToNew(state, fe->fnext, &fe->fnext)) {
1840 /* restore the fe->firstcb entry */
1841 if (cb_OldToNew(state, fe->firstcb, &fe->firstcb)) {
1848 /* restore indices in the CallBack structures */
1849 for (i = 1; i < state->cb_map.len; i++) {
1850 if (state->cb_map.entries[i].new_idx) {
1851 cb = itocb(state->cb_map.entries[i].new_idx);
1853 /* restore the cb->cnext entry */
1854 if (cb_OldToNew(state, cb->cnext, &cb->cnext)) {
1859 /* restore the cb->fhead entry */
1860 if (fe_OldToNew(state, cb->fhead, &cb->fhead)) {
1865 /* restore the cb->hhead entry */
1866 if (h_OldToNew(state, cb->hhead, &cb->hhead)) {
1871 /* restore the cb->tprev entry */
1872 if (cb_OldToNew(state, cb->tprev, &cb->tprev)) {
1877 /* restore the cb->tnext entry */
1878 if (cb_OldToNew(state, cb->tnext, &cb->tnext)) {
1883 /* restore the cb->hprev entry */
1884 if (cb_OldToNew(state, cb->hprev, &cb->hprev)) {
1889 /* restore the cb->hnext entry */
1890 if (cb_OldToNew(state, cb->hnext, &cb->hnext)) {
1897 /* restore the timeout queue head indices */
1898 for (i = 0; i < state->cb_timeout_hdr->records; i++) {
1899 if (cb_OldToNew(state, timeout[i], &timeout[i])) {
1905 /* restore the FE hash table queue heads */
1906 for (i = 0; i < state->cb_fehash_hdr->records; i++) {
1907 if (fe_OldToNew(state, HashTable[i], &HashTable[i])) {
1918 cb_stateVerify(struct fs_dump_state * state)
1922 if (cb_stateVerifyFEHash(state)) {
1926 if (cb_stateVerifyTimeoutQueues(state)) {
1935 cb_stateVerifyFEHash(struct fs_dump_state * state)
1938 struct FileEntry * fe;
1939 afs_uint32 fei, chain_len;
1941 for (i = 0; i < FEHASH_SIZE; i++) {
1943 for (fei = HashTable[i], fe = itofe(fei);
1945 fei = fe->fnext, fe = itofe(fei)) {
1946 if (fei > cbstuff.nblks) {
1947 ViceLog(0, ("cb_stateVerifyFEHash: error: index out of range (fei=%d)\n", fei));
1951 if (cb_stateVerifyFE(state, fe)) {
1954 if (chain_len > FS_STATE_FE_MAX_HASH_CHAIN_LEN) {
1955 ViceLog(0, ("cb_stateVerifyFEHash: error: hash chain %d length exceeds %d; assuming there's a loop\n",
1956 i, FS_STATE_FE_MAX_HASH_CHAIN_LEN));
1969 cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe)
1973 if ((fe->firstcb && !fe->ncbs) ||
1974 (!fe->firstcb && fe->ncbs)) {
1975 ViceLog(0, ("cb_stateVerifyFE: error: fe->firstcb does not agree with fe->ncbs (fei=%d, fe->firstcb=%d, fe->ncbs=%d)\n",
1976 fetoi(fe), fe->firstcb, fe->ncbs));
1979 if (cb_stateVerifyFCBList(state, fe)) {
1980 ViceLog(0, ("cb_stateVerifyFE: error: FCBList failed verification (fei=%d)\n", fetoi(fe)));
1989 cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe)
1992 afs_uint32 cbi, fei, chain_len = 0;
1993 struct CallBack * cb;
1997 for (cbi = fe->firstcb, cb = itocb(cbi);
1999 cbi = cb->cnext, cb = itocb(cbi)) {
2000 if (cbi > cbstuff.nblks) {
2001 ViceLog(0, ("cb_stateVerifyFCBList: error: list index out of range (cbi=%d, ncbs=%d)\n",
2002 cbi, cbstuff.nblks));
2006 if (cb->fhead != fei) {
2007 ViceLog(0, ("cb_stateVerifyFCBList: error: cb->fhead != fei (fei=%d, cb->fhead=%d)\n",
2011 if (chain_len > FS_STATE_FCB_MAX_LIST_LEN) {
2012 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (fei=%d); assuming there's a loop\n",
2013 FS_STATE_FCB_MAX_LIST_LEN, fei));
2020 if (fe->ncbs != chain_len) {
2021 ViceLog(0, ("cb_stateVerifyFCBList: error: list length mismatch (len=%d, fe->ncbs=%d)\n",
2022 chain_len, fe->ncbs));
2031 cb_stateVerifyHCBList(struct fs_dump_state * state, struct host * host)
2034 afs_uint32 hi, chain_len, cbi;
2035 struct CallBack *cb, *ncb;
2040 for (cbi = host->cblist, cb = itocb(cbi);
2042 cbi = cb->hnext, cb = ncb) {
2043 if (chain_len && (host->cblist == cbi)) {
2044 /* we've wrapped around the circular list, and everything looks ok */
2047 if (cb->hhead != hi) {
2048 ViceLog(0, ("cb_stateVerifyHCBList: error: incorrect cb->hhead (cbi=%d, h->index=%d, cb->hhead=%d)\n",
2049 cbi, hi, cb->hhead));
2052 if (!cb->hprev || !cb->hnext) {
2053 ViceLog(0, ("cb_stateVerifyHCBList: error: null index in circular list (cbi=%d, h->index=%d)\n",
2058 if ((cb->hprev > cbstuff.nblks) ||
2059 (cb->hnext > cbstuff.nblks)) {
2060 ViceLog(0, ("cb_stateVerifyHCBList: error: list index out of range (cbi=%d, h->index=%d, cb->hprev=%d, cb->hnext=%d, nCBs=%d)\n",
2061 cbi, hi, cb->hprev, cb->hnext, cbstuff.nblks));
2065 ncb = itocb(cb->hnext);
2066 if (cbi != ncb->hprev) {
2067 ViceLog(0, ("cb_stateVerifyHCBList: error: corrupt linked list (cbi=%d, h->index=%d)\n",
2072 if (chain_len > FS_STATE_HCB_MAX_LIST_LEN) {
2073 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (h->index=%d); assuming there's a loop\n",
2074 FS_STATE_HCB_MAX_LIST_LEN, hi));
2086 cb_stateVerifyTimeoutQueues(struct fs_dump_state * state)
2089 afs_uint32 cbi, chain_len;
2090 struct CallBack *cb, *ncb;
2092 for (i = 0; i < CB_NUM_TIMEOUT_QUEUES; i++) {
2094 for (cbi = timeout[i], cb = itocb(cbi);
2096 cbi = cb->tnext, cb = ncb) {
2097 if (chain_len && (cbi == timeout[i])) {
2098 /* we've wrapped around the circular list, and everything looks ok */
2101 if (cbi > cbstuff.nblks) {
2102 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: list index out of range (cbi=%d, tindex=%d)\n",
2107 if (itot(cb->thead) != &timeout[i]) {
2108 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: cb->thead points to wrong timeout queue (tindex=%d, cbi=%d, cb->thead=%d)\n",
2109 i, cbi, cb->thead));
2112 if (!cb->tprev || !cb->tnext) {
2113 ViceLog(0, ("cb_stateVerifyTimeoutQueues: null index in circular list (cbi=%d, tindex=%d)\n",
2118 if ((cb->tprev > cbstuff.nblks) ||
2119 (cb->tnext > cbstuff.nblks)) {
2120 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list index out of range (cbi=%d, tindex=%d, cb->tprev=%d, cb->tnext=%d, nCBs=%d)\n",
2121 cbi, i, cb->tprev, cb->tnext, cbstuff.nblks));
2125 ncb = itocb(cb->tnext);
2126 if (cbi != ncb->tprev) {
2127 ViceLog(0, ("cb_stateVerifyTimeoutQueues: corrupt linked list (cbi=%d, tindex=%d)\n",
2132 if (chain_len > FS_STATE_TCB_MAX_LIST_LEN) {
2133 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list length exceeds %d (tindex=%d); assuming there's a loop\n",
2134 FS_STATE_TCB_MAX_LIST_LEN, i));
2147 cb_stateSaveTimeouts(struct fs_dump_state * state)
2150 struct iovec iov[2];
2152 AssignInt64(state->eof_offset, &state->cb_hdr->timeout_offset);
2154 memset(state->cb_timeout_hdr, 0, sizeof(struct callback_state_fehash_header));
2155 state->cb_timeout_hdr->magic = CALLBACK_STATE_TIMEOUT_MAGIC;
2156 state->cb_timeout_hdr->records = CB_NUM_TIMEOUT_QUEUES;
2157 state->cb_timeout_hdr->len = sizeof(struct callback_state_timeout_header) +
2158 (state->cb_timeout_hdr->records * sizeof(afs_uint32));
2160 iov[0].iov_base = (char *)state->cb_timeout_hdr;
2161 iov[0].iov_len = sizeof(struct callback_state_timeout_header);
2162 iov[1].iov_base = (char *)timeout;
2163 iov[1].iov_len = sizeof(timeout);
2165 if (fs_stateSeek(state, &state->cb_hdr->timeout_offset)) {
2170 if (fs_stateWriteV(state, iov, 2)) {
2175 fs_stateIncEOF(state, state->cb_timeout_hdr->len);
2182 cb_stateRestoreTimeouts(struct fs_dump_state * state)
2186 if (fs_stateReadHeader(state, &state->cb_hdr->timeout_offset,
2187 state->cb_timeout_hdr,
2188 sizeof(struct callback_state_timeout_header))) {
2193 if (state->cb_timeout_hdr->magic != CALLBACK_STATE_TIMEOUT_MAGIC) {
2197 if (state->cb_timeout_hdr->records != CB_NUM_TIMEOUT_QUEUES) {
2202 len = state->cb_timeout_hdr->records * sizeof(afs_uint32);
2204 if (state->cb_timeout_hdr->len !=
2205 (sizeof(struct callback_state_timeout_header) + len)) {
2210 if (fs_stateRead(state, timeout, len)) {
2220 cb_stateSaveFEHash(struct fs_dump_state * state)
2223 struct iovec iov[2];
2225 AssignInt64(state->eof_offset, &state->cb_hdr->fehash_offset);
2227 memset(state->cb_fehash_hdr, 0, sizeof(struct callback_state_fehash_header));
2228 state->cb_fehash_hdr->magic = CALLBACK_STATE_FEHASH_MAGIC;
2229 state->cb_fehash_hdr->records = FEHASH_SIZE;
2230 state->cb_fehash_hdr->len = sizeof(struct callback_state_fehash_header) +
2231 (state->cb_fehash_hdr->records * sizeof(afs_uint32));
2233 iov[0].iov_base = (char *)state->cb_fehash_hdr;
2234 iov[0].iov_len = sizeof(struct callback_state_fehash_header);
2235 iov[1].iov_base = (char *)HashTable;
2236 iov[1].iov_len = sizeof(HashTable);
2238 if (fs_stateSeek(state, &state->cb_hdr->fehash_offset)) {
2243 if (fs_stateWriteV(state, iov, 2)) {
2248 fs_stateIncEOF(state, state->cb_fehash_hdr->len);
2255 cb_stateRestoreFEHash(struct fs_dump_state * state)
2259 if (fs_stateReadHeader(state, &state->cb_hdr->fehash_offset,
2260 state->cb_fehash_hdr,
2261 sizeof(struct callback_state_fehash_header))) {
2266 if (state->cb_fehash_hdr->magic != CALLBACK_STATE_FEHASH_MAGIC) {
2270 if (state->cb_fehash_hdr->records != FEHASH_SIZE) {
2275 len = state->cb_fehash_hdr->records * sizeof(afs_uint32);
2277 if (state->cb_fehash_hdr->len !=
2278 (sizeof(struct callback_state_fehash_header) + len)) {
2283 if (fs_stateRead(state, HashTable, len)) {
2293 cb_stateSaveFEs(struct fs_dump_state * state)
2296 register int fei, hash;
2297 register struct FileEntry *fe;
2299 AssignInt64(state->eof_offset, &state->cb_hdr->fe_offset);
2301 for (hash = 0; hash < FEHASH_SIZE ; hash++) {
2302 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
2304 if (cb_stateSaveFE(state, fe)) {
2316 cb_stateRestoreFEs(struct fs_dump_state * state)
2318 int count, nFEs, ret = 0;
2320 nFEs = state->cb_hdr->nFEs;
2322 for (count = 0; count < nFEs; count++) {
2323 if (cb_stateRestoreFE(state)) {
2334 cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe)
2336 int ret = 0, iovcnt, cbi, written = 0;
2338 struct callback_state_entry_header hdr;
2339 struct FEDiskEntry fedsk;
2340 struct CBDiskEntry cbdsk[16];
2341 struct iovec iov[16];
2342 struct CallBack *cb;
2345 if (fei > state->cb_hdr->fe_max) {
2346 state->cb_hdr->fe_max = fei;
2349 memset(&hdr, 0, sizeof(struct callback_state_entry_header));
2351 if (cb_stateFEToDiskEntry(fe, &fedsk)) {
2356 iov[0].iov_base = (char *)&hdr;
2357 iov[0].iov_len = sizeof(hdr);
2358 iov[1].iov_base = (char *)&fedsk;
2359 iov[1].iov_len = sizeof(struct FEDiskEntry);
2362 for (cbi = fe->firstcb, cb = itocb(cbi);
2364 cbi = cb->cnext, cb = itocb(cbi), hdr.nCBs++) {
2365 if (cbi > state->cb_hdr->cb_max) {
2366 state->cb_hdr->cb_max = cbi;
2368 if (cb_stateCBToDiskEntry(cb, &cbdsk[iovcnt])) {
2372 cbdsk[iovcnt].index = cbi;
2373 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2374 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2376 if ((iovcnt == 16) || (!cb->cnext)) {
2377 if (fs_stateWriteV(state, iov, iovcnt)) {
2386 hdr.magic = CALLBACK_STATE_ENTRY_MAGIC;
2387 hdr.len = sizeof(hdr) + sizeof(struct FEDiskEntry) +
2388 (hdr.nCBs * sizeof(struct CBDiskEntry));
2391 if (fs_stateWriteV(state, iov, iovcnt)) {
2396 if (fs_stateWriteHeader(state, &state->eof_offset, &hdr, sizeof(hdr))) {
2402 fs_stateIncEOF(state, hdr.len);
2405 if (fs_stateSeek(state, &state->eof_offset)) {
2411 state->cb_hdr->nFEs++;
2412 state->cb_hdr->nCBs += hdr.nCBs;
2419 cb_stateRestoreFE(struct fs_dump_state * state)
2421 int ret = 0, iovcnt, nCBs;
2422 struct callback_state_entry_header hdr;
2423 struct FEDiskEntry fedsk;
2424 struct CBDiskEntry cbdsk[16];
2425 struct iovec iov[16];
2426 struct FileEntry * fe;
2427 struct CallBack * cb;
2429 iov[0].iov_base = (char *)&hdr;
2430 iov[0].iov_len = sizeof(hdr);
2431 iov[1].iov_base = (char *)&fedsk;
2432 iov[1].iov_len = sizeof(fedsk);
2435 if (fs_stateReadV(state, iov, iovcnt)) {
2440 if (hdr.magic != CALLBACK_STATE_ENTRY_MAGIC) {
2447 ViceLog(0, ("cb_stateRestoreFE: ran out of free FileEntry structures\n"));
2452 if (cb_stateDiskEntryToFE(state, &fedsk, fe)) {
2458 for (iovcnt = 0, nCBs = 0;
2461 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2462 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2464 if ((iovcnt == 16) || (nCBs == hdr.nCBs - 1)) {
2465 if (fs_stateReadV(state, iov, iovcnt)) {
2469 if (cb_stateRestoreCBs(state, fe, iov, iovcnt)) {
2483 cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
2484 struct iovec * iov, int niovecs)
2487 register struct CallBack * cb;
2488 struct CBDiskEntry * cbdsk;
2493 for (idx = 0; idx < niovecs; idx++) {
2494 cbdsk = (struct CBDiskEntry *) iov[idx].iov_base;
2495 if ((cb = GetCB()) == NULL) {
2496 ViceLog(0, ("cb_stateRestoreCBs: ran out of free CallBack structures\n"));
2500 if (cb_stateDiskEntryToCB(state, cbdsk, cb)) {
2501 ViceLog(0, ("cb_stateRestoreCBs: corrupt CallBack disk entry\n"));
2513 cb_stateFillHeader(struct callback_state_header * hdr)
2515 hdr->stamp.magic = CALLBACK_STATE_MAGIC;
2516 hdr->stamp.version = CALLBACK_STATE_VERSION;
2517 hdr->tfirst = tfirst;
2522 cb_stateCheckHeader(struct callback_state_header * hdr)
2526 if (hdr->stamp.magic != CALLBACK_STATE_MAGIC) {
2528 } else if (hdr->stamp.version != CALLBACK_STATE_VERSION) {
2530 } else if ((hdr->nFEs > cbstuff.nblks) || (hdr->nCBs > cbstuff.nblks)) {
2531 ViceLog(0, ("cb_stateCheckHeader: saved callback state larger than callback memory allocation\n"));
2537 /* disk entry conversion routines */
2539 cb_stateFEToDiskEntry(struct FileEntry * in, struct FEDiskEntry * out)
2541 memcpy(&out->fe, in, sizeof(struct FileEntry));
2542 out->index = fetoi(in);
2547 cb_stateDiskEntryToFE(struct fs_dump_state * state,
2548 struct FEDiskEntry * in, struct FileEntry * out)
2552 memcpy(out, &in->fe, sizeof(struct FileEntry));
2554 /* setup FE map entry */
2555 if (!in->index || (in->index >= state->fe_map.len)) {
2556 ViceLog(0, ("cb_stateDiskEntryToFE: index (%d) out of range",
2561 state->fe_map.entries[in->index].old_idx = in->index;
2562 state->fe_map.entries[in->index].new_idx = fetoi(out);
2569 cb_stateCBToDiskEntry(struct CallBack * in, struct CBDiskEntry * out)
2571 memcpy(&out->cb, in, sizeof(struct CallBack));
2572 out->index = cbtoi(in);
2577 cb_stateDiskEntryToCB(struct fs_dump_state * state,
2578 struct CBDiskEntry * in, struct CallBack * out)
2582 memcpy(out, &in->cb, sizeof(struct CallBack));
2584 /* setup CB map entry */
2585 if (!in->index || (in->index >= state->cb_map.len)) {
2586 ViceLog(0, ("cb_stateDiskEntryToCB: index (%d) out of range\n",
2591 state->cb_map.entries[in->index].old_idx = in->index;
2592 state->cb_map.entries[in->index].new_idx = cbtoi(out);
2598 /* index map routines */
2600 cb_stateAllocMap(struct fs_dump_state * state)
2602 state->fe_map.len = state->cb_hdr->fe_max + 1;
2603 state->cb_map.len = state->cb_hdr->cb_max + 1;
2604 state->fe_map.entries = (struct idx_map_entry_t *)
2605 calloc(state->fe_map.len, sizeof(struct idx_map_entry_t));
2606 state->cb_map.entries = (struct idx_map_entry_t *)
2607 calloc(state->cb_map.len, sizeof(struct idx_map_entry_t));
2608 return ((state->fe_map.entries != NULL) && (state->cb_map.entries != NULL)) ? 0 : 1;
2612 fe_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2616 /* FEs use a one-based indexing system, so old==0 implies no mapping */
2622 if (old >= state->fe_map.len) {
2623 ViceLog(0, ("fe_OldToNew: index %d is out of range\n", old));
2625 } else if (state->fe_map.entries[old].old_idx != old) { /* sanity check */
2626 ViceLog(0, ("fe_OldToNew: index %d points to an invalid FileEntry record\n", old));
2629 *new = state->fe_map.entries[old].new_idx;
2637 cb_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2641 /* CBs use a one-based indexing system, so old==0 implies no mapping */
2647 if (old >= state->cb_map.len) {
2648 ViceLog(0, ("cb_OldToNew: index %d is out of range\n", old));
2650 } else if (state->cb_map.entries[old].old_idx != old) { /* sanity check */
2651 ViceLog(0, ("cb_OldToNew: index %d points to an invalid CallBack record\n", old));
2654 *new = state->cb_map.entries[old].new_idx;
2660 #endif /* AFS_DEMAND_ATTACH_FS */
2663 DumpCallBackState(void)
2666 afs_uint32 magic = MAGIC, freelisthead;
2667 time_t now = FT_ApproxTime();
2669 oflag = O_WRONLY | O_CREAT | O_TRUNC;
2673 fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, oflag, 0666);
2676 ("Couldn't create callback dump file %s\n",
2677 AFSDIR_SERVER_CBKDUMP_FILEPATH));
2680 (void)write(fd, &magic, sizeof(magic));
2681 (void)write(fd, &now, sizeof(now));
2682 (void)write(fd, &cbstuff, sizeof(cbstuff));
2683 (void)write(fd, TimeOuts, sizeof(TimeOuts));
2684 (void)write(fd, timeout, sizeof(timeout));
2685 (void)write(fd, &tfirst, sizeof(tfirst));
2686 freelisthead = cbtoi((struct CallBack *)CBfree);
2687 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2688 freelisthead = fetoi((struct FileEntry *)FEfree);
2689 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2690 (void)write(fd, HashTable, sizeof(HashTable));
2691 (void)write(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2692 (void)write(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2698 #endif /* !INTERPRET_DUMP */
2700 #ifdef INTERPRET_DUMP
2702 /* This is only compiled in for the callback analyzer program */
2703 /* Returns the time of the dump */
2705 ReadDump(char *file)
2708 afs_uint32 magic, freelisthead;
2715 fd = open(file, oflag);
2717 fprintf(stderr, "Couldn't read dump file %s\n", file);
2720 read(fd, &magic, sizeof(magic));
2721 if (magic != MAGIC) {
2723 "Magic number of %s is invalid. You might be trying to\n",
2726 "run this program on a machine type with a different byte ordering.\n");
2729 read(fd, &now, sizeof(now));
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"
2759 main(int argc, char **argv)
2761 int err = 0, cbi = 0, stats = 0, noptions = 0, all = 0, vol = 0, raw = 0;
2763 register struct FileEntry *fe;
2764 register struct CallBack *cb;
2767 memset(&fid, 0, sizeof(fid));
2770 while (argc && **argv == '-') {
2773 if (!strcmp(*argv, "-host")) {
2779 cbi = atoi(*++argv);
2780 } else if (!strcmp(*argv, "-fid")) {
2786 fid.Volume = atoi(*++argv);
2787 fid.Vnode = atoi(*++argv);
2788 fid.Unique = atoi(*++argv);
2789 } else if (!strcmp(*argv, "-time")) {
2790 fprintf(stderr, "-time not supported\n");
2792 } else if (!strcmp(*argv, "-stats")) {
2794 } else if (!strcmp(*argv, "-all")) {
2796 } else if (!strcmp(*argv, "-raw")) {
2798 } else if (!strcmp(*argv, "-volume")) {
2804 vol = atoi(*++argv);
2809 if (err || argc != 1) {
2811 "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] callbackdumpfile\n");
2813 "[cbid is shown for each host in the hosts.dump file]\n");
2816 now = ReadDump(*argv);
2817 if (stats || noptions == 0) {
2818 time_t uxtfirst = UXtime(tfirst);
2819 printf("The time of the dump was %u %s", (unsigned int) now, ctime(&now));
2820 printf("The last time cleanup ran was %u %s", (unsigned int) uxtfirst,
2822 PrintCallBackStats();
2827 struct CallBack *cb;
2828 struct FileEntry *fe;
2830 for (hash = 0; hash < FEHASH_SIZE; hash++) {
2831 for (feip = &HashTable[hash]; fe = itofe(*feip);) {
2832 if (!vol || (fe->volid == vol)) {
2833 register struct CallBack *cbnext;
2834 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
2836 cbnext = itocb(cb->cnext);
2846 afs_uint32 cfirst = cbi;
2851 } while (cbi != cfirst);
2856 printf("No callback entries for %u.%u\n", fid.Volume, fid.Vnode);
2859 cb = itocb(fe->firstcb);
2862 cb = itocb(cb->cnext);
2867 for (i = 1; i < cbstuff.nblks; i++) {
2868 p = (afs_int32 *) & FE[i];
2869 printf("%d:%12x%12x%12x%12x\n", i, p[0], p[1], p[2], p[3]);
2876 PrintCB(register struct CallBack *cb, afs_uint32 now)
2878 struct FileEntry *fe = itofe(cb->fhead);
2879 time_t expires = TIndexToTime(cb->thead);
2884 printf("vol=%u vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %lu secs at %s",
2885 fe->volid, fe->vnode, fe->ncbs, cb->hhead, cb->status, fe->status,
2886 expires - now, ctime(&expires));
2891 #if !defined(INTERPRET_DUMP)
2893 ** try breaking calbacks on afidp from host. Use multi_rx.
2894 ** return 0 on success, non-zero on failure
2897 MultiBreakCallBackAlternateAddress(struct host *host, struct AFSCBFids *afidp)
2901 retVal = MultiBreakCallBackAlternateAddress_r(host, afidp);
2907 MultiBreakCallBackAlternateAddress_r(struct host *host,
2908 struct AFSCBFids *afidp)
2911 struct rx_connection **conns;
2912 struct rx_connection *connSuccess = 0;
2913 struct AddrPort *interfaces;
2914 static struct rx_securityClass *sc = 0;
2915 static struct AFSCBs tc = { 0, 0 };
2918 /* nothing more can be done */
2919 if (!host->interface)
2920 return 1; /* failure */
2922 /* the only address is the primary interface */
2923 if (host->interface->numberOfInterfaces <= 1)
2924 return 1; /* failure */
2926 /* initialise a security object only once */
2928 sc = rxnull_NewClientSecurityObject();
2930 i = host->interface->numberOfInterfaces;
2931 interfaces = calloc(i, sizeof(struct AddrPort));
2932 conns = calloc(i, sizeof(struct rx_connection *));
2933 if (!interfaces || !conns) {
2935 ("Failed malloc in MultiBreakCallBackAlternateAddress_r\n"));
2939 /* initialize alternate rx connections */
2940 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
2941 /* this is the current primary address */
2942 if (host->host == host->interface->interface[i].addr &&
2943 host->port == host->interface->interface[i].port)
2946 interfaces[j] = host->interface->interface[i];
2948 rx_NewConnection(interfaces[j].addr,
2949 interfaces[j].port, 1, sc, 0);
2950 rx_SetConnDeadTime(conns[j], 2);
2951 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
2955 assert(j); /* at least one alternate address */
2957 ("Starting multibreakcall back on all addr for host %x (%s:%d)\n",
2958 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
2960 multi_Rx(conns, j) {
2961 multi_RXAFSCB_CallBack(afidp, &tc);
2965 if (host->callback_rxcon)
2966 rx_DestroyConnection(host->callback_rxcon);
2967 host->callback_rxcon = conns[multi_i];
2968 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
2969 host->host = interfaces[multi_i].addr;
2970 host->port = interfaces[multi_i].port;
2971 h_AddHostToAddrHashTable_r(host->host, host->port, host);
2972 connSuccess = conns[multi_i];
2973 rx_SetConnDeadTime(host->callback_rxcon, 50);
2974 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
2976 ("multibreakcall success with addr %s:%d\n",
2977 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
2978 ntohs(interfaces[multi_i].port)));
2985 /* Destroy all connections except the one on which we succeeded */
2986 for (i = 0; i < j; i++)
2987 if (conns[i] != connSuccess)
2988 rx_DestroyConnection(conns[i]);
2994 return 0; /* success */
2996 return 1; /* failure */
3001 ** try multi_RX probes to host.
3002 ** return 0 on success, non-0 on failure
3005 MultiProbeAlternateAddress_r(struct host *host)
3008 struct rx_connection **conns;
3009 struct rx_connection *connSuccess = 0;
3010 struct AddrPort *interfaces;
3011 static struct rx_securityClass *sc = 0;
3014 /* nothing more can be done */
3015 if (!host->interface)
3016 return 1; /* failure */
3018 /* the only address is the primary interface */
3019 if (host->interface->numberOfInterfaces <= 1)
3020 return 1; /* failure */
3022 /* initialise a security object only once */
3024 sc = rxnull_NewClientSecurityObject();
3026 i = host->interface->numberOfInterfaces;
3027 interfaces = calloc(i, sizeof(struct AddrPort));
3028 conns = calloc(i, sizeof(struct rx_connection *));
3029 if (!interfaces || !conns) {
3030 ViceLog(0, ("Failed malloc in MultiProbeAlternateAddress_r\n"));
3034 /* initialize alternate rx connections */
3035 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3036 /* this is the current primary address */
3037 if (host->host == host->interface->interface[i].addr &&
3038 host->port == host->interface->interface[i].port)
3041 interfaces[j] = host->interface->interface[i];
3043 rx_NewConnection(interfaces[j].addr,
3044 interfaces[j].port, 1, sc, 0);
3045 rx_SetConnDeadTime(conns[j], 2);
3046 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3050 assert(j); /* at least one alternate address */
3052 ("Starting multiprobe on all addr for host %x (%s:%d)\n",
3053 host, afs_inet_ntoa_r(host->host, hoststr),
3054 ntohs(host->port)));
3056 multi_Rx(conns, j) {
3057 multi_RXAFSCB_ProbeUuid(&host->interface->uuid);
3061 if (host->callback_rxcon)
3062 rx_DestroyConnection(host->callback_rxcon);
3063 host->callback_rxcon = conns[multi_i];
3064 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
3065 host->host = interfaces[multi_i].addr;
3066 host->port = interfaces[multi_i].port;
3067 h_AddHostToAddrHashTable_r(host->host, host->port, host);
3068 connSuccess = conns[multi_i];
3069 rx_SetConnDeadTime(host->callback_rxcon, 50);
3070 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3072 ("multiprobe success with addr %s:%d\n",
3073 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3074 ntohs(interfaces[multi_i].port)));
3079 ("multiprobe failure with addr %s:%d\n",
3080 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3081 ntohs(interfaces[multi_i].port)));
3083 /* This is less than desirable but its the best we can do.
3084 * The AFS Cache Manager will return either 0 for a Uuid
3085 * match and a 1 for a non-match. If the error is 1 we
3086 * therefore know that our mapping of IP address to Uuid
3087 * is wrong. We should attempt to find the correct
3088 * Uuid and fix the host tables.
3090 if (multi_error == 1) {
3091 /* remove the current alternate address from this host */
3093 removeInterfaceAddr_r(host, interfaces[multi_i].addr, interfaces[multi_i].port);
3097 #ifdef AFS_DEMAND_ATTACH_FS
3098 /* try to bail ASAP if the fileserver is shutting down */
3100 if (fs_state.mode == FS_MODE_SHUTDOWN) {
3109 /* Destroy all connections except the one on which we succeeded */
3110 for (i = 0; i < j; i++)
3111 if (conns[i] != connSuccess)
3112 rx_DestroyConnection(conns[i]);
3118 return 0; /* success */
3120 return 1; /* failure */
3123 #endif /* !defined(INTERPRET_DUMP) */