2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
9 * Portions Copyright (c) 2006 Sine Nomine Associates
13 * NEW callback package callback.c (replaces vicecb.c)
14 * Updated call back routines, NOW with:
16 * Faster DeleteVenus (Now called DeleteAllCallBacks)
17 * Call back breaking for volumes
18 * Adaptive timeouts on call backs
19 * Architected for Multi RPC
20 * No locks (currently implicit vnode locks--these will go, to)
21 * Delayed call back when rpc connection down.
22 * Bulk break of delayed call backs when rpc connection
24 * Strict limit on number of call backs.
26 * InitCallBack(nblocks)
27 * Initialize: nblocks is max number # of file entries + # of callback entries
28 * nblocks must be < 65536
29 * Space used is nblocks*16 bytes
30 * Note that space will be reclaimed by breaking callbacks of old hosts
32 * time = AddCallBack(host, fid)
34 * Returns the expiration time at the workstation.
36 * BreakCallBack(host, fid)
37 * Break all call backs for fid, except for the specified host.
40 * BreakVolumeCallBacksLater(volume)
41 * Break all call backs on volume, using single call to each host
42 * Delete all the call backs.
44 * DeleteCallBack(host,fid)
45 * Delete (do not break) single call back for fid.
47 * DeleteFileCallBacks(fid)
48 * Delete (do not break) all call backs for fid.
50 * DeleteAllCallBacks(host)
51 * Delete (do not break) all call backs for host.
53 * CleanupTimedOutCallBacks()
54 * Delete all timed out call back entries
55 * Must be called periodically by file server.
57 * BreakDelayedCallBacks(host)
58 * Break all delayed call backs for host.
59 * Returns 1: one or more failed, 0: success.
61 * PrintCallBackStats()
62 * Print statistics about call backs to stdout.
64 * DumpCallBacks() ---wishful thinking---
65 * Dump call back state to /tmp/callback.state.
66 * This is separately interpretable by the program pcb.
68 * Notes: In general, if a call back to a host doesn't get through,
69 * then HostDown, supplied elsewhere, is called. BreakDelayedCallBacks,
70 * however, does not call HostDown, but instead returns an indication of
71 * success if all delayed call backs were finally broken.
73 * BreakDelayedCallBacks MUST be called at the first sign of activity
74 * from the host after HostDown has been called (or a previous
75 * BreakDelayedCallBacks failed). The BreakDelayedCallBacks must be
76 * allowed to complete before any requests from that host are handled.
77 * If BreakDelayedCallBacks fails, then the host should remain
78 * down (and the request should be failed).
80 * CleanupCallBacks MUST be called periodically by the file server for
81 * this package to work correctly. Every 5 minutes is suggested.
84 #include <afsconfig.h>
85 #include <afs/param.h>
90 #ifdef HAVE_SYS_FILE_H
96 #include <afs/nfs.h> /* yuck. This is an abomination. */
98 #include <rx/rx_queue.h>
99 #include <afs/afscbint.h>
100 #include <afs/afsutil.h>
101 #include <afs/ihandle.h>
102 #include <afs/partition.h>
103 #include <afs/vnode.h>
104 #include <afs/volume.h>
105 #include "viced_prototypes.h"
108 #include <afs/ptclient.h> /* need definition of prlist for host.h */
110 #include "callback.h"
111 #ifdef AFS_DEMAND_ATTACH_FS
112 #include "serialize_state.h"
113 #endif /* AFS_DEMAND_ATTACH_FS */
116 extern afsUUID FS_HostUUID;
117 extern int hostCount;
119 #ifndef INTERPRET_DUMP
120 static int ShowProblems = 1;
123 struct cbcounters cbstuff;
125 static struct FileEntry * FE = NULL; /* don't use FE[0] */
126 static struct CallBack * CB = NULL; /* don't use CB[0] */
128 static struct CallBack * CBfree = NULL;
129 static struct FileEntry * FEfree = NULL;
132 /* Time to live for call backs depends upon number of users of the file.
133 * TimeOuts is indexed by this number/8 (using TimeOut macro). Times
134 * in this table are for the workstation; server timeouts, add
137 static int TimeOuts[] = {
138 /* Note: don't make the first entry larger than 4 hours (see above) */
139 4 * 60 * 60, /* 0-7 users */
140 1 * 60 * 60, /* 8-15 users */
141 30 * 60, /* 16-23 users */
142 15 * 60, /* 24-31 users */
143 15 * 60, /* 32-39 users */
144 10 * 60, /* 40-47 users */
145 10 * 60, /* 48-55 users */
146 10 * 60, /* 56-63 users */
147 }; /* Anything more: MinTimeOut */
149 /* minimum time given for a call back */
150 #ifndef INTERPRET_DUMP
151 static int MinTimeOut = (7 * 60);
154 /* Heads of CB queues; a timeout index is 1+index into this array */
155 static afs_uint32 timeout[CB_NUM_TIMEOUT_QUEUES];
157 static afs_int32 tfirst; /* cbtime of oldest unexpired call back time queue */
160 /* 16 byte object get/free routines */
165 /* Prototypes for static routines */
166 static struct FileEntry *FindFE(AFSFid * fid);
168 #ifndef INTERPRET_DUMP
169 static struct CallBack *iGetCB(int *nused);
170 static int iFreeCB(struct CallBack *cb, int *nused);
171 static struct FileEntry *iGetFE(int *nused);
172 static int iFreeFE(struct FileEntry *fe, int *nused);
173 static int TAdd(struct CallBack *cb, afs_uint32 * thead);
174 static int TDel(struct CallBack *cb);
175 static int HAdd(struct CallBack *cb, struct host *host);
176 static int HDel(struct CallBack *cb);
177 static int CDel(struct CallBack *cb, int deletefe);
178 static int CDelPtr(struct FileEntry *fe, afs_uint32 * cbp,
180 static afs_uint32 *FindCBPtr(struct FileEntry *fe, struct host *host);
181 static int FDel(struct FileEntry *fe);
182 static int AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead,
183 int type, int locked);
184 static void MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
185 struct AFSCBFids *afidp);
186 static int MultiBreakVolumeCallBack_r(struct host *host,
187 struct VCBParams *parms, int deletefe);
188 static int MultiBreakVolumeLaterCallBack(struct host *host, void *rock);
189 static int GetSomeSpace_r(struct host *hostp, int locked);
190 static int ClearHostCallbacks_r(struct host *hp, int locked);
191 static int DumpCallBackState_r(void);
194 #define GetCB() ((struct CallBack *)iGetCB(&cbstuff.nCBs))
195 #define GetFE() ((struct FileEntry *)iGetFE(&cbstuff.nFEs))
196 #define FreeCB(cb) iFreeCB((struct CallBack *)cb, &cbstuff.nCBs)
197 #define FreeFE(fe) iFreeFE((struct FileEntry *)fe, &cbstuff.nFEs)
200 /* Other protos - move out sometime */
201 void PrintCB(struct CallBack *cb, afs_uint32 now);
203 static afs_uint32 HashTable[FEHASH_SIZE]; /* File entry hash table */
205 static struct FileEntry *
210 struct FileEntry *fe;
212 hash = FEHash(fid->Volume, fid->Unique);
213 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
215 if (fe->volid == fid->Volume && fe->unique == fid->Unique
216 && fe->vnode == fid->Vnode && (fe->status & FE_LATER) != FE_LATER)
222 #ifndef INTERPRET_DUMP
224 static struct CallBack *
227 struct CallBack *ret;
229 if ((ret = CBfree)) {
230 CBfree = (struct CallBack *)(((struct object *)ret)->next);
237 iFreeCB(struct CallBack *cb, int *nused)
239 ((struct object *)cb)->next = (struct object *)CBfree;
245 static struct FileEntry *
248 struct FileEntry *ret;
250 if ((ret = FEfree)) {
251 FEfree = (struct FileEntry *)(((struct object *)ret)->next);
258 iFreeFE(struct FileEntry *fe, int *nused)
260 ((struct object *)fe)->next = (struct object *)FEfree;
266 /* Add cb to end of specified timeout list */
268 TAdd(struct CallBack *cb, afs_uint32 * thead)
271 (*thead) = cb->tnext = cb->tprev = cbtoi(cb);
273 struct CallBack *thp = itocb(*thead);
275 cb->tprev = thp->tprev;
279 thp->tprev = (itocb(thp->tprev)->tnext = cbtoi(cb));
281 thp->tprev = cbtoi(cb);
284 cb->thead = ttoi(thead);
288 /* Delete call back entry from timeout list */
290 TDel(struct CallBack *cb)
292 afs_uint32 *thead = itot(cb->thead);
294 if (*thead == cbtoi(cb))
295 *thead = (*thead == cb->tnext ? 0 : cb->tnext);
296 if (itocb(cb->tprev))
297 itocb(cb->tprev)->tnext = cb->tnext;
298 if (itocb(cb->tnext))
299 itocb(cb->tnext)->tprev = cb->tprev;
303 /* Add cb to end of specified host list */
305 HAdd(struct CallBack *cb, struct host *host)
307 cb->hhead = h_htoi(host);
309 host->cblist = cb->hnext = cb->hprev = cbtoi(cb);
311 struct CallBack *fcb = itocb(host->cblist);
313 cb->hprev = fcb->hprev;
314 cb->hnext = cbtoi(fcb);
315 fcb->hprev = (itocb(fcb->hprev)->hnext = cbtoi(cb));
320 /* Delete call back entry from host list */
322 HDel(struct CallBack *cb)
324 afs_uint32 *hhead = &h_itoh(cb->hhead)->cblist;
326 if (*hhead == cbtoi(cb))
327 *hhead = (*hhead == cb->hnext ? 0 : cb->hnext);
328 itocb(cb->hprev)->hnext = cb->hnext;
329 itocb(cb->hnext)->hprev = cb->hprev;
333 /* Delete call back entry from fid's chain of cb's */
334 /* N.B. This one also deletes the CB, and also possibly parent FE, so
335 * make sure that it is not on any other list before calling this
338 CDel(struct CallBack *cb, int deletefe)
341 struct FileEntry *fe = itofe(cb->fhead);
345 for (safety = 0, cbp = &fe->firstcb; *cbp && *cbp != cbi;
346 cbp = &itocb(*cbp)->cnext, safety++) {
347 if (safety > cbstuff.nblks + 10) {
348 ViceLogThenPanic(0, ("CDel: Internal Error -- shutting down: "
349 "wanted %d from %d, now at %d\n",
350 cbi, fe->firstcb, *cbp));
351 DumpCallBackState_r();
352 ShutDownAndCore(PANIC);
355 CDelPtr(fe, cbp, deletefe);
359 /* Same as CDel, but pointer to parent pointer to CB entry is passed,
360 * as well as file entry */
361 /* N.B. This one also deletes the CB, and also possibly parent FE, so
362 * make sure that it is not on any other list before calling this
364 static int Ccdelpt = 0, CcdelB = 0;
367 CDelPtr(struct FileEntry *fe, afs_uint32 * cbp,
380 if ((--fe->ncbs == 0) && deletefe)
386 FindCBPtr(struct FileEntry *fe, struct host *host)
388 afs_uint32 hostindex = h_htoi(host);
393 for (safety = 0, cbp = &fe->firstcb; *cbp; cbp = &cb->cnext, safety++) {
394 if (safety > cbstuff.nblks) {
395 ViceLog(0, ("FindCBPtr: Internal Error -- shutting down.\n"));
396 DumpCallBackState_r();
397 ShutDownAndCore(PANIC);
400 if (cb->hhead == hostindex)
406 /* Delete file entry from hash table */
408 FDel(struct FileEntry *fe)
411 afs_uint32 *p = &HashTable[FEHash(fe->volid, fe->unique)];
413 while (*p && *p != fei)
414 p = &itofe(*p)->fnext;
421 /* initialize the callback package */
423 InitCallBack(int nblks)
425 opr_Assert(nblks > 0);
428 tfirst = CBtime(time(NULL));
429 /* N.B. The "-1", below, is because
430 * FE[0] and CB[0] are not used--and not allocated */
431 FE = calloc(nblks, sizeof(struct FileEntry));
433 ViceLogThenPanic(0, ("Failed malloc in InitCallBack\n"));
435 FE--; /* FE[0] is supposed to point to junk */
436 cbstuff.nFEs = nblks;
438 FreeFE(&FE[cbstuff.nFEs]); /* This is correct */
439 CB = calloc(nblks, sizeof(struct CallBack));
441 ViceLogThenPanic(0, ("Failed malloc in InitCallBack\n"));
443 CB--; /* CB[0] is supposed to point to junk */
444 cbstuff.nCBs = nblks;
446 FreeCB(&CB[cbstuff.nCBs]); /* This is correct */
447 cbstuff.nblks = nblks;
448 cbstuff.nbreakers = 0;
454 XCallBackBulk_r(struct host * ahost, struct AFSFid * fids, afs_int32 nfids)
456 struct AFSCallBack tcbs[AFSCBMAX];
462 struct rx_connection *cb_conn = NULL;
464 rx_SetConnDeadTime(ahost->callback_rxcon, 4);
465 rx_SetConnHardDeadTime(ahost->callback_rxcon, AFS_HARDDEADTIME);
471 for (i = 0; i < nfids && i < AFSCBMAX; i++) {
472 tcbs[i].CallBackVersion = CALLBACK_VERSION;
473 tcbs[i].ExpirationTime = 0;
474 tcbs[i].CallBackType = CB_DROPPED;
476 tf.AFSCBFids_len = i;
477 tf.AFSCBFids_val = &(fids[j]);
481 tc.AFSCBs_val = tcbs;
483 cb_conn = ahost->callback_rxcon;
484 rx_GetConnection(cb_conn);
486 code |= RXAFSCB_CallBack(cb_conn, &tf, &tc);
487 rx_PutConnection(cb_conn);
495 /* the locked flag tells us if the host entry has already been locked
496 * by our parent. I don't think anybody actually calls us with the
497 * host locked, but here's how to make that work: GetSomeSpace has to
498 * change so that it doesn't attempt to lock any hosts < "host". That
499 * means that it might be unable to free any objects, so it has to
500 * return an exit status. If it fails, then AddCallBack1 might fail,
501 * as well. If so, the host->ResetDone should probably be set to 0,
502 * and we probably don't want to return a callback promise to the
503 * cache manager, either. */
505 AddCallBack1(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
513 if (!(host->hostFlags & HOSTDELETED))
514 retVal = AddCallBack1_r(host, fid, thead, type, 1);
524 AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
527 struct FileEntry *fe;
528 struct CallBack *cb = 0, *lastcb = 0;
529 struct FileEntry *newfe = 0;
530 afs_uint32 time_out = 0;
531 afs_uint32 *Thead = thead;
532 struct CallBack *newcb = 0;
535 cbstuff.AddCallBacks++;
539 /* allocate these guys first, since we can't call the allocator with
540 * the host structure locked -- or we might deadlock. However, we have
541 * to avoid races with FindFE... */
542 while (!(newcb = GetCB())) {
543 GetSomeSpace_r(host, locked);
545 while (!(newfe = GetFE())) { /* Get it now, so we don't have to call */
546 /* GetSomeSpace with the host locked, later. This might turn out to */
547 /* have been unneccessary, but that's actually kind of unlikely, since */
548 /* most files are not shared. */
549 GetSomeSpace_r(host, locked);
553 h_Lock_r(host); /* this can yield, so do it before we get any */
555 if (host->hostFlags & HOSTDELETED) {
563 if (type == CB_NORMAL) {
565 TimeCeiling(time(NULL) + TimeOut(fe ? fe->ncbs : 0) +
567 Thead = THead(CBtime(time_out));
568 } else if (type == CB_VOLUME) {
569 time_out = TimeCeiling((60 * 120 + time(NULL)) + ServerBias);
570 Thead = THead(CBtime(time_out));
571 } else if (type == CB_BULK) {
572 /* bulk status can get so many callbacks all at once, and most of them
573 * are probably not for things that will be used for long.
576 TimeCeiling(time(NULL) + ServerBias +
577 TimeOut(22 + (fe ? fe->ncbs : 0)));
578 Thead = THead(CBtime(time_out));
589 fe->volid = fid->Volume;
590 fe->vnode = fid->Vnode;
591 fe->unique = fid->Unique;
594 hash = FEHash(fid->Volume, fid->Unique);
595 fe->fnext = HashTable[hash];
596 HashTable[hash] = fetoi(fe);
598 for (safety = 0, lastcb = cb = itocb(fe->firstcb); cb;
599 lastcb = cb, cb = itocb(cb->cnext), safety++) {
600 if (safety > cbstuff.nblks) {
601 ViceLog(0, ("AddCallBack1: Internal Error -- shutting down.\n"));
602 DumpCallBackState_r();
603 ShutDownAndCore(PANIC);
605 if (cb->hhead == h_htoi(host))
608 if (cb) { /* Already have call back: move to new timeout list */
609 /* don't change delayed callbacks back to normal ones */
610 if (cb->status != CB_DELAYED)
612 /* Only move if new timeout is longer */
613 if (TNorm(ttoi(Thead)) > TNorm(cb->thead)) {
617 if (newfe == NULL) { /* we are using the new FE */
618 fe->firstcb = cbtoi(cb);
620 cb->fhead = fetoi(fe);
625 *(lastcb ? &lastcb->cnext : &fe->firstcb) = cbtoi(cb);
628 cb->fhead = fetoi(fe);
634 /* now free any still-unused callback or host entries */
640 if (!locked) /* freecb and freefe might(?) yield */
643 if (type == CB_NORMAL || type == CB_VOLUME || type == CB_BULK)
644 return time_out - ServerBias; /* Expires sooner at workstation */
650 CompareCBA(const void *e1, const void *e2)
652 const struct cbstruct *cba1 = (const struct cbstruct *)e1;
653 const struct cbstruct *cba2 = (const struct cbstruct *)e2;
654 return ((cba1->hp)->index - (cba2->hp)->index);
657 /* Take an array full of hosts, all held. Break callbacks to them, and
658 * release the holds once you're done.
659 * Currently only works for a single Fid in afidp array.
660 * If you want to make this work with multiple fids, you need to fix
661 * the error handling. One approach would be to force a reset if a
662 * multi-fid call fails, or you could add delayed callbacks for each
663 * fid. You probably also need to sort and remove duplicate hosts.
664 * When this is called from the BreakVolumeCallBacks path, it does NOT
665 * force a reset if the RPC fails, it just marks the host down and tries
666 * to create a delayed callback. */
667 /* N.B. be sure that code works when ncbas == 0 */
668 /* N.B. requires all the cba[*].hp pointers to be valid... */
669 /* This routine does not hold a lock on the host for the duration of
670 * the BreakCallBack RPC, which is a significant deviation from tradition.
671 * It _does_ get a lock on the host before setting VenusDown = 1,
672 * which is sufficient only if VenusDown = 0 only happens when the
673 * lock is held over the RPC and the subsequent VenusDown == 0
674 * wherever that is done. */
676 MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
677 struct AFSCBFids *afidp)
680 struct rx_connection *conns[MAX_CB_HOSTS];
681 static struct AFSCBs tc = { 0, 0 };
682 int multi_to_cba_map[MAX_CB_HOSTS];
684 opr_Assert(ncbas <= MAX_CB_HOSTS);
687 * When we issue a multi_Rx callback break, we must rx_NewCall a call for
688 * each host before we do anything. If there are no call channels
689 * available on the conn, we must wait for one of the existing calls to
690 * finish. If another thread is breaking callbacks at the same time, it is
691 * possible for us to be waiting on NewCall for one of their multi_Rx
692 * CallBack calls to finish, but they are waiting on NewCall for one of
693 * our calls to finish. So we deadlock.
695 * This can be thought of as similar to obtaining multiple locks at the
696 * same time. So if we establish an ordering, the possibility of deadlock
697 * goes away. Here we provide such an ordering, by sorting our CBAs
698 * according to CompareCBA.
700 qsort(cba, ncbas, sizeof(struct cbstruct), CompareCBA);
702 /* set up conns for multi-call */
703 for (i = 0, j = 0; i < ncbas; i++) {
704 struct host *thishost = cba[i].hp;
705 if (!thishost || (thishost->hostFlags & HOSTDELETED)) {
708 rx_GetConnection(thishost->callback_rxcon);
709 multi_to_cba_map[j] = i;
710 conns[j++] = thishost->callback_rxcon;
712 rx_SetConnDeadTime(thishost->callback_rxcon, 4);
713 rx_SetConnHardDeadTime(thishost->callback_rxcon, AFS_HARDDEADTIME);
716 if (j) { /* who knows what multi would do with 0 conns? */
720 multi_RXAFSCB_CallBack(afidp, &tc);
726 i = multi_to_cba_map[multi_i];
732 ("BCB: INTERNAL ERROR: hp=%p, cba=%p, thead=%u\n",
736 ** try breaking callbacks on alternate interface addresses
738 if (MultiBreakCallBackAlternateAddress(hp, afidp)) {
741 ("BCB: Failed on file %u.%u.%u, "
742 "Host %p (%s:%d) is down\n",
743 afidp->AFSCBFids_val->Volume,
744 afidp->AFSCBFids_val->Vnode,
745 afidp->AFSCBFids_val->Unique,
747 afs_inet_ntoa_r(hp->host, hoststr),
753 if (!(hp->hostFlags & HOSTDELETED)) {
754 hp->hostFlags |= VENUSDOWN;
756 * We always go into AddCallBack1_r with the host locked
758 AddCallBack1_r(hp, afidp->AFSCBFids_val, itot(idx),
772 for (i = 0; i < ncbas; i++) {
780 /* H_UNLOCK around this so h_FreeConnection does not deadlock.
781 h_FreeConnection should *never* be called on a callback connection,
782 but on 10/27/04 a deadlock occurred where it was, when we know why,
783 this should be reverted. -- shadow */
785 for (i = 0; i < j; i++) {
786 rx_PutConnection(conns[i]);
794 * Break all call backs for fid, except for the specified host (unless flag
795 * is true, in which case all get a callback message. Assumption: the specified
796 * host is h_Held, by the caller; the others aren't.
797 * Specified host may be bogus, that's ok. This used to check to see if the
798 * host was down in two places, once right after the host was h_held, and
799 * again after it was locked. That race condition is incredibly rare and
800 * relatively harmless even when it does occur, so we don't check for it now.
802 /* if flag is true, send a break callback msg to "host", too */
804 BreakCallBack(struct host *xhost, AFSFid * fid, int flag)
806 struct FileEntry *fe;
807 struct CallBack *cb, *nextcb;
808 struct cbstruct cba[MAX_CB_HOSTS];
816 ("BCB: BreakCallBack(Host %p all but %s:%d, (%u,%u,%u))\n",
817 xhost, afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
818 fid->Volume, fid->Vnode, fid->Unique));
821 ("BCB: BreakCallBack(No Host, (%u,%u,%u))\n",
822 fid->Volume, fid->Vnode, fid->Unique));
825 cbstuff.BreakCallBacks++;
830 hostindex = xhost ? h_htoi(xhost) : 0;
831 cb = itocb(fe->firstcb);
832 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
833 /* the most common case is what follows the || */
836 tf.AFSCBFids_len = 1;
837 tf.AFSCBFids_val = fid;
840 for (ncbas = 0; cb && ncbas < MAX_CB_HOSTS; cb = nextcb) {
841 nextcb = itocb(cb->cnext);
842 if ((cb->hhead != hostindex || flag)
843 && (cb->status == CB_BULK || cb->status == CB_NORMAL
844 || cb->status == CB_VOLUME)) {
845 struct host *thishost = h_itoh(cb->hhead);
847 ViceLog(0, ("BCB: BOGUS! cb->hhead is NULL!\n"));
848 } else if (thishost->hostFlags & VENUSDOWN) {
850 ("BCB: %p (%s:%d) is down; delaying break call back\n",
851 thishost, afs_inet_ntoa_r(thishost->host, hoststr),
852 ntohs(thishost->port)));
853 cb->status = CB_DELAYED;
855 if (!(thishost->hostFlags & HOSTDELETED)) {
857 cba[ncbas].hp = thishost;
858 cba[ncbas].thead = cb->thead;
863 CDel(cb, 1); /* Usually first; so this delete
864 * is reasonably inexpensive */
870 MultiBreakCallBack_r(cba, ncbas, &tf);
872 /* we need to to all these initializations again because MultiBreakCallBack may block */
877 cb = itocb(fe->firstcb);
878 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
879 /* the most common case is what follows the || */
890 /* Delete (do not break) single call back for fid */
892 DeleteCallBack(struct host *host, AFSFid * fid)
894 struct FileEntry *fe;
899 cbstuff.DeleteCallBacks++;
902 /* do not care if the host has been HOSTDELETED */
908 ("DCB: No call backs for fid (%u, %u, %u)\n", fid->Volume,
909 fid->Vnode, fid->Unique));
912 pcb = FindCBPtr(fe, host);
915 ("DCB: No call back for host %p (%s:%d), (%u, %u, %u)\n",
916 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
917 fid->Volume, fid->Vnode, fid->Unique));
931 * Delete (do not break) all call backs for fid. This call doesn't
932 * set all of the various host locks, but it shouldn't really matter
933 * since we're not adding callbacks, but deleting them. I'm not sure
934 * why it doesn't set the lock, however; perhaps it should.
937 DeleteFileCallBacks(AFSFid * fid)
939 struct FileEntry *fe;
945 cbstuff.DeleteFiles++;
950 ("DF: No fid (%u,%u,%u) to delete\n", fid->Volume, fid->Vnode,
954 for (n = 0, cbi = fe->firstcb; cbi; n++) {
967 /* Delete (do not break) all call backs for host. The host should be
970 DeleteAllCallBacks_r(struct host *host, int deletefe)
975 cbstuff.DeleteAllCallBacks++;
976 cbi = first = host->cblist;
978 ViceLog(8, ("DV: no call backs\n"));
986 } while (cbi != first);
992 * Break all delayed call backs for host. Returns 1 if all call backs
993 * successfully broken; 0 otherwise. Assumes host is h_Held and h_Locked.
994 * Must be called with VenusDown set for this host
997 BreakDelayedCallBacks(struct host *host)
1001 retVal = BreakDelayedCallBacks_r(host);
1007 BreakDelayedCallBacks_r(struct host *host)
1009 struct AFSFid fids[AFSCBMAX];
1010 int cbi, first, nfids;
1011 struct CallBack *cb;
1014 struct rx_connection *cb_conn;
1016 cbstuff.nbreakers++;
1017 if (!(host->hostFlags & RESETDONE) && !(host->hostFlags & HOSTDELETED)) {
1018 host->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1019 cb_conn = host->callback_rxcon;
1020 rx_GetConnection(cb_conn);
1021 if (host->interface) {
1024 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1027 code = RXAFSCB_InitCallBackState(cb_conn);
1029 rx_PutConnection(cb_conn);
1032 host->hostFlags |= ALTADDR; /* alternate addresses are valid */
1036 ("CB: Call back connect back failed (in break delayed) "
1037 "for Host %p (%s:%d)\n",
1038 host, afs_inet_ntoa_r(host->host, hoststr),
1039 ntohs(host->port)));
1041 host->hostFlags |= VENUSDOWN;
1044 ("InitCallBackState success on %p (%s:%d)\n",
1045 host, afs_inet_ntoa_r(host->host, hoststr),
1046 ntohs(host->port)));
1047 /* reset was done successfully */
1048 host->hostFlags |= RESETDONE;
1049 host->hostFlags &= ~VENUSDOWN;
1052 while (!(host->hostFlags & HOSTDELETED)) {
1054 host->hostFlags &= ~VENUSDOWN; /* presume up */
1055 cbi = first = host->cblist;
1059 first = host->cblist;
1062 if (cb->status == CB_DELAYED) {
1063 struct FileEntry *fe = itofe(cb->fhead);
1064 fids[nfids].Volume = fe->volid;
1065 fids[nfids].Vnode = fe->vnode;
1066 fids[nfids].Unique = fe->unique;
1072 } while (cbi && cbi != first && nfids < AFSCBMAX);
1078 if (XCallBackBulk_r(host, fids, nfids)) {
1079 /* Failed, again: put them back, probably with old
1084 ("CB: XCallBackBulk failed, Host %p (%s:%d); "
1085 "callback list follows:\n",
1086 host, afs_inet_ntoa_r(host->host, hoststr),
1087 ntohs(host->port)));
1089 for (i = 0; i < nfids; i++) {
1092 ("CB: Host %p (%s:%d), file %u.%u.%u "
1093 "(part of bulk callback)\n",
1094 host, afs_inet_ntoa_r(host->host, hoststr),
1095 ntohs(host->port), fids[i].Volume,
1096 fids[i].Vnode, fids[i].Unique));
1099 * AddCallBack1_r(host, &fids[i], itot(thead[i]), CB_DELAYED, 1);
1100 * * but it turns out to cause too many tricky locking problems.
1101 * * now, if break delayed fails, screw it. */
1103 host->hostFlags |= VENUSDOWN; /* Failed */
1104 ClearHostCallbacks_r(host, 1 /* locked */ );
1108 if (nfids < AFSCBMAX)
1112 cbstuff.nbreakers--;
1113 /* If we succeeded it's always ok to unset HFE_LATER */
1114 if (!(host->hostFlags & VENUSDOWN))
1115 host->hostFlags &= ~HFE_LATER;
1116 return (host->hostFlags & VENUSDOWN);
1120 MultiBreakVolumeCallBack_r(struct host *host,
1121 struct VCBParams *parms, int deletefe)
1125 if (host->hostFlags & HOSTDELETED)
1128 if (!(host->hostFlags & HCBREAK))
1129 return 0; /* host is not flagged to notify */
1131 if (host->hostFlags & VENUSDOWN) {
1133 /* Do not care if the host is now HOSTDELETED */
1136 ("BVCB: volume callback for Host %p (%s:%d) failed\n",
1137 host, afs_inet_ntoa_r(host->host, hoststr),
1138 ntohs(host->port)));
1140 DeleteAllCallBacks_r(host, deletefe); /* Delete all callback state
1141 * rather than attempting to
1142 * selectively remember to
1143 * delete the volume callbacks
1145 host->hostFlags &= ~(RESETDONE|HCBREAK); /* Do InitCallBackState when host returns */
1149 opr_Assert(parms->ncbas <= MAX_CB_HOSTS);
1151 /* Do not call MultiBreakCallBack on the current host structure
1152 ** because it would prematurely release the hold on the host
1154 if (parms->ncbas == MAX_CB_HOSTS) {
1155 struct AFSCBFids tf;
1157 tf.AFSCBFids_len = 1;
1158 tf.AFSCBFids_val = parms->fid;
1160 /* this releases all the hosts */
1161 MultiBreakCallBack_r(parms->cba, parms->ncbas, &tf);
1165 parms->cba[parms->ncbas].hp = host;
1166 parms->cba[(parms->ncbas)++].thead = parms->thead;
1167 host->hostFlags &= ~HCBREAK;
1169 /* we have more work to do on this host, so make sure we keep a reference
1177 MultiBreakVolumeLaterCallBack(struct host *host, void *rock)
1179 struct VCBParams *parms = (struct VCBParams *)rock;
1182 retval = MultiBreakVolumeCallBack_r(host, parms, 0);
1188 * Break all call backs on a single volume. Don't call this with any
1189 * hosts h_held. Note that this routine clears the callbacks before
1190 * actually breaking them, and that the vnode isn't locked during this
1191 * operation, so that people might see temporary callback loss while
1192 * this function is executing. It is just a temporary state, however,
1193 * since the callback will be broken later by this same function.
1195 * Now uses multi-RX for CallBack RPC in a different thread,
1196 * only marking them here.
1198 extern pthread_cond_t fsync_cond;
1201 BreakVolumeCallBacksLater(VolumeId volume)
1205 struct FileEntry *fe;
1206 struct CallBack *cb;
1210 ViceLog(25, ("Setting later on volume %" AFS_VOLID_FMT "\n",
1211 afs_printable_VolumeId_lu(volume)));
1213 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1214 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1215 if (fe->volid == volume) {
1216 struct CallBack *cbnext;
1217 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1218 host = h_itoh(cb->hhead);
1219 host->hostFlags |= HFE_LATER;
1220 cb->status = CB_DELAYED;
1221 cbnext = itocb(cb->cnext);
1224 fe->status |= FE_LATER;
1233 /* didn't find any callbacks, so return right away. */
1237 ViceLog(25, ("Fsync thread wakeup\n"));
1239 opr_cv_broadcast(&fsync_cond);
1245 BreakLaterCallBacks(void)
1250 struct CallBack *cb;
1251 struct FileEntry *fe = NULL;
1252 struct FileEntry *myfe = NULL;
1254 struct VCBParams henumParms;
1255 unsigned short tthead = 0; /* zero is illegal value */
1259 ViceLog(25, ("Looking for FileEntries to unchain\n"));
1262 /* Pick the first volume we see to clean up */
1263 fid.Volume = fid.Vnode = fid.Unique = 0;
1265 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1266 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1267 if (fe && (fe->status & FE_LATER)
1268 && (fid.Volume == 0 || fid.Volume == fe->volid)) {
1269 /* Ugly, but used to avoid left side casting */
1270 struct object *tmpfe;
1272 ("Unchaining for %u:%u:%" AFS_VOLID_FMT "\n", fe->vnode,
1273 fe->unique, afs_printable_VolumeId_lu(fe->volid)));
1274 fid.Volume = fe->volid;
1276 fe->status &= ~FE_LATER; /* not strictly needed */
1277 /* Works since volid is deeper than the largest pointer */
1278 tmpfe = (struct object *)fe;
1279 tmpfe->next = (struct object *)myfe;
1292 /* loop over FEs from myfe and free/break */
1294 for (fe = myfe; fe;) {
1295 struct CallBack *cbnext;
1296 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1297 cbnext = itocb(cb->cnext);
1298 host = h_itoh(cb->hhead);
1299 if (cb->status == CB_DELAYED) {
1300 if (!(host->hostFlags & HOSTDELETED)) {
1301 /* mark this host for notification */
1302 host->hostFlags |= HCBREAK;
1303 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1309 CDel(cb, 0); /* Don't let CDel clean up the fe */
1310 /* leave flag for MultiBreakVolumeCallBack to clear */
1313 ("Found host %p (%s:%d) non-DELAYED cb for %u:%u:%" AFS_VOLID_FMT "\n",
1314 host, afs_inet_ntoa_r(host->host, hoststr),
1315 ntohs(host->port), fe->vnode, fe->unique,
1316 afs_printable_VolumeId_lu(fe->volid)));
1320 fe = (struct FileEntry *)((struct object *)fe)->next;
1325 ViceLog(125, ("Breaking volume %u\n", fid.Volume));
1326 henumParms.ncbas = 0;
1327 henumParms.fid = &fid;
1328 henumParms.thead = tthead;
1330 h_Enumerate(MultiBreakVolumeLaterCallBack, (char *)&henumParms);
1332 if (henumParms.ncbas) { /* do left-overs */
1333 struct AFSCBFids tf;
1334 tf.AFSCBFids_len = 1;
1335 tf.AFSCBFids_val = &fid;
1337 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf);
1338 henumParms.ncbas = 0;
1343 /* Arrange to be called again */
1348 * Delete all timed-out call back entries (to be called periodically by file
1352 CleanupTimedOutCallBacks(void)
1355 CleanupTimedOutCallBacks_r();
1361 CleanupTimedOutCallBacks_r(void)
1363 afs_uint32 now = CBtime(time(NULL));
1365 struct CallBack *cb;
1369 while (tfirst <= now) {
1371 cbi = *(thead = THead(tfirst));
1377 ("CCB: deleting timed out call back %x (%s:%d), (%" AFS_VOLID_FMT ",%u,%u)\n",
1378 h_itoh(cb->hhead)->host,
1379 afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
1380 h_itoh(cb->hhead)->port,
1381 afs_printable_VolumeId_lu(itofe(cb->fhead)->volid),
1382 itofe(cb->fhead)->vnode, itofe(cb->fhead)->unique));
1386 if (ntimedout > cbstuff.nblks) {
1387 ViceLog(0, ("CCB: Internal Error -- shutting down...\n"));
1388 DumpCallBackState_r();
1389 ShutDownAndCore(PANIC);
1391 } while (cbi != *thead);
1396 cbstuff.CBsTimedOut += ntimedout;
1397 ViceLog(7, ("CCB: deleted %d timed out callbacks\n", ntimedout));
1398 return (ntimedout > 0);
1402 * parameters to pass to lih*_r from h_Enumerate_r when trying to find a host
1403 * from which to clear callbacks.
1407 * Points to the least interesting host found; try to clear callbacks on
1408 * this host after h_Enumerate_r(lih*_r)'ing.
1413 * The last host we got from lih*_r, but we couldn't clear its callbacks
1414 * for some reason. Choose the next-best host after this one (with the
1415 * current lih*_r, this means to only select hosts that have an ActiveCall
1416 * newer than lastlih).
1418 struct host *lastlih;
1421 /* Value of host->refCount that allows us to reliably infer that
1422 * host may be held by some other thread */
1423 #define OTHER_MUSTHOLD_LIH 2
1425 /* This version does not allow 'host' to be selected unless its ActiveCall
1426 * is newer than 'params->lastlih' which is the host with the oldest
1427 * ActiveCall from the last pass (if it is provided). We filter out any hosts
1428 * that are are held by other threads.
1430 * There is a small problem here, but it may not be easily fixable. Say we
1431 * select some host A, and give it back to GetSomeSpace_r. GSS_r for some
1432 * reason cannot clear the callbacks on A, and so calls us again with
1433 * lastlih = A. Suppose there is another host B that has the same ActiveCall
1434 * time as A. We will now skip over host B, since
1435 * 'hostB->ActiveCall > hostA->ActiveCall' is not true. This could result in
1436 * us prematurely going to the GSS_r 2nd or 3rd pass, and making us a little
1437 * inefficient. This should be pretty rare, though, except perhaps in cases
1438 * with very small numbers of hosts.
1440 * Also filter out any hosts with HOSTDELETED set. h_Enumerate_r should in
1441 * theory not give these to us anyway, but be paranoid.
1444 lih0_r(struct host *host, void *rock)
1446 struct lih_params *params = (struct lih_params *)rock;
1448 /* OTHER_MUSTHOLD_LIH is because the h_Enum loop holds us once */
1450 && (!(host->hostFlags & HOSTDELETED))
1451 && (host->refCount < OTHER_MUSTHOLD_LIH)
1452 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1453 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1456 h_Release_r(params->lih); /* release prev host */
1465 /* same as lih0_r, except we do not prevent held hosts from being selected. */
1467 lih1_r(struct host *host, void *rock)
1469 struct lih_params *params = (struct lih_params *)rock;
1472 && (!(host->hostFlags & HOSTDELETED))
1473 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1474 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1477 h_Release_r(params->lih); /* release prev host */
1486 /* This could be upgraded to get more space each time */
1487 /* first pass: sequentially find the oldest host which isn't held by
1488 anyone for which we can clear callbacks;
1490 /* second pass: sequentially find the oldest host regardless of
1491 whether or not the host is held; skipping 'hostp' */
1492 /* third pass: attempt to clear callbacks from 'hostp' */
1493 /* always called with hostp unlocked */
1495 /* Note: hostlist is ordered most recently created host first and
1496 * its order has no relationship to the most recently used. */
1497 extern struct host *hostList;
1499 GetSomeSpace_r(struct host *hostp, int locked)
1502 struct lih_params params;
1505 if (cbstuff.GotSomeSpaces == 0) {
1506 /* only log this once; if GSS is getting called constantly, that's not
1507 * good but don't make things worse by spamming the log. */
1508 ViceLog(0, ("We have run out of callback space; forcing callback revocation. "
1509 "This suggests the fileserver is configured with insufficient "
1510 "callbacks; you probably want to increase the -cb fileserver "
1511 "parameter (current setting: %u). The fileserver will continue "
1512 "to operate, but this may indicate a severe performance problem\n",
1514 ViceLog(0, ("This message is logged at most once; for more information "
1515 "see the OpenAFS documentation and fileserver xstat collection 3\n"));
1518 cbstuff.GotSomeSpaces++;
1520 ("GSS: First looking for timed out call backs via CleanupCallBacks\n"));
1521 if (CleanupTimedOutCallBacks_r()) {
1527 params.lastlih = NULL;
1532 h_Enumerate_r(i == 0 ? lih0_r : lih1_r, hostList, ¶ms);
1535 if (params.lastlih) {
1536 h_Release_r(params.lastlih);
1537 params.lastlih = NULL;
1541 /* note that 'hp' was held by lih*_r; we will need to release it */
1543 if ((hp != hostp) && !ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
1548 params.lastlih = hp;
1549 /* params.lastlih will be released on the next iteration, after
1554 * Next time try getting callbacks from any host even if
1555 * it's held, since the only other option is starvation for
1556 * the file server (i.e. until the callback timeout arrives).
1559 params.lastlih = NULL;
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)
1585 struct rx_connection *cb_conn = NULL;
1588 ("GSS: Delete longest inactive host %p (%s:%d)\n",
1589 hp, afs_inet_ntoa_r(hp->host, hoststr), ntohs(hp->port)));
1591 if ((hp->hostFlags & HOSTDELETED)) {
1592 /* hp could go away after reacquiring H_LOCK in h_NBLock_r, so we can't
1593 * really use it; its callbacks will get cleared anyway when
1594 * h_TossStuff_r gets its hands on it */
1600 /** Try a non-blocking lock. If the lock is already held return
1601 * after releasing hold on hp
1604 if (h_NBLock_r(hp)) {
1609 if (hp->Console & 2) {
1611 * If the special console field is set it means that a thread
1612 * is waiting in AddCallBack1 after it set pointers to the
1613 * file entry and/or callback entry. Because of the bogus
1614 * usage of h_hold it won't prevent from another thread, this
1615 * one, to remove all the callbacks so just to be safe we keep
1616 * a reference. NOTE, on the last phase we'll free the calling
1617 * host's callbacks but that's ok...
1621 DeleteAllCallBacks_r(hp, 1);
1622 if (hp->hostFlags & VENUSDOWN) {
1623 hp->hostFlags &= ~RESETDONE; /* remember that we must do a reset */
1624 } else if (!(hp->hostFlags & HOSTDELETED)) {
1625 /* host is up, try a call */
1626 hp->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1627 cb_conn = hp->callback_rxcon;
1628 rx_GetConnection(hp->callback_rxcon);
1629 if (hp->interface) {
1632 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1635 code = RXAFSCB_InitCallBackState(cb_conn);
1637 rx_PutConnection(cb_conn);
1640 hp->hostFlags |= ALTADDR; /* alternate addresses are valid */
1642 /* failed, mark host down and need reset */
1643 hp->hostFlags |= VENUSDOWN;
1644 hp->hostFlags &= ~RESETDONE;
1646 /* reset succeeded, we're done */
1647 hp->hostFlags |= RESETDONE;
1656 #endif /* INTERPRET_DUMP */
1660 PrintCallBackStats(void)
1663 "%d add CB, %d break CB, %d del CB, %d del FE, %d CB's timed out, %d space reclaim, %d del host\n",
1664 cbstuff.AddCallBacks, cbstuff.BreakCallBacks,
1665 cbstuff.DeleteCallBacks, cbstuff.DeleteFiles, cbstuff.CBsTimedOut,
1666 cbstuff.GotSomeSpaces, cbstuff.DeleteAllCallBacks);
1667 fprintf(stderr, "%d CBs, %d FEs, (%d of total of %d 16-byte blocks)\n",
1668 cbstuff.nCBs, cbstuff.nFEs, cbstuff.nCBs + cbstuff.nFEs,
1670 fprintf(stderr, "%d GSS1, %d GSS2, %d GSS3, %d GSS4, %d GSS5 (internal counters)\n",
1671 cbstuff.GSS1, cbstuff.GSS2, cbstuff.GSS3, cbstuff.GSS4, cbstuff.GSS5);
1676 #define MAGIC 0x12345678 /* To check byte ordering of dump when it is read in */
1677 #define MAGICV2 0x12345679 /* To check byte ordering & version of dump when it is read in */
1680 #ifndef INTERPRET_DUMP
1682 #ifdef AFS_DEMAND_ATTACH_FS
1685 * callback state serialization
1687 static int cb_stateSaveTimeouts(struct fs_dump_state * state);
1688 static int cb_stateSaveFEHash(struct fs_dump_state * state);
1689 static int cb_stateSaveFEs(struct fs_dump_state * state);
1690 static int cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe);
1691 static int cb_stateRestoreTimeouts(struct fs_dump_state * state);
1692 static int cb_stateRestoreFEHash(struct fs_dump_state * state);
1693 static int cb_stateRestoreFEs(struct fs_dump_state * state);
1694 static int cb_stateRestoreFE(struct fs_dump_state * state);
1695 static int cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
1696 struct iovec * iov, int niovecs);
1698 static int cb_stateVerifyFEHash(struct fs_dump_state * state);
1699 static int cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe);
1700 static int cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe);
1701 static int cb_stateVerifyTimeoutQueues(struct fs_dump_state * state);
1703 static int cb_stateFEToDiskEntry(struct FileEntry *, struct FEDiskEntry *);
1704 static int cb_stateDiskEntryToFE(struct fs_dump_state * state,
1705 struct FEDiskEntry *, struct FileEntry *);
1707 static int cb_stateCBToDiskEntry(struct CallBack *, struct CBDiskEntry *);
1708 static int cb_stateDiskEntryToCB(struct fs_dump_state * state,
1709 struct CBDiskEntry *, struct CallBack *);
1711 static int cb_stateFillHeader(struct callback_state_header * hdr);
1712 static int cb_stateCheckHeader(struct callback_state_header * hdr);
1714 static int cb_stateAllocMap(struct fs_dump_state * state);
1717 cb_stateSave(struct fs_dump_state * state)
1721 AssignInt64(state->eof_offset, &state->hdr->cb_offset);
1723 /* invalidate callback state header */
1724 memset(state->cb_hdr, 0, sizeof(struct callback_state_header));
1725 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1726 sizeof(struct callback_state_header))) {
1731 fs_stateIncEOF(state, sizeof(struct callback_state_header));
1733 /* dump timeout state */
1734 if (cb_stateSaveTimeouts(state)) {
1739 /* dump fe hashtable state */
1740 if (cb_stateSaveFEHash(state)) {
1745 /* dump callback state */
1746 if (cb_stateSaveFEs(state)) {
1751 /* write the callback state header to disk */
1752 cb_stateFillHeader(state->cb_hdr);
1753 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1754 sizeof(struct callback_state_header))) {
1764 cb_stateRestore(struct fs_dump_state * state)
1768 if (fs_stateReadHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1769 sizeof(struct callback_state_header))) {
1774 if (cb_stateCheckHeader(state->cb_hdr)) {
1779 if (cb_stateAllocMap(state)) {
1784 if (cb_stateRestoreTimeouts(state)) {
1789 if (cb_stateRestoreFEHash(state)) {
1794 /* restore FEs and CBs from disk */
1795 if (cb_stateRestoreFEs(state)) {
1800 /* restore the timeout queue heads */
1801 tfirst = state->cb_hdr->tfirst;
1808 cb_stateRestoreIndices(struct fs_dump_state * state)
1811 struct FileEntry * fe;
1812 struct CallBack * cb;
1814 /* restore indices in the FileEntry structures */
1815 for (i = 1; i < state->fe_map.len; i++) {
1816 if (state->fe_map.entries[i].new_idx) {
1817 fe = itofe(state->fe_map.entries[i].new_idx);
1819 /* restore the fe->fnext entry */
1820 if (fe_OldToNew(state, fe->fnext, &fe->fnext)) {
1825 /* restore the fe->firstcb entry */
1826 if (cb_OldToNew(state, fe->firstcb, &fe->firstcb)) {
1833 /* restore indices in the CallBack structures */
1834 for (i = 1; i < state->cb_map.len; i++) {
1835 if (state->cb_map.entries[i].new_idx) {
1836 cb = itocb(state->cb_map.entries[i].new_idx);
1838 /* restore the cb->cnext entry */
1839 if (cb_OldToNew(state, cb->cnext, &cb->cnext)) {
1844 /* restore the cb->fhead entry */
1845 if (fe_OldToNew(state, cb->fhead, &cb->fhead)) {
1850 /* restore the cb->hhead entry */
1851 if (h_OldToNew(state, cb->hhead, &cb->hhead)) {
1856 /* restore the cb->tprev entry */
1857 if (cb_OldToNew(state, cb->tprev, &cb->tprev)) {
1862 /* restore the cb->tnext entry */
1863 if (cb_OldToNew(state, cb->tnext, &cb->tnext)) {
1868 /* restore the cb->hprev entry */
1869 if (cb_OldToNew(state, cb->hprev, &cb->hprev)) {
1874 /* restore the cb->hnext entry */
1875 if (cb_OldToNew(state, cb->hnext, &cb->hnext)) {
1882 /* restore the timeout queue head indices */
1883 for (i = 0; i < state->cb_timeout_hdr->records; i++) {
1884 if (cb_OldToNew(state, timeout[i], &timeout[i])) {
1890 /* restore the FE hash table queue heads */
1891 for (i = 0; i < state->cb_fehash_hdr->records; i++) {
1892 if (fe_OldToNew(state, HashTable[i], &HashTable[i])) {
1903 cb_stateVerify(struct fs_dump_state * state)
1907 if (cb_stateVerifyFEHash(state)) {
1911 if (cb_stateVerifyTimeoutQueues(state)) {
1919 cb_stateVerifyFEHash(struct fs_dump_state * state)
1922 struct FileEntry * fe;
1923 afs_uint32 fei, chain_len;
1925 for (i = 0; i < FEHASH_SIZE; i++) {
1927 for (fei = HashTable[i], fe = itofe(fei);
1929 fei = fe->fnext, fe = itofe(fei)) {
1930 if (fei > cbstuff.nblks) {
1931 ViceLog(0, ("cb_stateVerifyFEHash: error: index out of range (fei=%d)\n", fei));
1935 if (cb_stateVerifyFE(state, fe)) {
1938 if (chain_len > FS_STATE_FE_MAX_HASH_CHAIN_LEN) {
1939 ViceLog(0, ("cb_stateVerifyFEHash: error: hash chain %d length exceeds %d; assuming there's a loop\n",
1940 i, FS_STATE_FE_MAX_HASH_CHAIN_LEN));
1952 cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe)
1956 if ((fe->firstcb && !fe->ncbs) ||
1957 (!fe->firstcb && fe->ncbs)) {
1958 ViceLog(0, ("cb_stateVerifyFE: error: fe->firstcb does not agree with fe->ncbs (fei=%lu, fe->firstcb=%lu, fe->ncbs=%lu)\n",
1959 afs_printable_uint32_lu(fetoi(fe)),
1960 afs_printable_uint32_lu(fe->firstcb),
1961 afs_printable_uint32_lu(fe->ncbs)));
1964 if (cb_stateVerifyFCBList(state, fe)) {
1965 ViceLog(0, ("cb_stateVerifyFE: error: FCBList failed verification (fei=%lu)\n",
1966 afs_printable_uint32_lu(fetoi(fe))));
1974 cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe)
1977 afs_uint32 cbi, fei, chain_len = 0;
1978 struct CallBack * cb;
1982 for (cbi = fe->firstcb, cb = itocb(cbi);
1984 cbi = cb->cnext, cb = itocb(cbi)) {
1985 if (cbi > cbstuff.nblks) {
1986 ViceLog(0, ("cb_stateVerifyFCBList: error: list index out of range (cbi=%d, ncbs=%d)\n",
1987 cbi, cbstuff.nblks));
1991 if (cb->fhead != fei) {
1992 ViceLog(0, ("cb_stateVerifyFCBList: error: cb->fhead != fei (fei=%d, cb->fhead=%d)\n",
1996 if (chain_len > FS_STATE_FCB_MAX_LIST_LEN) {
1997 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (fei=%d); assuming there's a loop\n",
1998 FS_STATE_FCB_MAX_LIST_LEN, fei));
2005 if (fe->ncbs != chain_len) {
2006 ViceLog(0, ("cb_stateVerifyFCBList: error: list length mismatch (len=%d, fe->ncbs=%d)\n",
2007 chain_len, fe->ncbs));
2016 cb_stateVerifyHCBList(struct fs_dump_state * state, struct host * host)
2019 afs_uint32 hi, chain_len, cbi;
2020 struct CallBack *cb, *ncb;
2025 for (cbi = host->cblist, cb = itocb(cbi);
2027 cbi = cb->hnext, cb = ncb) {
2028 if (chain_len && (host->cblist == cbi)) {
2029 /* we've wrapped around the circular list, and everything looks ok */
2032 if (cb->hhead != hi) {
2033 ViceLog(0, ("cb_stateVerifyHCBList: error: incorrect cb->hhead (cbi=%d, h->index=%d, cb->hhead=%d)\n",
2034 cbi, hi, cb->hhead));
2037 if (!cb->hprev || !cb->hnext) {
2038 ViceLog(0, ("cb_stateVerifyHCBList: error: null index in circular list (cbi=%d, h->index=%d)\n",
2043 if ((cb->hprev > cbstuff.nblks) ||
2044 (cb->hnext > cbstuff.nblks)) {
2045 ViceLog(0, ("cb_stateVerifyHCBList: error: list index out of range (cbi=%d, h->index=%d, cb->hprev=%d, cb->hnext=%d, nCBs=%d)\n",
2046 cbi, hi, cb->hprev, cb->hnext, cbstuff.nblks));
2050 ncb = itocb(cb->hnext);
2051 if (cbi != ncb->hprev) {
2052 ViceLog(0, ("cb_stateVerifyHCBList: error: corrupt linked list (cbi=%d, h->index=%d)\n",
2057 if (chain_len > FS_STATE_HCB_MAX_LIST_LEN) {
2058 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (h->index=%d); assuming there's a loop\n",
2059 FS_STATE_HCB_MAX_LIST_LEN, hi));
2071 cb_stateVerifyTimeoutQueues(struct fs_dump_state * state)
2074 afs_uint32 cbi, chain_len;
2075 struct CallBack *cb, *ncb;
2077 for (i = 0; i < CB_NUM_TIMEOUT_QUEUES; i++) {
2079 for (cbi = timeout[i], cb = itocb(cbi);
2081 cbi = cb->tnext, cb = ncb) {
2082 if (chain_len && (cbi == timeout[i])) {
2083 /* we've wrapped around the circular list, and everything looks ok */
2086 if (cbi > cbstuff.nblks) {
2087 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: list index out of range (cbi=%d, tindex=%d)\n",
2092 if (itot(cb->thead) != &timeout[i]) {
2093 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: cb->thead points to wrong timeout queue (tindex=%d, cbi=%d, cb->thead=%d)\n",
2094 i, cbi, cb->thead));
2097 if (!cb->tprev || !cb->tnext) {
2098 ViceLog(0, ("cb_stateVerifyTimeoutQueues: null index in circular list (cbi=%d, tindex=%d)\n",
2103 if ((cb->tprev > cbstuff.nblks) ||
2104 (cb->tnext > cbstuff.nblks)) {
2105 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list index out of range (cbi=%d, tindex=%d, cb->tprev=%d, cb->tnext=%d, nCBs=%d)\n",
2106 cbi, i, cb->tprev, cb->tnext, cbstuff.nblks));
2110 ncb = itocb(cb->tnext);
2111 if (cbi != ncb->tprev) {
2112 ViceLog(0, ("cb_stateVerifyTimeoutQueues: corrupt linked list (cbi=%d, tindex=%d)\n",
2117 if (chain_len > FS_STATE_TCB_MAX_LIST_LEN) {
2118 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list length exceeds %d (tindex=%d); assuming there's a loop\n",
2119 FS_STATE_TCB_MAX_LIST_LEN, i));
2131 cb_stateSaveTimeouts(struct fs_dump_state * state)
2134 struct iovec iov[2];
2136 AssignInt64(state->eof_offset, &state->cb_hdr->timeout_offset);
2138 memset(state->cb_timeout_hdr, 0, sizeof(struct callback_state_fehash_header));
2139 state->cb_timeout_hdr->magic = CALLBACK_STATE_TIMEOUT_MAGIC;
2140 state->cb_timeout_hdr->records = CB_NUM_TIMEOUT_QUEUES;
2141 state->cb_timeout_hdr->len = sizeof(struct callback_state_timeout_header) +
2142 (state->cb_timeout_hdr->records * sizeof(afs_uint32));
2144 iov[0].iov_base = (char *)state->cb_timeout_hdr;
2145 iov[0].iov_len = sizeof(struct callback_state_timeout_header);
2146 iov[1].iov_base = (char *)timeout;
2147 iov[1].iov_len = sizeof(timeout);
2149 if (fs_stateSeek(state, &state->cb_hdr->timeout_offset)) {
2154 if (fs_stateWriteV(state, iov, 2)) {
2159 fs_stateIncEOF(state, state->cb_timeout_hdr->len);
2166 cb_stateRestoreTimeouts(struct fs_dump_state * state)
2170 if (fs_stateReadHeader(state, &state->cb_hdr->timeout_offset,
2171 state->cb_timeout_hdr,
2172 sizeof(struct callback_state_timeout_header))) {
2177 if (state->cb_timeout_hdr->magic != CALLBACK_STATE_TIMEOUT_MAGIC) {
2181 if (state->cb_timeout_hdr->records != CB_NUM_TIMEOUT_QUEUES) {
2186 len = state->cb_timeout_hdr->records * sizeof(afs_uint32);
2188 if (state->cb_timeout_hdr->len !=
2189 (sizeof(struct callback_state_timeout_header) + len)) {
2194 if (fs_stateRead(state, timeout, len)) {
2204 cb_stateSaveFEHash(struct fs_dump_state * state)
2207 struct iovec iov[2];
2209 AssignInt64(state->eof_offset, &state->cb_hdr->fehash_offset);
2211 memset(state->cb_fehash_hdr, 0, sizeof(struct callback_state_fehash_header));
2212 state->cb_fehash_hdr->magic = CALLBACK_STATE_FEHASH_MAGIC;
2213 state->cb_fehash_hdr->records = FEHASH_SIZE;
2214 state->cb_fehash_hdr->len = sizeof(struct callback_state_fehash_header) +
2215 (state->cb_fehash_hdr->records * sizeof(afs_uint32));
2217 iov[0].iov_base = (char *)state->cb_fehash_hdr;
2218 iov[0].iov_len = sizeof(struct callback_state_fehash_header);
2219 iov[1].iov_base = (char *)HashTable;
2220 iov[1].iov_len = sizeof(HashTable);
2222 if (fs_stateSeek(state, &state->cb_hdr->fehash_offset)) {
2227 if (fs_stateWriteV(state, iov, 2)) {
2232 fs_stateIncEOF(state, state->cb_fehash_hdr->len);
2239 cb_stateRestoreFEHash(struct fs_dump_state * state)
2243 if (fs_stateReadHeader(state, &state->cb_hdr->fehash_offset,
2244 state->cb_fehash_hdr,
2245 sizeof(struct callback_state_fehash_header))) {
2250 if (state->cb_fehash_hdr->magic != CALLBACK_STATE_FEHASH_MAGIC) {
2254 if (state->cb_fehash_hdr->records != FEHASH_SIZE) {
2259 len = state->cb_fehash_hdr->records * sizeof(afs_uint32);
2261 if (state->cb_fehash_hdr->len !=
2262 (sizeof(struct callback_state_fehash_header) + len)) {
2267 if (fs_stateRead(state, HashTable, len)) {
2277 cb_stateSaveFEs(struct fs_dump_state * state)
2281 struct FileEntry *fe;
2283 AssignInt64(state->eof_offset, &state->cb_hdr->fe_offset);
2285 for (hash = 0; hash < FEHASH_SIZE ; hash++) {
2286 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
2288 if (cb_stateSaveFE(state, fe)) {
2300 cb_stateRestoreFEs(struct fs_dump_state * state)
2302 int count, nFEs, ret = 0;
2304 nFEs = state->cb_hdr->nFEs;
2306 for (count = 0; count < nFEs; count++) {
2307 if (cb_stateRestoreFE(state)) {
2318 cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe)
2320 int ret = 0, iovcnt, cbi, written = 0;
2322 struct callback_state_entry_header hdr;
2323 struct FEDiskEntry fedsk;
2324 struct CBDiskEntry cbdsk[16];
2325 struct iovec iov[16];
2326 struct CallBack *cb;
2329 if (fei > state->cb_hdr->fe_max) {
2330 state->cb_hdr->fe_max = fei;
2333 memset(&hdr, 0, sizeof(struct callback_state_entry_header));
2335 if (cb_stateFEToDiskEntry(fe, &fedsk)) {
2340 iov[0].iov_base = (char *)&hdr;
2341 iov[0].iov_len = sizeof(hdr);
2342 iov[1].iov_base = (char *)&fedsk;
2343 iov[1].iov_len = sizeof(struct FEDiskEntry);
2346 for (cbi = fe->firstcb, cb = itocb(cbi);
2348 cbi = cb->cnext, cb = itocb(cbi), hdr.nCBs++) {
2349 if (cbi > state->cb_hdr->cb_max) {
2350 state->cb_hdr->cb_max = cbi;
2352 if (cb_stateCBToDiskEntry(cb, &cbdsk[iovcnt])) {
2356 cbdsk[iovcnt].index = cbi;
2357 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2358 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2360 if ((iovcnt == 16) || (!cb->cnext)) {
2361 if (fs_stateWriteV(state, iov, iovcnt)) {
2370 hdr.magic = CALLBACK_STATE_ENTRY_MAGIC;
2371 hdr.len = sizeof(hdr) + sizeof(struct FEDiskEntry) +
2372 (hdr.nCBs * sizeof(struct CBDiskEntry));
2375 if (fs_stateWriteV(state, iov, iovcnt)) {
2380 if (fs_stateWriteHeader(state, &state->eof_offset, &hdr, sizeof(hdr))) {
2386 fs_stateIncEOF(state, hdr.len);
2389 if (fs_stateSeek(state, &state->eof_offset)) {
2395 state->cb_hdr->nFEs++;
2396 state->cb_hdr->nCBs += hdr.nCBs;
2403 cb_stateRestoreFE(struct fs_dump_state * state)
2405 int ret = 0, iovcnt, nCBs;
2406 struct callback_state_entry_header hdr;
2407 struct FEDiskEntry fedsk;
2408 struct CBDiskEntry cbdsk[16];
2409 struct iovec iov[16];
2410 struct FileEntry * fe;
2412 iov[0].iov_base = (char *)&hdr;
2413 iov[0].iov_len = sizeof(hdr);
2414 iov[1].iov_base = (char *)&fedsk;
2415 iov[1].iov_len = sizeof(fedsk);
2418 if (fs_stateReadV(state, iov, iovcnt)) {
2423 if (hdr.magic != CALLBACK_STATE_ENTRY_MAGIC) {
2430 ViceLog(0, ("cb_stateRestoreFE: ran out of free FileEntry structures\n"));
2435 if (cb_stateDiskEntryToFE(state, &fedsk, fe)) {
2441 for (iovcnt = 0, nCBs = 0;
2444 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2445 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2447 if ((iovcnt == 16) || (nCBs == hdr.nCBs - 1)) {
2448 if (fs_stateReadV(state, iov, iovcnt)) {
2452 if (cb_stateRestoreCBs(state, fe, iov, iovcnt)) {
2466 cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
2467 struct iovec * iov, int niovecs)
2470 struct CallBack * cb;
2471 struct CBDiskEntry * cbdsk;
2473 for (idx = 0; idx < niovecs; idx++) {
2474 cbdsk = (struct CBDiskEntry *) iov[idx].iov_base;
2476 if (cbdsk->cb.hhead < state->h_map.len &&
2477 state->h_map.entries[cbdsk->cb.hhead].valid == FS_STATE_IDX_SKIPPED) {
2481 if ((cb = GetCB()) == NULL) {
2482 ViceLog(0, ("cb_stateRestoreCBs: ran out of free CallBack structures\n"));
2486 if (cb_stateDiskEntryToCB(state, cbdsk, cb)) {
2487 ViceLog(0, ("cb_stateRestoreCBs: corrupt CallBack disk entry\n"));
2499 cb_stateFillHeader(struct callback_state_header * hdr)
2501 hdr->stamp.magic = CALLBACK_STATE_MAGIC;
2502 hdr->stamp.version = CALLBACK_STATE_VERSION;
2503 hdr->tfirst = tfirst;
2508 cb_stateCheckHeader(struct callback_state_header * hdr)
2512 if (hdr->stamp.magic != CALLBACK_STATE_MAGIC) {
2514 } else if (hdr->stamp.version != CALLBACK_STATE_VERSION) {
2516 } else if ((hdr->nFEs > cbstuff.nblks) || (hdr->nCBs > cbstuff.nblks)) {
2517 ViceLog(0, ("cb_stateCheckHeader: saved callback state larger than callback memory allocation\n"));
2523 /* disk entry conversion routines */
2525 cb_stateFEToDiskEntry(struct FileEntry * in, struct FEDiskEntry * out)
2527 memcpy(&out->fe, in, sizeof(struct FileEntry));
2528 out->index = fetoi(in);
2533 cb_stateDiskEntryToFE(struct fs_dump_state * state,
2534 struct FEDiskEntry * in, struct FileEntry * out)
2538 memcpy(out, &in->fe, sizeof(struct FileEntry));
2540 /* setup FE map entry */
2541 if (!in->index || (in->index >= state->fe_map.len)) {
2542 ViceLog(0, ("cb_stateDiskEntryToFE: index (%d) out of range",
2547 state->fe_map.entries[in->index].valid = FS_STATE_IDX_VALID;
2548 state->fe_map.entries[in->index].old_idx = in->index;
2549 state->fe_map.entries[in->index].new_idx = fetoi(out);
2556 cb_stateCBToDiskEntry(struct CallBack * in, struct CBDiskEntry * out)
2558 memcpy(&out->cb, in, sizeof(struct CallBack));
2559 out->index = cbtoi(in);
2564 cb_stateDiskEntryToCB(struct fs_dump_state * state,
2565 struct CBDiskEntry * in, struct CallBack * out)
2569 memcpy(out, &in->cb, sizeof(struct CallBack));
2571 /* setup CB map entry */
2572 if (!in->index || (in->index >= state->cb_map.len)) {
2573 ViceLog(0, ("cb_stateDiskEntryToCB: index (%d) out of range\n",
2578 state->cb_map.entries[in->index].valid = FS_STATE_IDX_VALID;
2579 state->cb_map.entries[in->index].old_idx = in->index;
2580 state->cb_map.entries[in->index].new_idx = cbtoi(out);
2586 /* index map routines */
2588 cb_stateAllocMap(struct fs_dump_state * state)
2590 state->fe_map.len = state->cb_hdr->fe_max + 1;
2591 state->cb_map.len = state->cb_hdr->cb_max + 1;
2592 state->fe_map.entries = (struct idx_map_entry_t *)
2593 calloc(state->fe_map.len, sizeof(struct idx_map_entry_t));
2594 state->cb_map.entries = (struct idx_map_entry_t *)
2595 calloc(state->cb_map.len, sizeof(struct idx_map_entry_t));
2596 return ((state->fe_map.entries != NULL) && (state->cb_map.entries != NULL)) ? 0 : 1;
2600 fe_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2604 /* FEs use a one-based indexing system, so old==0 implies no mapping */
2610 if (old >= state->fe_map.len) {
2611 ViceLog(0, ("fe_OldToNew: index %d is out of range\n", old));
2613 } else if (state->fe_map.entries[old].valid != FS_STATE_IDX_VALID ||
2614 state->fe_map.entries[old].old_idx != old) { /* sanity check */
2615 ViceLog(0, ("fe_OldToNew: index %d points to an invalid FileEntry record\n", old));
2618 *new = state->fe_map.entries[old].new_idx;
2626 cb_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2630 /* CBs use a one-based indexing system, so old==0 implies no mapping */
2636 if (old >= state->cb_map.len) {
2637 ViceLog(0, ("cb_OldToNew: index %d is out of range\n", old));
2639 } else if (state->cb_map.entries[old].valid != FS_STATE_IDX_VALID ||
2640 state->cb_map.entries[old].old_idx != old) { /* sanity check */
2641 ViceLog(0, ("cb_OldToNew: index %d points to an invalid CallBack record\n", old));
2644 *new = state->cb_map.entries[old].new_idx;
2650 #endif /* AFS_DEMAND_ATTACH_FS */
2652 #define DumpBytes(fd,buf,req) if (write(fd, buf, req) < 0) ; /* don't care */
2655 DumpCallBackState_r(void)
2658 afs_uint32 magic = MAGICV2, now = (afs_int32) time(NULL), freelisthead;
2660 oflag = O_WRONLY | O_CREAT | O_TRUNC;
2664 fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, oflag, 0666);
2667 ("Couldn't create callback dump file %s\n",
2668 AFSDIR_SERVER_CBKDUMP_FILEPATH));
2672 * Collect but ignoring the return value of write(2) here,
2673 * to avoid compiler warnings on some platforms.
2675 DumpBytes(fd, &magic, sizeof(magic));
2676 DumpBytes(fd, &now, sizeof(now));
2677 DumpBytes(fd, &cbstuff, sizeof(cbstuff));
2678 DumpBytes(fd, TimeOuts, sizeof(TimeOuts));
2679 DumpBytes(fd, timeout, sizeof(timeout));
2680 DumpBytes(fd, &tfirst, sizeof(tfirst));
2681 freelisthead = cbtoi((struct CallBack *)CBfree);
2682 DumpBytes(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2683 freelisthead = fetoi((struct FileEntry *)FEfree);
2684 DumpBytes(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2685 DumpBytes(fd, HashTable, sizeof(HashTable));
2686 DumpBytes(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2687 DumpBytes(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2694 DumpCallBackState(void) {
2698 rc = DumpCallBackState_r();
2704 #endif /* !INTERPRET_DUMP */
2706 #ifdef INTERPRET_DUMP
2709 ReadBytes(int fd, void *buf, size_t req)
2713 count = read(fd, buf, req);
2717 } else if (count != req) {
2718 fprintf(stderr, "read: premature EOF (expected %lu, got %lu)\n",
2719 (unsigned long)req, (unsigned long)count);
2724 /* This is only compiled in for the callback analyzer program */
2725 /* Returns the time of the dump */
2727 ReadDump(char *file, int timebits)
2730 afs_uint32 magic, freelisthead;
2738 fd = open(file, oflag);
2740 fprintf(stderr, "Couldn't read dump file %s\n", file);
2743 ReadBytes(fd, &magic, sizeof(magic));
2744 if (magic == MAGICV2) {
2747 if (magic != MAGIC) {
2749 "Magic number of %s is invalid. You might be trying to\n",
2752 "run this program on a machine type with a different byte ordering.\n");
2756 if (timebits == 64) {
2757 ReadBytes(fd, &now64, sizeof(afs_int64));
2758 now = (afs_int32) now64;
2760 ReadBytes(fd, &now, sizeof(afs_int32));
2762 ReadBytes(fd, &cbstuff, sizeof(cbstuff));
2763 ReadBytes(fd, TimeOuts, sizeof(TimeOuts));
2764 ReadBytes(fd, timeout, sizeof(timeout));
2765 ReadBytes(fd, &tfirst, sizeof(tfirst));
2766 ReadBytes(fd, &freelisthead, sizeof(freelisthead));
2767 CB = ((struct CallBack
2768 *)(calloc(cbstuff.nblks, sizeof(struct CallBack)))) - 1;
2769 FE = ((struct FileEntry
2770 *)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
2771 CBfree = (struct CallBack *)itocb(freelisthead);
2772 ReadBytes(fd, &freelisthead, sizeof(freelisthead));
2773 FEfree = (struct FileEntry *)itofe(freelisthead);
2774 ReadBytes(fd, HashTable, sizeof(HashTable));
2775 ReadBytes(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2776 ReadBytes(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2778 perror("Error reading dumpfile");
2785 #include "AFS_component_version_number.h"
2787 #include "AFS_component_version_number.c"
2790 static afs_uint32 *cbTrack;
2793 main(int argc, char **argv)
2795 int err = 0, cbi = 0, stats = 0, noptions = 0, all = 0, vol = 0, raw = 0;
2797 struct FileEntry *fe;
2798 struct CallBack *cb;
2802 memset(&fid, 0, sizeof(fid));
2805 while (argc && **argv == '-') {
2808 if (!strcmp(*argv, "-host")) {
2814 cbi = atoi(*++argv);
2815 } else if (!strcmp(*argv, "-fid")) {
2821 fid.Volume = atoi(*++argv);
2822 fid.Vnode = atoi(*++argv);
2823 fid.Unique = atoi(*++argv);
2824 } else if (!strcmp(*argv, "-time")) {
2825 fprintf(stderr, "-time not supported\n");
2827 } else if (!strcmp(*argv, "-stats")) {
2829 } else if (!strcmp(*argv, "-all")) {
2831 } else if (!strcmp(*argv, "-raw")) {
2833 } else if (!strcmp(*argv, "-timebits")) {
2839 timebits = atoi(*++argv);
2840 if ((timebits != 32)
2844 } else if (!strcmp(*argv, "-volume")) {
2850 vol = atoi(*++argv);
2855 if (err || argc != 1) {
2857 "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] [-timebits 32"
2859 "] callbackdumpfile\n");
2861 "[cbid is shown for each host in the hosts.dump file]\n");
2864 now = ReadDump(*argv, timebits);
2865 if (stats || noptions == 0) {
2866 time_t uxtfirst = UXtime(tfirst), tnow = now;
2867 printf("The time of the dump was %u %s", (unsigned int) now, ctime(&tnow));
2868 printf("The last time cleanup ran was %u %s", (unsigned int) uxtfirst,
2870 PrintCallBackStats();
2873 cbTrack = calloc(cbstuff.nblks, sizeof(cbTrack[0]));
2878 struct CallBack *cb;
2879 struct FileEntry *fe;
2881 for (hash = 0; hash < FEHASH_SIZE; hash++) {
2882 for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
2883 if (!vol || (fe->volid == vol)) {
2884 afs_uint32 fe_i = fetoi(fe);
2886 for (cb = itocb(fe->firstcb); cb; cb = itocb(cb->cnext)) {
2887 afs_uint32 cb_i = cbtoi(cb);
2889 if (cb_i > cbstuff.nblks) {
2890 printf("CB index out of range (%u > %d), stopped for this FE\n",
2891 cb_i, cbstuff.nblks);
2895 if (cbTrack[cb_i]) {
2896 printf("CB entry already claimed for FE[%u] (this is FE[%u]), stopped\n",
2897 cbTrack[cb_i], fe_i);
2900 cbTrack[cb_i] = fe_i;
2912 afs_uint32 cfirst = cbi;
2917 } while (cbi != cfirst);
2922 printf("No callback entries for %u.%u\n", fid.Volume, fid.Vnode);
2925 cb = itocb(fe->firstcb);
2928 cb = itocb(cb->cnext);
2933 for (i = 1; i < cbstuff.nblks; i++) {
2934 p = (afs_int32 *) & FE[i];
2935 printf("%d:%12x%12x%12x%12x\n", i, p[0], p[1], p[2], p[3]);
2944 PrintCB(struct CallBack *cb, afs_uint32 now)
2946 struct FileEntry *fe = itofe(cb->fhead);
2947 time_t expires = TIndexToTime(cb->thead);
2952 printf("vol=%" AFS_VOLID_FMT " vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %lu secs at %s",
2953 afs_printable_VolumeId_lu(fe->volid), fe->vnode, fe->ncbs,
2954 cb->hhead, cb->status, fe->status, expires - now, ctime(&expires));
2959 #if !defined(INTERPRET_DUMP)
2961 ** try breaking calbacks on afidp from host. Use multi_rx.
2962 ** return 0 on success, non-zero on failure
2965 MultiBreakCallBackAlternateAddress(struct host *host, struct AFSCBFids *afidp)
2969 retVal = MultiBreakCallBackAlternateAddress_r(host, afidp);
2975 MultiBreakCallBackAlternateAddress_r(struct host *host,
2976 struct AFSCBFids *afidp)
2979 struct rx_connection **conns;
2980 struct rx_connection *connSuccess = 0;
2981 struct AddrPort *interfaces;
2982 static struct rx_securityClass *sc = 0;
2983 static struct AFSCBs tc = { 0, 0 };
2986 /* nothing more can be done */
2987 if (!host->interface)
2988 return 1; /* failure */
2990 /* the only address is the primary interface */
2991 if (host->interface->numberOfInterfaces <= 1)
2992 return 1; /* failure */
2994 /* initialise a security object only once */
2996 sc = rxnull_NewClientSecurityObject();
2998 i = host->interface->numberOfInterfaces;
2999 interfaces = calloc(i, sizeof(struct AddrPort));
3000 conns = calloc(i, sizeof(struct rx_connection *));
3001 if (!interfaces || !conns) {
3002 ViceLogThenPanic(0, ("Failed malloc in "
3003 "MultiBreakCallBackAlternateAddress_r\n"));
3006 /* initialize alternate rx connections */
3007 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3008 /* this is the current primary address */
3009 if (host->host == host->interface->interface[i].addr &&
3010 host->port == host->interface->interface[i].port)
3013 interfaces[j] = host->interface->interface[i];
3015 rx_NewConnection(interfaces[j].addr,
3016 interfaces[j].port, 1, sc, 0);
3017 rx_SetConnDeadTime(conns[j], 2);
3018 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3022 opr_Assert(j); /* at least one alternate address */
3024 ("Starting multibreakcall back on all addr for host %p (%s:%d)\n",
3025 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
3027 multi_Rx(conns, j) {
3028 multi_RXAFSCB_CallBack(afidp, &tc);
3032 if (host->callback_rxcon)
3033 rx_DestroyConnection(host->callback_rxcon);
3034 host->callback_rxcon = conns[multi_i];
3035 /* add then remove */
3036 addInterfaceAddr_r(host, interfaces[multi_i].addr,
3037 interfaces[multi_i].port);
3038 removeInterfaceAddr_r(host, host->host, host->port);
3039 host->host = interfaces[multi_i].addr;
3040 host->port = interfaces[multi_i].port;
3041 connSuccess = conns[multi_i];
3042 rx_SetConnDeadTime(host->callback_rxcon, 50);
3043 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3045 ("multibreakcall success with addr %s:%d\n",
3046 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3047 ntohs(interfaces[multi_i].port)));
3054 /* Destroy all connections except the one on which we succeeded */
3055 for (i = 0; i < j; i++)
3056 if (conns[i] != connSuccess)
3057 rx_DestroyConnection(conns[i]);
3063 return 0; /* success */
3065 return 1; /* failure */
3070 ** try multi_RX probes to host.
3071 ** return 0 on success, non-0 on failure
3074 MultiProbeAlternateAddress_r(struct host *host)
3077 struct rx_connection **conns;
3078 struct rx_connection *connSuccess = 0;
3079 struct AddrPort *interfaces;
3080 static struct rx_securityClass *sc = 0;
3083 /* nothing more can be done */
3084 if (!host->interface)
3085 return 1; /* failure */
3087 /* the only address is the primary interface */
3088 if (host->interface->numberOfInterfaces <= 1)
3089 return 1; /* failure */
3091 /* initialise a security object only once */
3093 sc = rxnull_NewClientSecurityObject();
3095 i = host->interface->numberOfInterfaces;
3096 interfaces = calloc(i, sizeof(struct AddrPort));
3097 conns = calloc(i, sizeof(struct rx_connection *));
3098 if (!interfaces || !conns) {
3099 ViceLogThenPanic(0, ("Failed malloc in "
3100 "MultiProbeAlternateAddress_r\n"));
3103 /* initialize alternate rx connections */
3104 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3105 /* this is the current primary address */
3106 if (host->host == host->interface->interface[i].addr &&
3107 host->port == host->interface->interface[i].port)
3110 interfaces[j] = host->interface->interface[i];
3112 rx_NewConnection(interfaces[j].addr,
3113 interfaces[j].port, 1, sc, 0);
3114 rx_SetConnDeadTime(conns[j], 2);
3115 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3119 opr_Assert(j); /* at least one alternate address */
3121 ("Starting multiprobe on all addr for host %p (%s:%d)\n",
3122 host, afs_inet_ntoa_r(host->host, hoststr),
3123 ntohs(host->port)));
3125 multi_Rx(conns, j) {
3126 multi_RXAFSCB_ProbeUuid(&host->interface->uuid);
3130 if (host->callback_rxcon)
3131 rx_DestroyConnection(host->callback_rxcon);
3132 host->callback_rxcon = conns[multi_i];
3133 /* add then remove */
3134 addInterfaceAddr_r(host, interfaces[multi_i].addr,
3135 interfaces[multi_i].port);
3136 removeInterfaceAddr_r(host, host->host, host->port);
3137 host->host = interfaces[multi_i].addr;
3138 host->port = interfaces[multi_i].port;
3139 connSuccess = conns[multi_i];
3140 rx_SetConnDeadTime(host->callback_rxcon, 50);
3141 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3143 ("multiprobe success with addr %s:%d\n",
3144 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3145 ntohs(interfaces[multi_i].port)));
3150 ("multiprobe failure with addr %s:%d\n",
3151 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3152 ntohs(interfaces[multi_i].port)));
3154 /* This is less than desirable but its the best we can do.
3155 * The AFS Cache Manager will return either 0 for a Uuid
3156 * match and a 1 for a non-match. If the error is 1 we
3157 * therefore know that our mapping of IP address to Uuid
3158 * is wrong. We should attempt to find the correct
3159 * Uuid and fix the host tables.
3161 if (multi_error == 1) {
3162 /* remove the current alternate address from this host */
3164 removeInterfaceAddr_r(host, interfaces[multi_i].addr, interfaces[multi_i].port);
3168 #ifdef AFS_DEMAND_ATTACH_FS
3169 /* try to bail ASAP if the fileserver is shutting down */
3171 if (fs_state.mode == FS_MODE_SHUTDOWN) {
3180 /* Destroy all connections except the one on which we succeeded */
3181 for (i = 0; i < j; i++)
3182 if (conns[i] != connSuccess)
3183 rx_DestroyConnection(conns[i]);
3189 return 0; /* success */
3191 return 1; /* failure */
3194 #endif /* !defined(INTERPRET_DUMP) */