2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
9 * Portions Copyright (c) 2006 Sine Nomine Associates
13 * NEW callback package callback.c (replaces vicecb.c)
14 * Updated call back routines, NOW with:
16 * Faster DeleteVenus (Now called DeleteAllCallBacks)
17 * Call back breaking for volumes
18 * Adaptive timeouts on call backs
19 * Architected for Multi RPC
20 * No locks (currently implicit vnode locks--these will go, to)
21 * Delayed call back when rpc connection down.
22 * Bulk break of delayed call backs when rpc connection
24 * Strict limit on number of call backs.
26 * InitCallBack(nblocks)
27 * Initialize: nblocks is max number # of file entries + # of callback entries
28 * nblocks must be < 65536
29 * Space used is nblocks*16 bytes
30 * Note that space will be reclaimed by breaking callbacks of old hosts
32 * time = AddCallBack(host, fid)
34 * Returns the expiration time at the workstation.
36 * BreakCallBack(host, fid)
37 * Break all call backs for fid, except for the specified host.
40 * BreakVolumeCallBacks(volume)
41 * Break all call backs on volume, using single call to each host
42 * Delete all the call backs.
44 * DeleteCallBack(host,fid)
45 * Delete (do not break) single call back for fid.
47 * DeleteFileCallBacks(fid)
48 * Delete (do not break) all call backs for fid.
50 * DeleteAllCallBacks(host)
51 * Delete (do not break) all call backs for host.
53 * CleanupTimedOutCallBacks()
54 * Delete all timed out call back entries
55 * Must be called periodically by file server.
57 * BreakDelayedCallBacks(host)
58 * Break all delayed call backs for host.
59 * Returns 1: one or more failed, 0: success.
61 * PrintCallBackStats()
62 * Print statistics about call backs to stdout.
64 * DumpCallBacks() ---wishful thinking---
65 * Dump call back state to /tmp/callback.state.
66 * This is separately interpretable by the program pcb.
68 * Notes: In general, if a call back to a host doesn't get through,
69 * then HostDown, supplied elsewhere, is called. BreakDelayedCallBacks,
70 * however, does not call HostDown, but instead returns an indication of
71 * success if all delayed call backs were finally broken.
73 * BreakDelayedCallBacks MUST be called at the first sign of activity
74 * from the host after HostDown has been called (or a previous
75 * BreakDelayedCallBacks failed). The BreakDelayedCallBacks must be
76 * allowed to complete before any requests from that host are handled.
77 * If BreakDelayedCallBacks fails, then the host should remain
78 * down (and the request should be failed).
80 * CleanupCallBacks MUST be called periodically by the file server for
81 * this package to work correctly. Every 5 minutes is suggested.
84 #include <afsconfig.h>
85 #include <afs/param.h>
91 #include <stdlib.h> /* for malloc() */
92 #include <time.h> /* ANSI standard location for time stuff */
101 #include <afs/assert.h>
103 #include <afs/stds.h>
105 #include <afs/nfs.h> /* yuck. This is an abomination. */
108 #include <afscbint.h>
109 #include <afs/afsutil.h>
111 #include <afs/ihandle.h>
112 #include <afs/vnode.h>
113 #include <afs/volume.h>
114 #include "viced_prototypes.h"
117 #include <afs/ptclient.h> /* need definition of prlist for host.h */
119 #include "callback.h"
120 #ifdef AFS_DEMAND_ATTACH_FS
121 #include "../tviced/serialize_state.h"
122 #endif /* AFS_DEMAND_ATTACH_FS */
125 extern afsUUID FS_HostUUID;
126 extern int hostCount;
127 static int ShowProblems = 1;
129 struct cbcounters cbstuff;
131 static struct FileEntry * FE = NULL; /* don't use FE[0] */
132 static struct CallBack * CB = NULL; /* don't use CB[0] */
134 static struct CallBack * CBfree = NULL;
135 static struct FileEntry * FEfree = NULL;
138 /* Time to live for call backs depends upon number of users of the file.
139 * TimeOuts is indexed by this number/8 (using TimeOut macro). Times
140 * in this table are for the workstation; server timeouts, add
143 static int TimeOuts[] = {
144 /* Note: don't make the first entry larger than 4 hours (see above) */
145 4 * 60 * 60, /* 0-7 users */
146 1 * 60 * 60, /* 8-15 users */
147 30 * 60, /* 16-23 users */
148 15 * 60, /* 24-31 users */
149 15 * 60, /* 32-39 users */
150 10 * 60, /* 40-47 users */
151 10 * 60, /* 48-55 users */
152 10 * 60, /* 56-63 users */
153 }; /* Anything more: MinTimeOut */
155 /* minimum time given for a call back */
156 static int MinTimeOut = (7 * 60);
158 /* Heads of CB queues; a timeout index is 1+index into this array */
159 static afs_uint32 timeout[CB_NUM_TIMEOUT_QUEUES];
161 static afs_int32 tfirst; /* cbtime of oldest unexpired call back time queue */
164 /* 16 byte object get/free routines */
169 /* Prototypes for static routines */
170 static struct FileEntry *FindFE(register AFSFid * fid);
171 static struct CallBack *iGetCB(register int *nused);
172 static int iFreeCB(register struct CallBack *cb, register int *nused);
173 static struct FileEntry *iGetFE(register int *nused);
174 static int iFreeFE(register struct FileEntry *fe, register int *nused);
175 static int TAdd(register struct CallBack *cb, register afs_uint32 * thead);
176 static int TDel(register struct CallBack *cb);
177 static int HAdd(register struct CallBack *cb, register struct host *host);
178 static int HDel(register struct CallBack *cb);
179 static int CDel(struct CallBack *cb, int deletefe);
180 static int CDelPtr(register struct FileEntry *fe, register afs_uint32 * cbp,
182 static afs_uint32 *FindCBPtr(struct FileEntry *fe, struct host *host);
183 static int FDel(register struct FileEntry *fe);
184 static int AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead,
185 int type, int locked);
186 static void MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
187 struct AFSCBFids *afidp, struct host *xhost);
188 static int MultiBreakVolumeCallBack_r(struct host *host, int isheld,
189 struct VCBParams *parms, int deletefe);
190 static int MultiBreakVolumeCallBack(struct host *host, int isheld,
191 struct VCBParams *parms);
192 static int MultiBreakVolumeLaterCallBack(struct host *host, int isheld,
193 struct VCBParams *parms);
194 static int GetSomeSpace_r(struct host *hostp, int locked);
195 static int ClearHostCallbacks_r(struct host *hp, int locked);
197 #define GetCB() ((struct CallBack *)iGetCB(&cbstuff.nCBs))
198 #define GetFE() ((struct FileEntry *)iGetFE(&cbstuff.nFEs))
199 #define FreeCB(cb) iFreeCB((struct CallBack *)cb, &cbstuff.nCBs)
200 #define FreeFE(fe) iFreeFE((struct FileEntry *)fe, &cbstuff.nFEs)
203 /* Other protos - move out sometime */
204 void PrintCB(register struct CallBack *cb, afs_uint32 now);
206 static afs_uint32 HashTable[FEHASH_SIZE]; /* File entry hash table */
208 static struct FileEntry *
209 FindFE(register AFSFid * fid)
213 register struct FileEntry *fe;
215 hash = FEHash(fid->Volume, fid->Unique);
216 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
218 if (fe->volid == fid->Volume && fe->unique == fid->Unique
219 && fe->vnode == fid->Vnode && (fe->status & FE_LATER) != FE_LATER)
225 #ifndef INTERPRET_DUMP
227 static struct CallBack *
228 iGetCB(register int *nused)
230 register struct CallBack *ret;
232 if ((ret = CBfree)) {
233 CBfree = (struct CallBack *)(((struct object *)ret)->next);
240 iFreeCB(register struct CallBack *cb, register int *nused)
242 ((struct object *)cb)->next = (struct object *)CBfree;
248 static struct FileEntry *
249 iGetFE(register int *nused)
251 register struct FileEntry *ret;
253 if ((ret = FEfree)) {
254 FEfree = (struct FileEntry *)(((struct object *)ret)->next);
261 iFreeFE(register struct FileEntry *fe, register int *nused)
263 ((struct object *)fe)->next = (struct object *)FEfree;
269 /* Add cb to end of specified timeout list */
271 TAdd(register struct CallBack *cb, register afs_uint32 * thead)
274 (*thead) = cb->tnext = cb->tprev = cbtoi(cb);
276 register struct CallBack *thp = itocb(*thead);
278 cb->tprev = thp->tprev;
282 thp->tprev = (itocb(thp->tprev)->tnext = cbtoi(cb));
284 thp->tprev = cbtoi(cb);
287 cb->thead = ttoi(thead);
291 /* Delete call back entry from timeout list */
293 TDel(register struct CallBack *cb)
295 register afs_uint32 *thead = itot(cb->thead);
297 if (*thead == cbtoi(cb))
298 *thead = (*thead == cb->tnext ? 0 : cb->tnext);
299 if (itocb(cb->tprev))
300 itocb(cb->tprev)->tnext = cb->tnext;
301 if (itocb(cb->tnext))
302 itocb(cb->tnext)->tprev = cb->tprev;
306 /* Add cb to end of specified host list */
308 HAdd(register struct CallBack *cb, register struct host *host)
310 cb->hhead = h_htoi(host);
312 host->cblist = cb->hnext = cb->hprev = cbtoi(cb);
314 register struct CallBack *fcb = itocb(host->cblist);
316 cb->hprev = fcb->hprev;
317 cb->hnext = cbtoi(fcb);
318 fcb->hprev = (itocb(fcb->hprev)->hnext = cbtoi(cb));
323 /* Delete call back entry from host list */
325 HDel(register struct CallBack *cb)
327 register afs_uint32 *hhead = &h_itoh(cb->hhead)->cblist;
329 if (*hhead == cbtoi(cb))
330 *hhead = (*hhead == cb->hnext ? 0 : cb->hnext);
331 itocb(cb->hprev)->hnext = cb->hnext;
332 itocb(cb->hnext)->hprev = cb->hprev;
336 /* Delete call back entry from fid's chain of cb's */
337 /* N.B. This one also deletes the CB, and also possibly parent FE, so
338 * make sure that it is not on any other list before calling this
341 CDel(struct CallBack *cb, int deletefe)
344 struct FileEntry *fe = itofe(cb->fhead);
345 register afs_uint32 *cbp;
348 for (safety = 0, cbp = &fe->firstcb; *cbp && *cbp != cbi;
349 cbp = &itocb(*cbp)->cnext, safety++) {
350 if (safety > cbstuff.nblks + 10) {
353 ("CDel: Internal Error -- shutting down: wanted %d from %d, now at %d\n",
354 cbi, fe->firstcb, *cbp));
356 ShutDownAndCore(PANIC);
359 CDelPtr(fe, cbp, deletefe);
363 /* Same as CDel, but pointer to parent pointer to CB entry is passed,
364 * as well as file entry */
365 /* N.B. This one also deletes the CB, and also possibly parent FE, so
366 * make sure that it is not on any other list before calling this
368 static int Ccdelpt = 0, CcdelB = 0;
371 CDelPtr(register struct FileEntry *fe, register afs_uint32 * cbp,
374 register struct CallBack *cb;
384 if ((--fe->ncbs == 0) && deletefe)
390 FindCBPtr(struct FileEntry *fe, struct host *host)
392 register afs_uint32 hostindex = h_htoi(host);
393 register struct CallBack *cb;
394 register afs_uint32 *cbp;
397 for (safety = 0, cbp = &fe->firstcb; *cbp; cbp = &cb->cnext, safety++) {
398 if (safety > cbstuff.nblks) {
399 ViceLog(0, ("FindCBPtr: Internal Error -- shutting down.\n"));
401 ShutDownAndCore(PANIC);
404 if (cb->hhead == hostindex)
410 /* Delete file entry from hash table */
412 FDel(register struct FileEntry *fe)
414 register int fei = fetoi(fe);
415 register afs_uint32 *p = &HashTable[FEHash(fe->volid, fe->unique)];
417 while (*p && *p != fei)
418 p = &itofe(*p)->fnext;
425 /* initialize the callback package */
427 InitCallBack(int nblks)
430 tfirst = CBtime(FT_ApproxTime());
431 /* N.B. The "-1", below, is because
432 * FE[0] and CB[0] are not used--and not allocated */
433 FE = ((struct FileEntry *)(calloc(nblks, sizeof(struct FileEntry))));
435 ViceLog(0, ("Failed malloc in InitCallBack\n"));
438 FE--; /* FE[0] is supposed to point to junk */
439 cbstuff.nFEs = nblks;
441 FreeFE(&FE[cbstuff.nFEs]); /* This is correct */
442 CB = ((struct CallBack *)(calloc(nblks, sizeof(struct CallBack))));
444 ViceLog(0, ("Failed malloc in InitCallBack\n"));
447 CB--; /* CB[0] is supposed to point to junk */
448 cbstuff.nCBs = nblks;
450 FreeCB(&CB[cbstuff.nCBs]); /* This is correct */
451 cbstuff.nblks = nblks;
452 cbstuff.nbreakers = 0;
458 XCallBackBulk_r(struct host * ahost, struct AFSFid * fids, afs_int32 nfids)
460 struct AFSCallBack tcbs[AFSCBMAX];
466 struct rx_connection *cb_conn = NULL;
469 rx_SetConnDeadTime(ahost->callback_rxcon, 4);
470 rx_SetConnHardDeadTime(ahost->callback_rxcon, AFS_HARDDEADTIME);
477 for (i = 0; i < nfids && i < AFSCBMAX; i++) {
478 tcbs[i].CallBackVersion = CALLBACK_VERSION;
479 tcbs[i].ExpirationTime = 0;
480 tcbs[i].CallBackType = CB_DROPPED;
482 tf.AFSCBFids_len = i;
483 tf.AFSCBFids_val = &(fids[j]);
487 tc.AFSCBs_val = tcbs;
489 cb_conn = ahost->callback_rxcon;
490 rx_GetConnection(cb_conn);
492 code |= RXAFSCB_CallBack(cb_conn, &tf, &tc);
493 rx_PutConnection(cb_conn);
501 /* the locked flag tells us if the host entry has already been locked
502 * by our parent. I don't think anybody actually calls us with the
503 * host locked, but here's how to make that work: GetSomeSpace has to
504 * change so that it doesn't attempt to lock any hosts < "host". That
505 * means that it might be unable to free any objects, so it has to
506 * return an exit status. If it fails, then AddCallBack1 might fail,
507 * as well. If so, the host->ResetDone should probably be set to 0,
508 * and we probably don't want to return a callback promise to the
509 * cache manager, either. */
511 AddCallBack1(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
519 retVal = AddCallBack1_r(host, fid, thead, type, 1);
529 AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
532 struct FileEntry *fe;
533 struct CallBack *cb = 0, *lastcb = 0;
534 struct FileEntry *newfe = 0;
536 afs_uint32 *Thead = thead;
537 struct CallBack *newcb = 0;
542 /* allocate these guys first, since we can't call the allocator with
543 * the host structure locked -- or we might deadlock. However, we have
544 * to avoid races with FindFE... */
545 while (!(newcb = GetCB())) {
546 GetSomeSpace_r(host, locked);
548 while (!(newfe = GetFE())) { /* Get it now, so we don't have to call */
549 /* GetSomeSpace with the host locked, later. This might turn out to */
550 /* have been unneccessary, but that's actually kind of unlikely, since */
551 /* most files are not shared. */
552 GetSomeSpace_r(host, locked);
556 h_Lock_r(host); /* this can yield, so do it before we get any */
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));
582 register afs_uint32 hash;
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"));
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 */
647 /* Take an array full of hosts, all held. Break callbacks to them, and
648 * release the holds once you're done, except don't release xhost. xhost
649 * may be NULL. Currently only works for a single Fid in afidp array.
650 * If you want to make this work with multiple fids, you need to fix
651 * the error handling. One approach would be to force a reset if a
652 * multi-fid call fails, or you could add delayed callbacks for each
653 * fid. You probably also need to sort and remove duplicate hosts.
654 * When this is called from the BreakVolumeCallBacks path, it does NOT
655 * force a reset if the RPC fails, it just marks the host down and tries
656 * to create a delayed callback. */
657 /* N.B. be sure that code works when ncbas == 0 */
658 /* N.B. requires all the cba[*].hp pointers to be valid... */
659 /* This routine does not hold a lock on the host for the duration of
660 * the BreakCallBack RPC, which is a significant deviation from tradition.
661 * It _does_ get a lock on the host before setting VenusDown = 1,
662 * which is sufficient only if VenusDown = 0 only happens when the
663 * lock is held over the RPC and the subsequent VenusDown == 0
664 * wherever that is done. */
666 MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
667 struct AFSCBFids *afidp, struct host *xhost)
670 struct rx_connection *conns[MAX_CB_HOSTS];
671 static struct AFSCBs tc = { 0, 0 };
672 int multi_to_cba_map[MAX_CB_HOSTS];
674 assert(ncbas <= MAX_CB_HOSTS);
676 /* set up conns for multi-call */
677 for (i = 0, j = 0; i < ncbas; i++) {
678 struct host *thishost = cba[i].hp;
679 if (!thishost || (thishost->hostFlags & HOSTDELETED)) {
682 rx_GetConnection(thishost->callback_rxcon);
683 multi_to_cba_map[j] = i;
684 conns[j++] = thishost->callback_rxcon;
687 rx_SetConnDeadTime(thishost->callback_rxcon, 4);
688 rx_SetConnHardDeadTime(thishost->callback_rxcon, AFS_HARDDEADTIME);
692 if (j) { /* who knows what multi would do with 0 conns? */
696 multi_RXAFSCB_CallBack(afidp, &tc);
702 i = multi_to_cba_map[multi_i];
708 ("BCB: INTERNAL ERROR: hp=%x, cba=%x, thead=%u\n",
712 ** try breaking callbacks on alternate interface addresses
714 if (MultiBreakCallBackAlternateAddress(hp, afidp)) {
717 ("BCB: Failed on file %u.%u.%u, Host %x (%s:%d) is down\n",
718 afidp->AFSCBFids_val->Volume,
719 afidp->AFSCBFids_val->Vnode,
720 afidp->AFSCBFids_val->Unique,
722 afs_inet_ntoa_r(hp->host, hoststr),
728 hp->hostFlags |= VENUSDOWN;
730 * We always go into AddCallBack1_r with the host locked
732 AddCallBack1_r(hp, afidp->AFSCBFids_val, itot(idx),
745 for (i = 0; i < ncbas; i++) {
748 if (hp && xhost != hp) {
753 /* H_UNLOCK around this so h_FreeConnection does not deadlock.
754 h_FreeConnection should *never* be called on a callback connection,
755 but on 10/27/04 a deadlock occurred where it was, when we know why,
756 this should be reverted. -- shadow */
758 for (i = 0; i < j; i++) {
759 rx_PutConnection(conns[i]);
767 * Break all call backs for fid, except for the specified host (unless flag
768 * is true, in which case all get a callback message. Assumption: the specified
769 * host is h_Held, by the caller; the others aren't.
770 * Specified host may be bogus, that's ok. This used to check to see if the
771 * host was down in two places, once right after the host was h_held, and
772 * again after it was locked. That race condition is incredibly rare and
773 * relatively harmless even when it does occur, so we don't check for it now.
775 /* if flag is true, send a break callback msg to "host", too */
777 BreakCallBack(struct host *xhost, AFSFid * fid, int flag)
779 struct FileEntry *fe;
780 struct CallBack *cb, *nextcb;
781 struct cbstruct cba[MAX_CB_HOSTS];
788 ("BCB: BreakCallBack(Host %x all but %s:%d, (%u,%u,%u))\n",
789 xhost, afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
790 fid->Volume, fid->Vnode, fid->Unique));
793 cbstuff.BreakCallBacks++;
798 hostindex = h_htoi(xhost);
799 cb = itocb(fe->firstcb);
800 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
801 /* the most common case is what follows the || */
804 tf.AFSCBFids_len = 1;
805 tf.AFSCBFids_val = fid;
808 for (ncbas = 0; cb && ncbas < MAX_CB_HOSTS; cb = nextcb) {
809 nextcb = itocb(cb->cnext);
810 if ((cb->hhead != hostindex || flag)
811 && (cb->status == CB_BULK || cb->status == CB_NORMAL
812 || cb->status == CB_VOLUME)) {
813 struct host *thishost = h_itoh(cb->hhead);
815 ViceLog(0, ("BCB: BOGUS! cb->hhead is NULL!\n"));
816 } else if (thishost->hostFlags & VENUSDOWN) {
818 ("BCB: %x (%s:%d) is down; delaying break call back\n",
819 thishost, afs_inet_ntoa_r(thishost->host, hoststr),
820 ntohs(thishost->port)));
821 cb->status = CB_DELAYED;
824 cba[ncbas].hp = thishost;
825 cba[ncbas].thead = cb->thead;
829 CDel(cb, 1); /* Usually first; so this delete
830 * is reasonably inexpensive */
836 MultiBreakCallBack_r(cba, ncbas, &tf, xhost);
838 /* we need to to all these initializations again because MultiBreakCallBack may block */
843 cb = itocb(fe->firstcb);
844 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
845 /* the most common case is what follows the || */
856 /* Delete (do not break) single call back for fid */
858 DeleteCallBack(struct host *host, AFSFid * fid)
860 register struct FileEntry *fe;
861 register afs_uint32 *pcb;
864 cbstuff.DeleteCallBacks++;
873 ("DCB: No call backs for fid (%u, %u, %u)\n", fid->Volume,
874 fid->Vnode, fid->Unique));
877 pcb = FindCBPtr(fe, host);
880 ("DCB: No call back for host %x (%s:%d), (%u, %u, %u)\n",
881 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
882 fid->Volume, fid->Vnode, fid->Unique));
896 * Delete (do not break) all call backs for fid. This call doesn't
897 * set all of the various host locks, but it shouldn't really matter
898 * since we're not adding callbacks, but deleting them. I'm not sure
899 * why it doesn't set the lock, however; perhaps it should.
902 DeleteFileCallBacks(AFSFid * fid)
904 register struct FileEntry *fe;
905 register struct CallBack *cb;
906 register afs_uint32 cbi;
910 cbstuff.DeleteFiles++;
915 ("DF: No fid (%u,%u,%u) to delete\n", fid->Volume, fid->Vnode,
919 for (n = 0, cbi = fe->firstcb; cbi; n++) {
932 /* Delete (do not break) all call backs for host. The host should be
935 DeleteAllCallBacks_r(struct host *host, int deletefe)
937 register struct CallBack *cb;
938 register int cbi, first;
940 cbstuff.DeleteAllCallBacks++;
941 cbi = first = host->cblist;
943 ViceLog(8, ("DV: no call backs\n"));
951 } while (cbi != first);
957 * Break all delayed call backs for host. Returns 1 if all call backs
958 * successfully broken; 0 otherwise. Assumes host is h_Held and h_Locked.
959 * Must be called with VenusDown set for this host
962 BreakDelayedCallBacks(struct host *host)
966 retVal = BreakDelayedCallBacks_r(host);
972 BreakDelayedCallBacks_r(struct host *host)
974 struct AFSFid fids[AFSCBMAX];
975 u_byte thead[AFSCBMAX]; /* This should match thead in struct Callback */
976 int cbi, first, nfids;
980 struct rx_connection *cb_conn;
983 if (!(host->hostFlags & RESETDONE) && !(host->hostFlags & HOSTDELETED)) {
984 host->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
985 cb_conn = host->callback_rxcon;
986 rx_GetConnection(cb_conn);
987 if (host->interface) {
990 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
993 code = RXAFSCB_InitCallBackState(cb_conn);
995 rx_PutConnection(cb_conn);
998 host->hostFlags |= ALTADDR; /* alternate addresses are valid */
1002 ("CB: Call back connect back failed (in break delayed) for Host %x (%s:%d)\n",
1003 host, afs_inet_ntoa_r(host->host, hoststr),
1004 ntohs(host->port)));
1006 host->hostFlags |= VENUSDOWN;
1009 ("InitCallBackState success on %x (%s:%d)\n",
1010 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
1011 /* reset was done successfully */
1012 host->hostFlags |= RESETDONE;
1013 host->hostFlags &= ~VENUSDOWN;
1016 while (!(host->hostFlags & HOSTDELETED)) {
1018 host->hostFlags &= ~VENUSDOWN; /* presume up */
1019 cbi = first = host->cblist;
1023 first = host->cblist;
1026 if (cb->status == CB_DELAYED) {
1027 register struct FileEntry *fe = itofe(cb->fhead);
1028 thead[nfids] = cb->thead;
1029 fids[nfids].Volume = fe->volid;
1030 fids[nfids].Vnode = fe->vnode;
1031 fids[nfids].Unique = fe->unique;
1037 } while (cbi && cbi != first && nfids < AFSCBMAX);
1043 if (XCallBackBulk_r(host, fids, nfids)) {
1044 /* Failed, again: put them back, probably with old
1049 ("CB: XCallBackBulk failed, Host %x (%s:%d); callback list follows:\n",
1050 host, afs_inet_ntoa_r(host->host, hoststr),
1051 ntohs(host->port)));
1053 for (i = 0; i < nfids; i++) {
1056 ("CB: Host %x (%s:%d), file %u.%u.%u (part of bulk callback)\n",
1057 host, afs_inet_ntoa_r(host->host, hoststr),
1058 ntohs(host->port), fids[i].Volume,
1059 fids[i].Vnode, fids[i].Unique));
1062 * AddCallBack1_r(host, &fids[i], itot(thead[i]), CB_DELAYED, 1);
1063 * * but it turns out to cause too many tricky locking problems.
1064 * * now, if break delayed fails, screw it. */
1066 host->hostFlags |= VENUSDOWN; /* Failed */
1067 ClearHostCallbacks_r(host, 1 /* locked */ );
1071 if (nfids < AFSCBMAX)
1075 cbstuff.nbreakers--;
1076 /* If we succeeded it's always ok to unset HFE_LATER */
1077 if (!(host->hostFlags & VENUSDOWN))
1078 host->hostFlags &= ~HFE_LATER;
1079 return (host->hostFlags & VENUSDOWN);
1083 ** isheld is 0 if the host is held in h_Enumerate
1084 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1087 MultiBreakVolumeCallBack_r(struct host *host, int isheld,
1088 struct VCBParams *parms, int deletefe)
1093 return isheld; /* host is held only by h_Enumerate, do nothing */
1094 if (host->hostFlags & HOSTDELETED)
1095 return 0; /* host is deleted, release hold */
1097 if (host->hostFlags & VENUSDOWN) {
1099 if (host->hostFlags & HOSTDELETED) {
1101 return 0; /* Release hold */
1104 ("BVCB: volume call back for Host %x (%s:%d) failed\n",
1105 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
1108 ("CB: volume callback for Host %x (%s:%d) failed\n",
1109 host, afs_inet_ntoa_r(host->host, hoststr),
1110 ntohs(host->port)));
1112 DeleteAllCallBacks_r(host, deletefe); /* Delete all callback state
1113 * rather than attempting to
1114 * selectively remember to
1115 * delete the volume callbacks
1117 host->hostFlags &= ~RESETDONE; /* Do InitCallBackState when host returns */
1119 return 0; /* release hold */
1121 assert(parms->ncbas <= MAX_CB_HOSTS);
1123 /* Do not call MultiBreakCallBack on the current host structure
1124 ** because it would prematurely release the hold on the host
1126 if (parms->ncbas == MAX_CB_HOSTS) {
1127 struct AFSCBFids tf;
1129 tf.AFSCBFids_len = 1;
1130 tf.AFSCBFids_val = parms->fid;
1132 /* this releases all the hosts */
1133 MultiBreakCallBack_r(parms->cba, parms->ncbas, &tf, 0 /* xhost */ );
1137 parms->cba[parms->ncbas].hp = host;
1138 parms->cba[(parms->ncbas)++].thead = parms->thead;
1139 return 1; /* DON'T release hold, because we still need it. */
1143 ** isheld is 0 if the host is held in h_Enumerate
1144 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1147 MultiBreakVolumeCallBack(struct host *host, int isheld,
1148 struct VCBParams *parms)
1152 retval = MultiBreakVolumeCallBack_r(host, isheld, parms, 1);
1158 ** isheld is 0 if the host is held in h_Enumerate
1159 ** isheld is 1 if the host is held in BreakVolumeCallBacks
1162 MultiBreakVolumeLaterCallBack(struct host *host, int isheld,
1163 struct VCBParams *parms)
1167 retval = MultiBreakVolumeCallBack_r(host, isheld, parms, 0);
1173 * Break all call backs on a single volume. Don't call this with any
1174 * hosts h_held. Note that this routine clears the callbacks before
1175 * actually breaking them, and that the vnode isn't locked during this
1176 * operation, so that people might see temporary callback loss while
1177 * this function is executing. It is just a temporary state, however,
1178 * since the callback will be broken later by this same function.
1180 * Now uses multi-RX for CallBack RPC. Note that the
1181 * multiBreakCallBacks routine does not force a reset if the RPC
1182 * fails, unlike the previous version of this routine, but does create
1183 * a delayed callback. Resets will be forced if the host is
1184 * determined to be down before the RPC is executed.
1187 BreakVolumeCallBacks(afs_uint32 volume)
1192 struct CallBack *cb;
1193 struct FileEntry *fe;
1195 struct VCBParams henumParms;
1196 afs_uint32 tthead = 0; /* zero is illegal value */
1199 fid.Volume = volume, fid.Vnode = fid.Unique = 0;
1200 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1201 for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
1202 if (fe->volid == volume) {
1203 register struct CallBack *cbnext;
1204 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1205 host = h_itoh(cb->hhead);
1207 cbnext = itocb(cb->cnext);
1208 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1214 /* leave hold for MultiBreakVolumeCallBack to clear */
1225 /* didn't find any callbacks, so return right away. */
1229 henumParms.ncbas = 0;
1230 henumParms.fid = &fid;
1231 henumParms.thead = tthead;
1233 h_Enumerate(MultiBreakVolumeCallBack, (char *)&henumParms);
1235 if (henumParms.ncbas) { /* do left-overs */
1236 struct AFSCBFids tf;
1237 tf.AFSCBFids_len = 1;
1238 tf.AFSCBFids_val = &fid;
1240 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf, 0);
1242 henumParms.ncbas = 0;
1248 #ifdef AFS_PTHREAD_ENV
1249 extern pthread_cond_t fsync_cond;
1251 extern char fsync_wait[];
1255 BreakVolumeCallBacksLater(afs_uint32 volume)
1259 struct FileEntry *fe;
1260 struct CallBack *cb;
1264 ViceLog(25, ("Setting later on volume %u\n", volume));
1266 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1267 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1268 if (fe->volid == volume) {
1269 register struct CallBack *cbnext;
1270 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1271 host = h_itoh(cb->hhead);
1272 host->hostFlags |= HFE_LATER;
1273 cb->status = CB_DELAYED;
1274 cbnext = itocb(cb->cnext);
1277 fe->status |= FE_LATER;
1286 /* didn't find any callbacks, so return right away. */
1290 ViceLog(25, ("Fsync thread wakeup\n"));
1291 #ifdef AFS_PTHREAD_ENV
1293 assert(pthread_cond_broadcast(&fsync_cond) == 0);
1296 LWP_NoYieldSignal(fsync_wait);
1302 BreakLaterCallBacks(void)
1307 struct CallBack *cb;
1308 struct FileEntry *fe = NULL;
1309 struct FileEntry *myfe = NULL;
1311 struct VCBParams henumParms;
1312 unsigned short tthead = 0; /* zero is illegal value */
1316 ViceLog(25, ("Looking for FileEntries to unchain\n"));
1319 /* Pick the first volume we see to clean up */
1320 fid.Volume = fid.Vnode = fid.Unique = 0;
1322 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1323 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1324 if (fe && (fe->status & FE_LATER)
1325 && (fid.Volume == 0 || fid.Volume == fe->volid)) {
1326 /* Ugly, but used to avoid left side casting */
1327 struct object *tmpfe;
1329 ("Unchaining for %u:%u:%u\n", fe->vnode, fe->unique,
1331 fid.Volume = fe->volid;
1333 fe->status &= ~FE_LATER;
1334 /* Works since volid is deeper than the largest pointer */
1335 tmpfe = (struct object *)fe;
1336 tmpfe->next = (struct object *)myfe;
1349 /* loop over FEs from myfe and free/break */
1351 for (fe = myfe; fe;) {
1352 register struct CallBack *cbnext;
1353 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1354 cbnext = itocb(cb->cnext);
1355 host = h_itoh(cb->hhead);
1356 if (cb->status == CB_DELAYED) {
1358 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1363 CDel(cb, 0); /* Don't let CDel clean up the fe */
1364 /* leave hold for MultiBreakVolumeCallBack to clear */
1367 ("Found host %x (%s:%d) non-DELAYED cb for %u:%u:%u\n",
1368 host, afs_inet_ntoa_r(host->host, hoststr),
1369 ntohs(host->port), fe->vnode, fe->unique, fe->volid));
1373 fe = (struct FileEntry *)((struct object *)fe)->next;
1378 ViceLog(125, ("Breaking volume %u\n", fid.Volume));
1379 henumParms.ncbas = 0;
1380 henumParms.fid = &fid;
1381 henumParms.thead = tthead;
1383 h_Enumerate(MultiBreakVolumeLaterCallBack, (char *)&henumParms);
1385 if (henumParms.ncbas) { /* do left-overs */
1386 struct AFSCBFids tf;
1387 tf.AFSCBFids_len = 1;
1388 tf.AFSCBFids_val = &fid;
1390 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf, 0);
1391 henumParms.ncbas = 0;
1396 /* Arrange to be called again */
1401 * Delete all timed-out call back entries (to be called periodically by file
1405 CleanupTimedOutCallBacks(void)
1408 CleanupTimedOutCallBacks_r();
1414 CleanupTimedOutCallBacks_r(void)
1416 afs_uint32 now = CBtime(FT_ApproxTime());
1417 register afs_uint32 *thead;
1418 register struct CallBack *cb;
1419 register int ntimedout = 0;
1422 while (tfirst <= now) {
1424 cbi = *(thead = THead(tfirst));
1430 ("CCB: deleting timed out call back %x (%s:%d), (%u,%u,%u)\n",
1431 h_itoh(cb->hhead)->host,
1432 afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
1433 h_itoh(cb->hhead)->port, itofe(cb->fhead)->volid,
1434 itofe(cb->fhead)->vnode, itofe(cb->fhead)->unique));
1438 if (ntimedout > cbstuff.nblks) {
1439 ViceLog(0, ("CCB: Internal Error -- shutting down...\n"));
1440 DumpCallBackState();
1441 ShutDownAndCore(PANIC);
1443 } while (cbi != *thead);
1448 cbstuff.CBsTimedOut += ntimedout;
1449 ViceLog(7, ("CCB: deleted %d timed out callbacks\n", ntimedout));
1450 return (ntimedout > 0);
1453 static struct host *lih_host;
1454 static int lih_host_held;
1456 /* This version does not allow 'host' to be selected unless its ActiveCall
1457 * is newer than 'hostp' which is the host with the oldest ActiveCall from
1458 * the last pass (if it is provided). We filter out any hosts that are
1459 * are held by other threads.
1462 lih0_r(register struct host *host, register int held,
1463 register struct host *hostp)
1466 && (hostp && host != hostp)
1467 && (!held && !h_OtherHolds_r(host))
1468 && (!lih_host || host->ActiveCall < lih_host->ActiveCall)
1469 && (!hostp || host->ActiveCall > hostp->ActiveCall)) {
1470 if (lih_host != NULL && lih_host_held) {
1471 h_Release_r(lih_host);
1474 lih_host_held = !held;
1480 /* This version does not allow 'host' to be selected unless its ActiveCall
1481 * is newer than 'hostp' which is the host with the oldest ActiveCall from
1482 * the last pass (if it is provided). In this second varient, we do not
1483 * prevent held hosts from being selected.
1486 lih1_r(register struct host *host, register int held,
1487 register struct host *hostp)
1490 && (hostp && host != hostp)
1491 && (!lih_host || host->ActiveCall < lih_host->ActiveCall)
1492 && (!hostp || host->ActiveCall > hostp->ActiveCall)) {
1493 if (lih_host != NULL && lih_host_held) {
1494 h_Release_r(lih_host);
1497 lih_host_held = !held;
1503 /* This could be upgraded to get more space each time */
1504 /* first pass: sequentially find the oldest host which isn't held by
1505 anyone for which we can clear callbacks;
1507 /* second pass: sequentially find the oldest host regardless of
1508 whether or not the host is held; skipping 'hostp' */
1509 /* third pass: attempt to clear callbacks from 'hostp' */
1510 /* always called with hostp unlocked */
1512 /* Note: hostlist is ordered most recently created host first and
1513 * its order has no relationship to the most recently used. */
1514 extern struct host *hostList;
1516 GetSomeSpace_r(struct host *hostp, int locked)
1518 register struct host *hp, *hp1, *hp2;
1521 cbstuff.GotSomeSpaces++;
1523 ("GSS: First looking for timed out call backs via CleanupCallBacks\n"));
1524 if (CleanupTimedOutCallBacks_r()) {
1534 h_Enumerate_r(i == 0 ? lih0_r : lih1_r, hp2, (char *)hp1);
1537 /* set in lih_r! private copy before giving up H_LOCK */
1538 int lih_host_held2=lih_host_held;
1540 if ((hp != hostp) && !ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
1551 * Next time try getting callbacks from any host even if
1552 * it's deleted (that's actually great since we can freely
1553 * remove its callbacks) or it's held since the only other
1554 * option is starvation for the file server (i.e. until the
1555 * callback timeout arrives).
1562 ("GSS: Try harder for longest inactive host cnt= %d\n",
1567 /* Could not obtain space from other hosts, clear hostp's callback state */
1572 ClearHostCallbacks_r(hostp, 1 /*already locked */ );
1579 /* locked - set if caller has already locked the host */
1581 ClearHostCallbacks_r(struct host *hp, int locked)
1586 struct rx_connection *cb_conn = NULL;
1589 ("GSS: Delete longest inactive host %x (%s:%d)\n",
1590 hp, afs_inet_ntoa_r(hp->host, hoststr), ntohs(hp->port)));
1591 if (!(held = h_Held_r(hp)))
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)) {
1604 if (hp->Console & 2) {
1606 * If the special console field is set it means that a thread
1607 * is waiting in AddCallBack1 after it set pointers to the
1608 * file entry and/or callback entry. Because of the bogus
1609 * usage of h_hold it won't prevent from another thread, this
1610 * one, to remove all the callbacks so just to be safe we keep
1611 * a reference. NOTE, on the last phase we'll free the calling
1612 * host's callbacks but that's ok...
1616 DeleteAllCallBacks_r(hp, 1);
1617 if (hp->hostFlags & VENUSDOWN) {
1618 hp->hostFlags &= ~RESETDONE; /* remember that we must do a reset */
1620 /* host is up, try a call */
1621 hp->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1622 cb_conn = hp->callback_rxcon;
1623 rx_GetConnection(hp->callback_rxcon);
1624 if (hp->interface) {
1627 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1630 code = RXAFSCB_InitCallBackState(cb_conn);
1632 rx_PutConnection(cb_conn);
1635 hp->hostFlags |= ALTADDR; /* alternate addresses are valid */
1637 /* failed, mark host down and need reset */
1638 hp->hostFlags |= VENUSDOWN;
1639 hp->hostFlags &= ~RESETDONE;
1641 /* reset succeeded, we're done */
1642 hp->hostFlags |= RESETDONE;
1653 #endif /* INTERPRET_DUMP */
1657 PrintCallBackStats(void)
1660 "%d add CB, %d break CB, %d del CB, %d del FE, %d CB's timed out, %d space reclaim, %d del host\n",
1661 cbstuff.AddCallBacks, cbstuff.BreakCallBacks,
1662 cbstuff.DeleteCallBacks, cbstuff.DeleteFiles, cbstuff.CBsTimedOut,
1663 cbstuff.GotSomeSpaces, cbstuff.DeleteAllCallBacks);
1664 fprintf(stderr, "%d CBs, %d FEs, (%d of total of %d 16-byte blocks)\n",
1665 cbstuff.nCBs, cbstuff.nFEs, cbstuff.nCBs + cbstuff.nFEs,
1671 #define MAGIC 0x12345678 /* To check byte ordering of dump when it is read in */
1673 #ifndef INTERPRET_DUMP
1675 #ifdef AFS_DEMAND_ATTACH_FS
1678 * callback state serialization
1680 static int cb_stateSaveTimeouts(struct fs_dump_state * state);
1681 static int cb_stateSaveFEHash(struct fs_dump_state * state);
1682 static int cb_stateSaveFEs(struct fs_dump_state * state);
1683 static int cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe);
1684 static int cb_stateRestoreTimeouts(struct fs_dump_state * state);
1685 static int cb_stateRestoreFEHash(struct fs_dump_state * state);
1686 static int cb_stateRestoreFEs(struct fs_dump_state * state);
1687 static int cb_stateRestoreFE(struct fs_dump_state * state);
1688 static int cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
1689 struct iovec * iov, int niovecs);
1691 static int cb_stateVerifyFEHash(struct fs_dump_state * state);
1692 static int cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe);
1693 static int cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe);
1694 static int cb_stateVerifyTimeoutQueues(struct fs_dump_state * state);
1696 static int cb_stateFEToDiskEntry(struct FileEntry *, struct FEDiskEntry *);
1697 static int cb_stateDiskEntryToFE(struct fs_dump_state * state,
1698 struct FEDiskEntry *, struct FileEntry *);
1700 static int cb_stateCBToDiskEntry(struct CallBack *, struct CBDiskEntry *);
1701 static int cb_stateDiskEntryToCB(struct fs_dump_state * state,
1702 struct CBDiskEntry *, struct CallBack *);
1704 static int cb_stateFillHeader(struct callback_state_header * hdr);
1705 static int cb_stateCheckHeader(struct callback_state_header * hdr);
1707 static int cb_stateAllocMap(struct fs_dump_state * state);
1710 cb_stateSave(struct fs_dump_state * state)
1714 AssignInt64(state->eof_offset, &state->hdr->cb_offset);
1716 /* invalidate callback state header */
1717 memset(state->cb_hdr, 0, sizeof(struct callback_state_header));
1718 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1719 sizeof(struct callback_state_header))) {
1724 fs_stateIncEOF(state, sizeof(struct callback_state_header));
1726 /* dump timeout state */
1727 if (cb_stateSaveTimeouts(state)) {
1732 /* dump fe hashtable state */
1733 if (cb_stateSaveFEHash(state)) {
1738 /* dump callback state */
1739 if (cb_stateSaveFEs(state)) {
1744 /* write the callback state header to disk */
1745 cb_stateFillHeader(state->cb_hdr);
1746 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1747 sizeof(struct callback_state_header))) {
1757 cb_stateRestore(struct fs_dump_state * state)
1761 if (fs_stateReadHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1762 sizeof(struct callback_state_header))) {
1767 if (cb_stateCheckHeader(state->cb_hdr)) {
1772 if (cb_stateAllocMap(state)) {
1777 if (cb_stateRestoreTimeouts(state)) {
1782 if (cb_stateRestoreFEHash(state)) {
1787 /* restore FEs and CBs from disk */
1788 if (cb_stateRestoreFEs(state)) {
1793 /* restore the timeout queue heads */
1794 tfirst = state->cb_hdr->tfirst;
1801 cb_stateRestoreIndices(struct fs_dump_state * state)
1804 struct FileEntry * fe;
1805 struct CallBack * cb;
1807 /* restore indices in the FileEntry structures */
1808 for (i = 1; i < state->fe_map.len; i++) {
1809 if (state->fe_map.entries[i].new_idx) {
1810 fe = itofe(state->fe_map.entries[i].new_idx);
1812 /* restore the fe->fnext entry */
1813 if (fe_OldToNew(state, fe->fnext, &fe->fnext)) {
1818 /* restore the fe->firstcb entry */
1819 if (cb_OldToNew(state, fe->firstcb, &fe->firstcb)) {
1826 /* restore indices in the CallBack structures */
1827 for (i = 1; i < state->cb_map.len; i++) {
1828 if (state->cb_map.entries[i].new_idx) {
1829 cb = itocb(state->cb_map.entries[i].new_idx);
1831 /* restore the cb->cnext entry */
1832 if (cb_OldToNew(state, cb->cnext, &cb->cnext)) {
1837 /* restore the cb->fhead entry */
1838 if (fe_OldToNew(state, cb->fhead, &cb->fhead)) {
1843 /* restore the cb->hhead entry */
1844 if (h_OldToNew(state, cb->hhead, &cb->hhead)) {
1849 /* restore the cb->tprev entry */
1850 if (cb_OldToNew(state, cb->tprev, &cb->tprev)) {
1855 /* restore the cb->tnext entry */
1856 if (cb_OldToNew(state, cb->tnext, &cb->tnext)) {
1861 /* restore the cb->hprev entry */
1862 if (cb_OldToNew(state, cb->hprev, &cb->hprev)) {
1867 /* restore the cb->hnext entry */
1868 if (cb_OldToNew(state, cb->hnext, &cb->hnext)) {
1875 /* restore the timeout queue head indices */
1876 for (i = 0; i < state->cb_timeout_hdr->records; i++) {
1877 if (cb_OldToNew(state, timeout[i], &timeout[i])) {
1883 /* restore the FE hash table queue heads */
1884 for (i = 0; i < state->cb_fehash_hdr->records; i++) {
1885 if (fe_OldToNew(state, HashTable[i], &HashTable[i])) {
1896 cb_stateVerify(struct fs_dump_state * state)
1900 if (cb_stateVerifyFEHash(state)) {
1904 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));
1947 cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe)
1951 if ((fe->firstcb && !fe->ncbs) ||
1952 (!fe->firstcb && fe->ncbs)) {
1953 ViceLog(0, ("cb_stateVerifyFE: error: fe->firstcb does not agree with fe->ncbs (fei=%d, fe->firstcb=%d, fe->ncbs=%d)\n",
1954 fetoi(fe), fe->firstcb, fe->ncbs));
1957 if (cb_stateVerifyFCBList(state, fe)) {
1958 ViceLog(0, ("cb_stateVerifyFE: error: FCBList failed verification (fei=%d)\n", fetoi(fe)));
1967 cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe)
1970 afs_uint32 cbi, fei, chain_len = 0;
1971 struct CallBack * cb;
1975 for (cbi = fe->firstcb, cb = itocb(cbi);
1977 cbi = cb->cnext, cb = itocb(cbi)) {
1978 if (cbi > cbstuff.nblks) {
1979 ViceLog(0, ("cb_stateVerifyFCBList: error: list index out of range (cbi=%d, ncbs=%d)\n",
1980 cbi, cbstuff.nblks));
1984 if (cb->fhead != fei) {
1985 ViceLog(0, ("cb_stateVerifyFCBList: error: cb->fhead != fei (fei=%d, cb->fhead=%d)\n",
1989 if (chain_len > FS_STATE_FCB_MAX_LIST_LEN) {
1990 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (fei=%d); assuming there's a loop\n",
1991 FS_STATE_FCB_MAX_LIST_LEN, fei));
1998 if (fe->ncbs != chain_len) {
1999 ViceLog(0, ("cb_stateVerifyFCBList: error: list length mismatch (len=%d, fe->ncbs=%d)\n",
2000 chain_len, fe->ncbs));
2009 cb_stateVerifyHCBList(struct fs_dump_state * state, struct host * host)
2012 afs_uint32 hi, chain_len, cbi;
2013 struct CallBack *cb, *ncb;
2018 for (cbi = host->cblist, cb = itocb(cbi);
2020 cbi = cb->hnext, cb = ncb) {
2021 if (chain_len && (host->cblist == cbi)) {
2022 /* we've wrapped around the circular list, and everything looks ok */
2025 if (cb->hhead != hi) {
2026 ViceLog(0, ("cb_stateVerifyHCBList: error: incorrect cb->hhead (cbi=%d, h->index=%d, cb->hhead=%d)\n",
2027 cbi, hi, cb->hhead));
2030 if (!cb->hprev || !cb->hnext) {
2031 ViceLog(0, ("cb_stateVerifyHCBList: error: null index in circular list (cbi=%d, h->index=%d)\n",
2036 if ((cb->hprev > cbstuff.nblks) ||
2037 (cb->hnext > cbstuff.nblks)) {
2038 ViceLog(0, ("cb_stateVerifyHCBList: error: list index out of range (cbi=%d, h->index=%d, cb->hprev=%d, cb->hnext=%d, nCBs=%d)\n",
2039 cbi, hi, cb->hprev, cb->hnext, cbstuff.nblks));
2043 ncb = itocb(cb->hnext);
2044 if (cbi != ncb->hprev) {
2045 ViceLog(0, ("cb_stateVerifyHCBList: error: corrupt linked list (cbi=%d, h->index=%d)\n",
2050 if (chain_len > FS_STATE_HCB_MAX_LIST_LEN) {
2051 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (h->index=%d); assuming there's a loop\n",
2052 FS_STATE_HCB_MAX_LIST_LEN, hi));
2064 cb_stateVerifyTimeoutQueues(struct fs_dump_state * state)
2067 afs_uint32 cbi, chain_len;
2068 struct CallBack *cb, *ncb;
2070 for (i = 0; i < CB_NUM_TIMEOUT_QUEUES; i++) {
2072 for (cbi = timeout[i], cb = itocb(cbi);
2074 cbi = cb->tnext, cb = ncb) {
2075 if (chain_len && (cbi == timeout[i])) {
2076 /* we've wrapped around the circular list, and everything looks ok */
2079 if (cbi > cbstuff.nblks) {
2080 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: list index out of range (cbi=%d, tindex=%d)\n",
2085 if (itot(cb->thead) != &timeout[i]) {
2086 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: cb->thead points to wrong timeout queue (tindex=%d, cbi=%d, cb->thead=%d)\n",
2087 i, cbi, cb->thead));
2090 if (!cb->tprev || !cb->tnext) {
2091 ViceLog(0, ("cb_stateVerifyTimeoutQueues: null index in circular list (cbi=%d, tindex=%d)\n",
2096 if ((cb->tprev > cbstuff.nblks) ||
2097 (cb->tnext > cbstuff.nblks)) {
2098 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list index out of range (cbi=%d, tindex=%d, cb->tprev=%d, cb->tnext=%d, nCBs=%d)\n",
2099 cbi, i, cb->tprev, cb->tnext, cbstuff.nblks));
2103 ncb = itocb(cb->tnext);
2104 if (cbi != ncb->tprev) {
2105 ViceLog(0, ("cb_stateVerifyTimeoutQueues: corrupt linked list (cbi=%d, tindex=%d)\n",
2110 if (chain_len > FS_STATE_TCB_MAX_LIST_LEN) {
2111 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list length exceeds %d (tindex=%d); assuming there's a loop\n",
2112 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)
2274 register int fei, hash;
2275 register 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;
2405 struct CallBack * cb;
2407 iov[0].iov_base = (char *)&hdr;
2408 iov[0].iov_len = sizeof(hdr);
2409 iov[1].iov_base = (char *)&fedsk;
2410 iov[1].iov_len = sizeof(fedsk);
2413 if (fs_stateReadV(state, iov, iovcnt)) {
2418 if (hdr.magic != CALLBACK_STATE_ENTRY_MAGIC) {
2425 ViceLog(0, ("cb_stateRestoreFE: ran out of free FileEntry structures\n"));
2430 if (cb_stateDiskEntryToFE(state, &fedsk, fe)) {
2436 for (iovcnt = 0, nCBs = 0;
2439 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2440 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2442 if ((iovcnt == 16) || (nCBs == hdr.nCBs - 1)) {
2443 if (fs_stateReadV(state, iov, iovcnt)) {
2447 if (cb_stateRestoreCBs(state, fe, iov, iovcnt)) {
2461 cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
2462 struct iovec * iov, int niovecs)
2465 register struct CallBack * cb;
2466 struct CBDiskEntry * cbdsk;
2471 for (idx = 0; idx < niovecs; idx++) {
2472 cbdsk = (struct CBDiskEntry *) iov[idx].iov_base;
2473 if ((cb = GetCB()) == NULL) {
2474 ViceLog(0, ("cb_stateRestoreCBs: ran out of free CallBack structures\n"));
2478 if (cb_stateDiskEntryToCB(state, cbdsk, cb)) {
2479 ViceLog(0, ("cb_stateRestoreCBs: corrupt CallBack disk entry\n"));
2491 cb_stateFillHeader(struct callback_state_header * hdr)
2493 hdr->stamp.magic = CALLBACK_STATE_MAGIC;
2494 hdr->stamp.version = CALLBACK_STATE_VERSION;
2495 hdr->tfirst = tfirst;
2500 cb_stateCheckHeader(struct callback_state_header * hdr)
2504 if (hdr->stamp.magic != CALLBACK_STATE_MAGIC) {
2506 } else if (hdr->stamp.version != CALLBACK_STATE_VERSION) {
2508 } else if ((hdr->nFEs > cbstuff.nblks) || (hdr->nCBs > cbstuff.nblks)) {
2509 ViceLog(0, ("cb_stateCheckHeader: saved callback state larger than callback memory allocation\n"));
2515 /* disk entry conversion routines */
2517 cb_stateFEToDiskEntry(struct FileEntry * in, struct FEDiskEntry * out)
2519 memcpy(&out->fe, in, sizeof(struct FileEntry));
2520 out->index = fetoi(in);
2525 cb_stateDiskEntryToFE(struct fs_dump_state * state,
2526 struct FEDiskEntry * in, struct FileEntry * out)
2530 memcpy(out, &in->fe, sizeof(struct FileEntry));
2532 /* setup FE map entry */
2533 if (!in->index || (in->index >= state->fe_map.len)) {
2534 ViceLog(0, ("cb_stateDiskEntryToFE: index (%d) out of range",
2539 state->fe_map.entries[in->index].old_idx = in->index;
2540 state->fe_map.entries[in->index].new_idx = fetoi(out);
2547 cb_stateCBToDiskEntry(struct CallBack * in, struct CBDiskEntry * out)
2549 memcpy(&out->cb, in, sizeof(struct CallBack));
2550 out->index = cbtoi(in);
2555 cb_stateDiskEntryToCB(struct fs_dump_state * state,
2556 struct CBDiskEntry * in, struct CallBack * out)
2560 memcpy(out, &in->cb, sizeof(struct CallBack));
2562 /* setup CB map entry */
2563 if (!in->index || (in->index >= state->cb_map.len)) {
2564 ViceLog(0, ("cb_stateDiskEntryToCB: index (%d) out of range\n",
2569 state->cb_map.entries[in->index].old_idx = in->index;
2570 state->cb_map.entries[in->index].new_idx = cbtoi(out);
2576 /* index map routines */
2578 cb_stateAllocMap(struct fs_dump_state * state)
2580 state->fe_map.len = state->cb_hdr->fe_max + 1;
2581 state->cb_map.len = state->cb_hdr->cb_max + 1;
2582 state->fe_map.entries = (struct idx_map_entry_t *)
2583 calloc(state->fe_map.len, sizeof(struct idx_map_entry_t));
2584 state->cb_map.entries = (struct idx_map_entry_t *)
2585 calloc(state->cb_map.len, sizeof(struct idx_map_entry_t));
2586 return ((state->fe_map.entries != NULL) && (state->cb_map.entries != NULL)) ? 0 : 1;
2590 fe_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2594 /* FEs use a one-based indexing system, so old==0 implies no mapping */
2600 if (old >= state->fe_map.len) {
2601 ViceLog(0, ("fe_OldToNew: index %d is out of range\n", old));
2603 } else if (state->fe_map.entries[old].old_idx != old) { /* sanity check */
2604 ViceLog(0, ("fe_OldToNew: index %d points to an invalid FileEntry record\n", old));
2607 *new = state->fe_map.entries[old].new_idx;
2615 cb_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2619 /* CBs use a one-based indexing system, so old==0 implies no mapping */
2625 if (old >= state->cb_map.len) {
2626 ViceLog(0, ("cb_OldToNew: index %d is out of range\n", old));
2628 } else if (state->cb_map.entries[old].old_idx != old) { /* sanity check */
2629 ViceLog(0, ("cb_OldToNew: index %d points to an invalid CallBack record\n", old));
2632 *new = state->cb_map.entries[old].new_idx;
2638 #endif /* AFS_DEMAND_ATTACH_FS */
2641 DumpCallBackState(void)
2644 afs_uint32 magic = MAGIC, now = FT_ApproxTime(), freelisthead;
2646 fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, O_WRONLY | O_CREAT | O_TRUNC,
2650 ("Couldn't create callback dump file %s\n",
2651 AFSDIR_SERVER_CBKDUMP_FILEPATH));
2654 (void)write(fd, &magic, sizeof(magic));
2655 (void)write(fd, &now, sizeof(now));
2656 (void)write(fd, &cbstuff, sizeof(cbstuff));
2657 (void)write(fd, TimeOuts, sizeof(TimeOuts));
2658 (void)write(fd, timeout, sizeof(timeout));
2659 (void)write(fd, &tfirst, sizeof(tfirst));
2660 freelisthead = cbtoi((struct CallBack *)CBfree);
2661 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2662 freelisthead = fetoi((struct FileEntry *)FEfree);
2663 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2664 (void)write(fd, HashTable, sizeof(HashTable));
2665 (void)write(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2666 (void)write(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2672 #endif /* !INTERPRET_DUMP */
2674 #ifdef INTERPRET_DUMP
2676 /* This is only compiled in for the callback analyzer program */
2677 /* Returns the time of the dump */
2679 ReadDump(char *file)
2682 afs_uint32 magic, freelisthead;
2685 fd = open(file, O_RDONLY);
2687 fprintf(stderr, "Couldn't read dump file %s\n", file);
2690 read(fd, &magic, sizeof(magic));
2691 if (magic != MAGIC) {
2693 "Magic number of %s is invalid. You might be trying to\n",
2696 "run this program on a machine type with a different byte ordering.\n");
2699 read(fd, &now, sizeof(now));
2700 read(fd, &cbstuff, sizeof(cbstuff));
2701 read(fd, TimeOuts, sizeof(TimeOuts));
2702 read(fd, timeout, sizeof(timeout));
2703 read(fd, &tfirst, sizeof(tfirst));
2704 read(fd, &freelisthead, sizeof(freelisthead));
2705 CB = ((struct CallBack
2706 *)(calloc(cbstuff.nblks, sizeof(struct CallBack)))) - 1;
2707 FE = ((struct FileEntry
2708 *)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
2709 CBfree = (struct CallBack *)itocb(freelisthead);
2710 read(fd, &freelisthead, sizeof(freelisthead));
2711 FEfree = (struct FileEntry *)itofe(freelisthead);
2712 read(fd, HashTable, sizeof(HashTable));
2713 read(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2714 read(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2716 perror("Error reading dumpfile");
2723 #include "AFS_component_version_number.h"
2725 #include "AFS_component_version_number.c"
2729 main(int argc, char **argv)
2731 int err = 0, cbi = 0, stats = 0, noptions = 0, all = 0, vol = 0, raw = 0;
2733 register struct FileEntry *fe;
2734 register struct CallBack *cb;
2737 memset(&fid, 0, sizeof(fid));
2740 while (argc && **argv == '-') {
2743 if (!strcmp(*argv, "-host")) {
2749 cbi = atoi(*++argv);
2750 } else if (!strcmp(*argv, "-fid")) {
2756 fid.Volume = atoi(*++argv);
2757 fid.Vnode = atoi(*++argv);
2758 fid.Unique = atoi(*++argv);
2759 } else if (!strcmp(*argv, "-time")) {
2760 fprintf(stderr, "-time not supported\n");
2762 } else if (!strcmp(*argv, "-stats")) {
2764 } else if (!strcmp(*argv, "-all")) {
2766 } else if (!strcmp(*argv, "-raw")) {
2768 } else if (!strcmp(*argv, "-volume")) {
2774 vol = atoi(*++argv);
2779 if (err || argc != 1) {
2781 "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] callbackdumpfile\n");
2783 "[cbid is shown for each host in the hosts.dump file]\n");
2786 now = ReadDump(*argv);
2787 if (stats || noptions == 0) {
2788 time_t uxtfirst = UXtime(tfirst);
2789 printf("The time of the dump was %u %s", now, ctime(&now));
2790 printf("The last time cleanup ran was %u %s", uxtfirst,
2792 PrintCallBackStats();
2797 struct CallBack *cb;
2798 struct FileEntry *fe;
2800 for (hash = 0; hash < FEHASH_SIZE; hash++) {
2801 for (feip = &HashTable[hash]; fe = itofe(*feip);) {
2802 if (!vol || (fe->volid == vol)) {
2803 register struct CallBack *cbnext;
2804 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
2806 cbnext = itocb(cb->cnext);
2816 afs_uint32 cfirst = cbi;
2821 } while (cbi != cfirst);
2826 printf("No callback entries for %u.%u\n", fid.Volume, fid.Vnode);
2829 cb = itocb(fe->firstcb);
2832 cb = itocb(cb->cnext);
2837 for (i = 1; i < cbstuff.nblks; i++) {
2838 p = (afs_int32 *) & FE[i];
2839 printf("%d:%12x%12x%12x%12x\n", i, p[0], p[1], p[2], p[3]);
2846 PrintCB(register struct CallBack *cb, afs_uint32 now)
2848 struct FileEntry *fe = itofe(cb->fhead);
2849 time_t expires = TIndexToTime(cb->thead);
2854 printf("vol=%u vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %d secs at %s",
2855 fe->volid, fe->vnode, fe->ncbs, cb->hhead, cb->status, fe->status,
2856 expires - now, ctime(&expires));
2861 #if !defined(INTERPRET_DUMP)
2863 ** try breaking calbacks on afidp from host. Use multi_rx.
2864 ** return 0 on success, non-zero on failure
2867 MultiBreakCallBackAlternateAddress(struct host *host, struct AFSCBFids *afidp)
2871 retVal = MultiBreakCallBackAlternateAddress_r(host, afidp);
2877 MultiBreakCallBackAlternateAddress_r(struct host *host,
2878 struct AFSCBFids *afidp)
2881 struct rx_connection **conns;
2882 struct rx_connection *connSuccess = 0;
2883 struct AddrPort *interfaces;
2884 static struct rx_securityClass *sc = 0;
2885 static struct AFSCBs tc = { 0, 0 };
2888 /* nothing more can be done */
2889 if (!host->interface)
2890 return 1; /* failure */
2892 /* the only address is the primary interface */
2893 if (host->interface->numberOfInterfaces <= 1)
2894 return 1; /* failure */
2896 /* initialise a security object only once */
2898 sc = rxnull_NewClientSecurityObject();
2900 i = host->interface->numberOfInterfaces;
2901 interfaces = calloc(i, sizeof(struct AddrPort));
2902 conns = calloc(i, sizeof(struct rx_connection *));
2903 if (!interfaces || !conns) {
2905 ("Failed malloc in MultiBreakCallBackAlternateAddress_r\n"));
2909 /* initialize alternate rx connections */
2910 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
2911 /* this is the current primary address */
2912 if (host->host == host->interface->interface[i].addr &&
2913 host->port == host->interface->interface[i].port)
2916 interfaces[j] = host->interface->interface[i];
2918 rx_NewConnection(interfaces[j].addr,
2919 interfaces[j].port, 1, sc, 0);
2920 rx_SetConnDeadTime(conns[j], 2);
2921 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
2925 assert(j); /* at least one alternate address */
2927 ("Starting multibreakcall back on all addr for host %x (%s:%d)\n",
2928 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
2930 multi_Rx(conns, j) {
2931 multi_RXAFSCB_CallBack(afidp, &tc);
2935 if (host->callback_rxcon)
2936 rx_DestroyConnection(host->callback_rxcon);
2937 host->callback_rxcon = conns[multi_i];
2938 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
2939 host->host = interfaces[multi_i].addr;
2940 host->port = interfaces[multi_i].port;
2941 h_AddHostToAddrHashTable_r(host->host, host->port, host);
2942 connSuccess = conns[multi_i];
2943 rx_SetConnDeadTime(host->callback_rxcon, 50);
2944 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
2946 ("multibreakcall success with addr %s:%d\n",
2947 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
2948 ntohs(interfaces[multi_i].port)));
2955 /* Destroy all connections except the one on which we succeeded */
2956 for (i = 0; i < j; i++)
2957 if (conns[i] != connSuccess)
2958 rx_DestroyConnection(conns[i]);
2964 return 0; /* success */
2966 return 1; /* failure */
2971 ** try multi_RX probes to host.
2972 ** return 0 on success, non-0 on failure
2975 MultiProbeAlternateAddress_r(struct host *host)
2978 struct rx_connection **conns;
2979 struct rx_connection *connSuccess = 0;
2980 struct AddrPort *interfaces;
2981 static struct rx_securityClass *sc = 0;
2984 /* nothing more can be done */
2985 if (!host->interface)
2986 return 1; /* failure */
2988 /* the only address is the primary interface */
2989 if (host->interface->numberOfInterfaces <= 1)
2990 return 1; /* failure */
2992 /* initialise a security object only once */
2994 sc = rxnull_NewClientSecurityObject();
2996 i = host->interface->numberOfInterfaces;
2997 interfaces = calloc(i, sizeof(struct AddrPort));
2998 conns = calloc(i, sizeof(struct rx_connection *));
2999 if (!interfaces || !conns) {
3000 ViceLog(0, ("Failed malloc in MultiProbeAlternateAddress_r\n"));
3004 /* initialize alternate rx connections */
3005 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3006 /* this is the current primary address */
3007 if (host->host == host->interface->interface[i].addr &&
3008 host->port == host->interface->interface[i].port)
3011 interfaces[j] = host->interface->interface[i];
3013 rx_NewConnection(interfaces[j].addr,
3014 interfaces[j].port, 1, sc, 0);
3015 rx_SetConnDeadTime(conns[j], 2);
3016 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3020 assert(j); /* at least one alternate address */
3022 ("Starting multiprobe on all addr for host %x (%s:%d)\n",
3023 host, afs_inet_ntoa_r(host->host, hoststr),
3024 ntohs(host->port)));
3026 multi_Rx(conns, j) {
3027 multi_RXAFSCB_ProbeUuid(&host->interface->uuid);
3031 if (host->callback_rxcon)
3032 rx_DestroyConnection(host->callback_rxcon);
3033 host->callback_rxcon = conns[multi_i];
3034 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
3035 host->host = interfaces[multi_i].addr;
3036 host->port = interfaces[multi_i].port;
3037 h_AddHostToAddrHashTable_r(host->host, host->port, host);
3038 connSuccess = conns[multi_i];
3039 rx_SetConnDeadTime(host->callback_rxcon, 50);
3040 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3042 ("multiprobe success with addr %s:%d\n",
3043 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3044 ntohs(interfaces[multi_i].port)));
3049 ("multiprobe failure with addr %s:%d\n",
3050 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3051 ntohs(interfaces[multi_i].port)));
3053 /* This is less than desirable but its the best we can do.
3054 * The AFS Cache Manager will return either 0 for a Uuid
3055 * match and a 1 for a non-match. If the error is 1 we
3056 * therefore know that our mapping of IP address to Uuid
3057 * is wrong. We should attempt to find the correct
3058 * Uuid and fix the host tables.
3060 if (multi_error == 1) {
3061 /* remove the current alternate address from this host */
3063 removeInterfaceAddr_r(host, interfaces[multi_i].addr, interfaces[multi_i].port);
3067 #ifdef AFS_DEMAND_ATTACH_FS
3068 /* try to bail ASAP if the fileserver is shutting down */
3070 if (fs_state.mode == FS_MODE_SHUTDOWN) {
3079 /* Destroy all connections except the one on which we succeeded */
3080 for (i = 0; i < j; i++)
3081 if (conns[i] != connSuccess)
3082 rx_DestroyConnection(conns[i]);
3088 return 0; /* success */
3090 return 1; /* failure */
3093 #endif /* !defined(INTERPRET_DUMP) */