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);
686 /* sort cba list to avoid makecall issues */
687 qsort(cba, ncbas, sizeof(struct cbstruct), CompareCBA);
689 /* set up conns for multi-call */
690 for (i = 0, j = 0; i < ncbas; i++) {
691 struct host *thishost = cba[i].hp;
692 if (!thishost || (thishost->hostFlags & HOSTDELETED)) {
695 rx_GetConnection(thishost->callback_rxcon);
696 multi_to_cba_map[j] = i;
697 conns[j++] = thishost->callback_rxcon;
699 rx_SetConnDeadTime(thishost->callback_rxcon, 4);
700 rx_SetConnHardDeadTime(thishost->callback_rxcon, AFS_HARDDEADTIME);
703 if (j) { /* who knows what multi would do with 0 conns? */
707 multi_RXAFSCB_CallBack(afidp, &tc);
713 i = multi_to_cba_map[multi_i];
719 ("BCB: INTERNAL ERROR: hp=%p, cba=%p, thead=%u\n",
723 ** try breaking callbacks on alternate interface addresses
725 if (MultiBreakCallBackAlternateAddress(hp, afidp)) {
728 ("BCB: Failed on file %u.%u.%u, "
729 "Host %p (%s:%d) is down\n",
730 afidp->AFSCBFids_val->Volume,
731 afidp->AFSCBFids_val->Vnode,
732 afidp->AFSCBFids_val->Unique,
734 afs_inet_ntoa_r(hp->host, hoststr),
740 if (!(hp->hostFlags & HOSTDELETED)) {
741 hp->hostFlags |= VENUSDOWN;
743 * We always go into AddCallBack1_r with the host locked
745 AddCallBack1_r(hp, afidp->AFSCBFids_val, itot(idx),
759 for (i = 0; i < ncbas; i++) {
767 /* H_UNLOCK around this so h_FreeConnection does not deadlock.
768 h_FreeConnection should *never* be called on a callback connection,
769 but on 10/27/04 a deadlock occurred where it was, when we know why,
770 this should be reverted. -- shadow */
772 for (i = 0; i < j; i++) {
773 rx_PutConnection(conns[i]);
781 * Break all call backs for fid, except for the specified host (unless flag
782 * is true, in which case all get a callback message. Assumption: the specified
783 * host is h_Held, by the caller; the others aren't.
784 * Specified host may be bogus, that's ok. This used to check to see if the
785 * host was down in two places, once right after the host was h_held, and
786 * again after it was locked. That race condition is incredibly rare and
787 * relatively harmless even when it does occur, so we don't check for it now.
789 /* if flag is true, send a break callback msg to "host", too */
791 BreakCallBack(struct host *xhost, AFSFid * fid, int flag)
793 struct FileEntry *fe;
794 struct CallBack *cb, *nextcb;
795 struct cbstruct cbaDef[MAX_CB_HOSTS], *cba = cbaDef;
796 unsigned int ncbas, cbaAlloc = MAX_CB_HOSTS;
803 ("BCB: BreakCallBack(Host %p all but %s:%d, (%u,%u,%u))\n",
804 xhost, afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
805 fid->Volume, fid->Vnode, fid->Unique));
808 ("BCB: BreakCallBack(No Host, (%u,%u,%u))\n",
809 fid->Volume, fid->Vnode, fid->Unique));
812 cbstuff.BreakCallBacks++;
817 hostindex = xhost ? h_htoi(xhost) : 0;
818 cb = itocb(fe->firstcb);
819 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
820 /* the most common case is what follows the || */
823 tf.AFSCBFids_len = 1;
824 tf.AFSCBFids_val = fid;
826 for (ncbas = 0; cb ; cb = nextcb) {
827 nextcb = itocb(cb->cnext);
828 if ((cb->hhead != hostindex || flag)
829 && (cb->status == CB_BULK || cb->status == CB_NORMAL
830 || cb->status == CB_VOLUME)) {
831 struct host *thishost = h_itoh(cb->hhead);
833 ViceLog(0, ("BCB: BOGUS! cb->hhead is NULL!\n"));
834 } else if (thishost->hostFlags & VENUSDOWN) {
836 ("BCB: %p (%s:%d) is down; delaying break call back\n",
837 thishost, afs_inet_ntoa_r(thishost->host, hoststr),
838 ntohs(thishost->port)));
839 cb->status = CB_DELAYED;
841 if (!(thishost->hostFlags & HOSTDELETED)) {
843 if (ncbas == cbaAlloc) { /* Need more space */
844 int curLen = cbaAlloc*sizeof(cba[0]);
845 struct cbstruct *cbaOld = (cba == cbaDef) ? NULL : cba;
847 /* There are logical contraints elsewhere that the number of hosts
848 (i.e. h_HTSPERBLOCK*h_MAXHOSTTABLES) remains in the realm of a signed "int".
849 cbaAlloc is defined unsigned int hence doubling below cannot overflow
851 cbaAlloc = cbaAlloc<<1; /* double */
852 cba = realloc(cbaOld, cbaAlloc * sizeof(cba[0]));
854 if (cbaOld == NULL) { /* realloc wouldn't have copied from cbaDef */
855 memcpy(cba, cbaDef, curLen);
858 cba[ncbas].hp = thishost;
859 cba[ncbas].thead = cb->thead;
864 CDel(cb, 1); /* Usually first; so this delete
865 * is reasonably inexpensive */
871 struct cbstruct *cba2;
874 for (cba2 = cba, num = ncbas; ncbas > 0; cba2 += num, ncbas -= num) {
875 num = (ncbas > MAX_CB_HOSTS) ? MAX_CB_HOSTS : ncbas;
876 MultiBreakCallBack_r(cba2, num, &tf);
880 if (cba != cbaDef) free(cba);
887 /* Delete (do not break) single call back for fid */
889 DeleteCallBack(struct host *host, AFSFid * fid)
891 struct FileEntry *fe;
896 cbstuff.DeleteCallBacks++;
899 /* do not care if the host has been HOSTDELETED */
905 ("DCB: No call backs for fid (%u, %u, %u)\n", fid->Volume,
906 fid->Vnode, fid->Unique));
909 pcb = FindCBPtr(fe, host);
912 ("DCB: No call back for host %p (%s:%d), (%u, %u, %u)\n",
913 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
914 fid->Volume, fid->Vnode, fid->Unique));
928 * Delete (do not break) all call backs for fid. This call doesn't
929 * set all of the various host locks, but it shouldn't really matter
930 * since we're not adding callbacks, but deleting them. I'm not sure
931 * why it doesn't set the lock, however; perhaps it should.
934 DeleteFileCallBacks(AFSFid * fid)
936 struct FileEntry *fe;
942 cbstuff.DeleteFiles++;
947 ("DF: No fid (%u,%u,%u) to delete\n", fid->Volume, fid->Vnode,
951 for (n = 0, cbi = fe->firstcb; cbi; n++) {
964 /* Delete (do not break) all call backs for host. The host should be
967 DeleteAllCallBacks_r(struct host *host, int deletefe)
972 cbstuff.DeleteAllCallBacks++;
973 cbi = first = host->cblist;
975 ViceLog(8, ("DV: no call backs\n"));
983 } while (cbi != first);
989 * Break all delayed call backs for host. Returns 1 if all call backs
990 * successfully broken; 0 otherwise. Assumes host is h_Held and h_Locked.
991 * Must be called with VenusDown set for this host
994 BreakDelayedCallBacks(struct host *host)
998 retVal = BreakDelayedCallBacks_r(host);
1004 BreakDelayedCallBacks_r(struct host *host)
1006 struct AFSFid fids[AFSCBMAX];
1007 int cbi, first, nfids;
1008 struct CallBack *cb;
1011 struct rx_connection *cb_conn;
1013 cbstuff.nbreakers++;
1014 if (!(host->hostFlags & RESETDONE) && !(host->hostFlags & HOSTDELETED)) {
1015 host->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1016 cb_conn = host->callback_rxcon;
1017 rx_GetConnection(cb_conn);
1018 if (host->interface) {
1021 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1024 code = RXAFSCB_InitCallBackState(cb_conn);
1026 rx_PutConnection(cb_conn);
1029 host->hostFlags |= ALTADDR; /* alternate addresses are valid */
1033 ("CB: Call back connect back failed (in break delayed) "
1034 "for Host %p (%s:%d)\n",
1035 host, afs_inet_ntoa_r(host->host, hoststr),
1036 ntohs(host->port)));
1038 host->hostFlags |= VENUSDOWN;
1041 ("InitCallBackState success on %p (%s:%d)\n",
1042 host, afs_inet_ntoa_r(host->host, hoststr),
1043 ntohs(host->port)));
1044 /* reset was done successfully */
1045 host->hostFlags |= RESETDONE;
1046 host->hostFlags &= ~VENUSDOWN;
1049 while (!(host->hostFlags & HOSTDELETED)) {
1051 host->hostFlags &= ~VENUSDOWN; /* presume up */
1052 cbi = first = host->cblist;
1056 first = host->cblist;
1059 if (cb->status == CB_DELAYED) {
1060 struct FileEntry *fe = itofe(cb->fhead);
1061 fids[nfids].Volume = fe->volid;
1062 fids[nfids].Vnode = fe->vnode;
1063 fids[nfids].Unique = fe->unique;
1069 } while (cbi && cbi != first && nfids < AFSCBMAX);
1075 if (XCallBackBulk_r(host, fids, nfids)) {
1076 /* Failed, again: put them back, probably with old
1081 ("CB: XCallBackBulk failed, Host %p (%s:%d); "
1082 "callback list follows:\n",
1083 host, afs_inet_ntoa_r(host->host, hoststr),
1084 ntohs(host->port)));
1086 for (i = 0; i < nfids; i++) {
1089 ("CB: Host %p (%s:%d), file %u.%u.%u "
1090 "(part of bulk callback)\n",
1091 host, afs_inet_ntoa_r(host->host, hoststr),
1092 ntohs(host->port), fids[i].Volume,
1093 fids[i].Vnode, fids[i].Unique));
1096 * AddCallBack1_r(host, &fids[i], itot(thead[i]), CB_DELAYED, 1);
1097 * * but it turns out to cause too many tricky locking problems.
1098 * * now, if break delayed fails, screw it. */
1100 host->hostFlags |= VENUSDOWN; /* Failed */
1101 ClearHostCallbacks_r(host, 1 /* locked */ );
1105 if (nfids < AFSCBMAX)
1109 cbstuff.nbreakers--;
1110 /* If we succeeded it's always ok to unset HFE_LATER */
1111 if (!(host->hostFlags & VENUSDOWN))
1112 host->hostFlags &= ~HFE_LATER;
1113 return (host->hostFlags & VENUSDOWN);
1117 MultiBreakVolumeCallBack_r(struct host *host,
1118 struct VCBParams *parms, int deletefe)
1122 if (host->hostFlags & HOSTDELETED)
1125 if (!(host->hostFlags & HCBREAK))
1126 return 0; /* host is not flagged to notify */
1128 if (host->hostFlags & VENUSDOWN) {
1130 /* Do not care if the host is now HOSTDELETED */
1133 ("BVCB: volume callback for Host %p (%s:%d) failed\n",
1134 host, afs_inet_ntoa_r(host->host, hoststr),
1135 ntohs(host->port)));
1137 DeleteAllCallBacks_r(host, deletefe); /* Delete all callback state
1138 * rather than attempting to
1139 * selectively remember to
1140 * delete the volume callbacks
1142 host->hostFlags &= ~(RESETDONE|HCBREAK); /* Do InitCallBackState when host returns */
1146 opr_Assert(parms->ncbas <= MAX_CB_HOSTS);
1148 /* Do not call MultiBreakCallBack on the current host structure
1149 ** because it would prematurely release the hold on the host
1151 if (parms->ncbas == MAX_CB_HOSTS) {
1152 struct AFSCBFids tf;
1154 tf.AFSCBFids_len = 1;
1155 tf.AFSCBFids_val = parms->fid;
1157 /* this releases all the hosts */
1158 MultiBreakCallBack_r(parms->cba, parms->ncbas, &tf);
1162 parms->cba[parms->ncbas].hp = host;
1163 parms->cba[(parms->ncbas)++].thead = parms->thead;
1164 host->hostFlags &= ~HCBREAK;
1166 /* we have more work to do on this host, so make sure we keep a reference
1174 MultiBreakVolumeLaterCallBack(struct host *host, void *rock)
1176 struct VCBParams *parms = (struct VCBParams *)rock;
1179 retval = MultiBreakVolumeCallBack_r(host, parms, 0);
1185 * Break all call backs on a single volume. Don't call this with any
1186 * hosts h_held. Note that this routine clears the callbacks before
1187 * actually breaking them, and that the vnode isn't locked during this
1188 * operation, so that people might see temporary callback loss while
1189 * this function is executing. It is just a temporary state, however,
1190 * since the callback will be broken later by this same function.
1192 * Now uses multi-RX for CallBack RPC in a different thread,
1193 * only marking them here.
1195 extern pthread_cond_t fsync_cond;
1198 BreakVolumeCallBacksLater(VolumeId volume)
1202 struct FileEntry *fe;
1203 struct CallBack *cb;
1207 ViceLog(25, ("Setting later on volume %" AFS_VOLID_FMT "\n",
1208 afs_printable_VolumeId_lu(volume)));
1210 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1211 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1212 if (fe->volid == volume) {
1213 struct CallBack *cbnext;
1214 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1215 host = h_itoh(cb->hhead);
1216 host->hostFlags |= HFE_LATER;
1217 cb->status = CB_DELAYED;
1218 cbnext = itocb(cb->cnext);
1221 fe->status |= FE_LATER;
1230 /* didn't find any callbacks, so return right away. */
1234 ViceLog(25, ("Fsync thread wakeup\n"));
1236 opr_cv_broadcast(&fsync_cond);
1242 BreakLaterCallBacks(void)
1247 struct CallBack *cb;
1248 struct FileEntry *fe = NULL;
1249 struct FileEntry *myfe = NULL;
1251 struct VCBParams henumParms;
1252 unsigned short tthead = 0; /* zero is illegal value */
1256 ViceLog(25, ("Looking for FileEntries to unchain\n"));
1259 /* Pick the first volume we see to clean up */
1260 fid.Volume = fid.Vnode = fid.Unique = 0;
1262 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1263 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1264 if (fe && (fe->status & FE_LATER)
1265 && (fid.Volume == 0 || fid.Volume == fe->volid)) {
1266 /* Ugly, but used to avoid left side casting */
1267 struct object *tmpfe;
1269 ("Unchaining for %u:%u:%" AFS_VOLID_FMT "\n", fe->vnode,
1270 fe->unique, afs_printable_VolumeId_lu(fe->volid)));
1271 fid.Volume = fe->volid;
1273 fe->status &= ~FE_LATER; /* not strictly needed */
1274 /* Works since volid is deeper than the largest pointer */
1275 tmpfe = (struct object *)fe;
1276 tmpfe->next = (struct object *)myfe;
1289 /* loop over FEs from myfe and free/break */
1291 for (fe = myfe; fe;) {
1292 struct CallBack *cbnext;
1293 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1294 cbnext = itocb(cb->cnext);
1295 host = h_itoh(cb->hhead);
1296 if (cb->status == CB_DELAYED) {
1297 if (!(host->hostFlags & HOSTDELETED)) {
1298 /* mark this host for notification */
1299 host->hostFlags |= HCBREAK;
1300 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1306 CDel(cb, 0); /* Don't let CDel clean up the fe */
1307 /* leave flag for MultiBreakVolumeCallBack to clear */
1310 ("Found host %p (%s:%d) non-DELAYED cb for %u:%u:%" AFS_VOLID_FMT "\n",
1311 host, afs_inet_ntoa_r(host->host, hoststr),
1312 ntohs(host->port), fe->vnode, fe->unique,
1313 afs_printable_VolumeId_lu(fe->volid)));
1317 fe = (struct FileEntry *)((struct object *)fe)->next;
1322 ViceLog(125, ("Breaking volume %u\n", fid.Volume));
1323 henumParms.ncbas = 0;
1324 henumParms.fid = &fid;
1325 henumParms.thead = tthead;
1327 h_Enumerate(MultiBreakVolumeLaterCallBack, (char *)&henumParms);
1329 if (henumParms.ncbas) { /* do left-overs */
1330 struct AFSCBFids tf;
1331 tf.AFSCBFids_len = 1;
1332 tf.AFSCBFids_val = &fid;
1334 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf);
1335 henumParms.ncbas = 0;
1340 /* Arrange to be called again */
1345 * Delete all timed-out call back entries (to be called periodically by file
1349 CleanupTimedOutCallBacks(void)
1352 CleanupTimedOutCallBacks_r();
1358 CleanupTimedOutCallBacks_r(void)
1360 afs_uint32 now = CBtime(time(NULL));
1362 struct CallBack *cb;
1366 while (tfirst <= now) {
1368 cbi = *(thead = THead(tfirst));
1374 ("CCB: deleting timed out call back %x (%s:%d), (%" AFS_VOLID_FMT ",%u,%u)\n",
1375 h_itoh(cb->hhead)->host,
1376 afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
1377 h_itoh(cb->hhead)->port,
1378 afs_printable_VolumeId_lu(itofe(cb->fhead)->volid),
1379 itofe(cb->fhead)->vnode, itofe(cb->fhead)->unique));
1383 if (ntimedout > cbstuff.nblks) {
1384 ViceLog(0, ("CCB: Internal Error -- shutting down...\n"));
1385 DumpCallBackState_r();
1386 ShutDownAndCore(PANIC);
1388 } while (cbi != *thead);
1393 cbstuff.CBsTimedOut += ntimedout;
1394 ViceLog(7, ("CCB: deleted %d timed out callbacks\n", ntimedout));
1395 return (ntimedout > 0);
1399 * parameters to pass to lih*_r from h_Enumerate_r when trying to find a host
1400 * from which to clear callbacks.
1404 * Points to the least interesting host found; try to clear callbacks on
1405 * this host after h_Enumerate_r(lih*_r)'ing.
1410 * The last host we got from lih*_r, but we couldn't clear its callbacks
1411 * for some reason. Choose the next-best host after this one (with the
1412 * current lih*_r, this means to only select hosts that have an ActiveCall
1413 * newer than lastlih).
1415 struct host *lastlih;
1418 /* Value of host->refCount that allows us to reliably infer that
1419 * host may be held by some other thread */
1420 #define OTHER_MUSTHOLD_LIH 2
1422 /* This version does not allow 'host' to be selected unless its ActiveCall
1423 * is newer than 'params->lastlih' which is the host with the oldest
1424 * ActiveCall from the last pass (if it is provided). We filter out any hosts
1425 * that are are held by other threads.
1427 * There is a small problem here, but it may not be easily fixable. Say we
1428 * select some host A, and give it back to GetSomeSpace_r. GSS_r for some
1429 * reason cannot clear the callbacks on A, and so calls us again with
1430 * lastlih = A. Suppose there is another host B that has the same ActiveCall
1431 * time as A. We will now skip over host B, since
1432 * 'hostB->ActiveCall > hostA->ActiveCall' is not true. This could result in
1433 * us prematurely going to the GSS_r 2nd or 3rd pass, and making us a little
1434 * inefficient. This should be pretty rare, though, except perhaps in cases
1435 * with very small numbers of hosts.
1437 * Also filter out any hosts with HOSTDELETED set. h_Enumerate_r should in
1438 * theory not give these to us anyway, but be paranoid.
1441 lih0_r(struct host *host, void *rock)
1443 struct lih_params *params = (struct lih_params *)rock;
1445 /* OTHER_MUSTHOLD_LIH is because the h_Enum loop holds us once */
1447 && (!(host->hostFlags & HOSTDELETED))
1448 && (host->refCount < OTHER_MUSTHOLD_LIH)
1449 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1450 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1453 h_Release_r(params->lih); /* release prev host */
1462 /* same as lih0_r, except we do not prevent held hosts from being selected. */
1464 lih1_r(struct host *host, void *rock)
1466 struct lih_params *params = (struct lih_params *)rock;
1469 && (!(host->hostFlags & HOSTDELETED))
1470 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1471 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1474 h_Release_r(params->lih); /* release prev host */
1483 /* This could be upgraded to get more space each time */
1484 /* first pass: sequentially find the oldest host which isn't held by
1485 anyone for which we can clear callbacks;
1487 /* second pass: sequentially find the oldest host regardless of
1488 whether or not the host is held; skipping 'hostp' */
1489 /* third pass: attempt to clear callbacks from 'hostp' */
1490 /* always called with hostp unlocked */
1492 /* Note: hostlist is ordered most recently created host first and
1493 * its order has no relationship to the most recently used. */
1494 extern struct host *hostList;
1496 GetSomeSpace_r(struct host *hostp, int locked)
1499 struct lih_params params;
1502 if (cbstuff.GotSomeSpaces == 0) {
1503 /* only log this once; if GSS is getting called constantly, that's not
1504 * good but don't make things worse by spamming the log. */
1505 ViceLog(0, ("We have run out of callback space; forcing callback revocation. "
1506 "This suggests the fileserver is configured with insufficient "
1507 "callbacks; you probably want to increase the -cb fileserver "
1508 "parameter (current setting: %u). The fileserver will continue "
1509 "to operate, but this may indicate a severe performance problem\n",
1511 ViceLog(0, ("This message is logged at most once; for more information "
1512 "see the OpenAFS documentation and fileserver xstat collection 3\n"));
1515 cbstuff.GotSomeSpaces++;
1517 ("GSS: First looking for timed out call backs via CleanupCallBacks\n"));
1518 if (CleanupTimedOutCallBacks_r()) {
1524 params.lastlih = NULL;
1529 h_Enumerate_r(i == 0 ? lih0_r : lih1_r, hostList, ¶ms);
1532 if (params.lastlih) {
1533 h_Release_r(params.lastlih);
1534 params.lastlih = NULL;
1538 /* note that 'hp' was held by lih*_r; we will need to release it */
1540 if ((hp != hostp) && !ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
1545 params.lastlih = hp;
1546 /* params.lastlih will be released on the next iteration, after
1551 * Next time try getting callbacks from any host even if
1552 * it's held, since the only other option is starvation for
1553 * the file server (i.e. until the callback timeout arrives).
1556 params.lastlih = NULL;
1559 ("GSS: Try harder for longest inactive host cnt= %d\n",
1564 /* Could not obtain space from other hosts, clear hostp's callback state */
1569 ClearHostCallbacks_r(hostp, 1 /*already locked */ );
1576 /* locked - set if caller has already locked the host */
1578 ClearHostCallbacks_r(struct host *hp, int locked)
1582 struct rx_connection *cb_conn = NULL;
1585 ("GSS: Delete longest inactive host %p (%s:%d)\n",
1586 hp, afs_inet_ntoa_r(hp->host, hoststr), ntohs(hp->port)));
1588 if ((hp->hostFlags & HOSTDELETED)) {
1589 /* hp could go away after reacquiring H_LOCK in h_NBLock_r, so we can't
1590 * really use it; its callbacks will get cleared anyway when
1591 * h_TossStuff_r gets its hands on it */
1597 /** Try a non-blocking lock. If the lock is already held return
1598 * after releasing hold on hp
1601 if (h_NBLock_r(hp)) {
1606 if (hp->Console & 2) {
1608 * If the special console field is set it means that a thread
1609 * is waiting in AddCallBack1 after it set pointers to the
1610 * file entry and/or callback entry. Because of the bogus
1611 * usage of h_hold it won't prevent from another thread, this
1612 * one, to remove all the callbacks so just to be safe we keep
1613 * a reference. NOTE, on the last phase we'll free the calling
1614 * host's callbacks but that's ok...
1618 DeleteAllCallBacks_r(hp, 1);
1619 if (hp->hostFlags & VENUSDOWN) {
1620 hp->hostFlags &= ~RESETDONE; /* remember that we must do a reset */
1621 } else if (!(hp->hostFlags & HOSTDELETED)) {
1622 /* host is up, try a call */
1623 hp->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1624 cb_conn = hp->callback_rxcon;
1625 rx_GetConnection(hp->callback_rxcon);
1626 if (hp->interface) {
1629 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1632 code = RXAFSCB_InitCallBackState(cb_conn);
1634 rx_PutConnection(cb_conn);
1637 hp->hostFlags |= ALTADDR; /* alternate addresses are valid */
1639 /* failed, mark host down and need reset */
1640 hp->hostFlags |= VENUSDOWN;
1641 hp->hostFlags &= ~RESETDONE;
1643 /* reset succeeded, we're done */
1644 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,
1667 fprintf(stderr, "%d GSS1, %d GSS2, %d GSS3, %d GSS4, %d GSS5 (internal counters)\n",
1668 cbstuff.GSS1, cbstuff.GSS2, cbstuff.GSS3, cbstuff.GSS4, cbstuff.GSS5);
1673 #define MAGIC 0x12345678 /* To check byte ordering of dump when it is read in */
1674 #define MAGICV2 0x12345679 /* To check byte ordering & version of dump when it is read in */
1677 #ifndef INTERPRET_DUMP
1679 #ifdef AFS_DEMAND_ATTACH_FS
1682 * callback state serialization
1684 static int cb_stateSaveTimeouts(struct fs_dump_state * state);
1685 static int cb_stateSaveFEHash(struct fs_dump_state * state);
1686 static int cb_stateSaveFEs(struct fs_dump_state * state);
1687 static int cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe);
1688 static int cb_stateRestoreTimeouts(struct fs_dump_state * state);
1689 static int cb_stateRestoreFEHash(struct fs_dump_state * state);
1690 static int cb_stateRestoreFEs(struct fs_dump_state * state);
1691 static int cb_stateRestoreFE(struct fs_dump_state * state);
1692 static int cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
1693 struct iovec * iov, int niovecs);
1695 static int cb_stateVerifyFEHash(struct fs_dump_state * state);
1696 static int cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe);
1697 static int cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe);
1698 static int cb_stateVerifyTimeoutQueues(struct fs_dump_state * state);
1700 static int cb_stateFEToDiskEntry(struct FileEntry *, struct FEDiskEntry *);
1701 static int cb_stateDiskEntryToFE(struct fs_dump_state * state,
1702 struct FEDiskEntry *, struct FileEntry *);
1704 static int cb_stateCBToDiskEntry(struct CallBack *, struct CBDiskEntry *);
1705 static int cb_stateDiskEntryToCB(struct fs_dump_state * state,
1706 struct CBDiskEntry *, struct CallBack *);
1708 static int cb_stateFillHeader(struct callback_state_header * hdr);
1709 static int cb_stateCheckHeader(struct callback_state_header * hdr);
1711 static int cb_stateAllocMap(struct fs_dump_state * state);
1714 cb_stateSave(struct fs_dump_state * state)
1718 AssignInt64(state->eof_offset, &state->hdr->cb_offset);
1720 /* invalidate callback state header */
1721 memset(state->cb_hdr, 0, sizeof(struct callback_state_header));
1722 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1723 sizeof(struct callback_state_header))) {
1728 fs_stateIncEOF(state, sizeof(struct callback_state_header));
1730 /* dump timeout state */
1731 if (cb_stateSaveTimeouts(state)) {
1736 /* dump fe hashtable state */
1737 if (cb_stateSaveFEHash(state)) {
1742 /* dump callback state */
1743 if (cb_stateSaveFEs(state)) {
1748 /* write the callback state header to disk */
1749 cb_stateFillHeader(state->cb_hdr);
1750 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1751 sizeof(struct callback_state_header))) {
1761 cb_stateRestore(struct fs_dump_state * state)
1765 if (fs_stateReadHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1766 sizeof(struct callback_state_header))) {
1771 if (cb_stateCheckHeader(state->cb_hdr)) {
1776 if (cb_stateAllocMap(state)) {
1781 if (cb_stateRestoreTimeouts(state)) {
1786 if (cb_stateRestoreFEHash(state)) {
1791 /* restore FEs and CBs from disk */
1792 if (cb_stateRestoreFEs(state)) {
1797 /* restore the timeout queue heads */
1798 tfirst = state->cb_hdr->tfirst;
1805 cb_stateRestoreIndices(struct fs_dump_state * state)
1808 struct FileEntry * fe;
1809 struct CallBack * cb;
1811 /* restore indices in the FileEntry structures */
1812 for (i = 1; i < state->fe_map.len; i++) {
1813 if (state->fe_map.entries[i].new_idx) {
1814 fe = itofe(state->fe_map.entries[i].new_idx);
1816 /* restore the fe->fnext entry */
1817 if (fe_OldToNew(state, fe->fnext, &fe->fnext)) {
1822 /* restore the fe->firstcb entry */
1823 if (cb_OldToNew(state, fe->firstcb, &fe->firstcb)) {
1830 /* restore indices in the CallBack structures */
1831 for (i = 1; i < state->cb_map.len; i++) {
1832 if (state->cb_map.entries[i].new_idx) {
1833 cb = itocb(state->cb_map.entries[i].new_idx);
1835 /* restore the cb->cnext entry */
1836 if (cb_OldToNew(state, cb->cnext, &cb->cnext)) {
1841 /* restore the cb->fhead entry */
1842 if (fe_OldToNew(state, cb->fhead, &cb->fhead)) {
1847 /* restore the cb->hhead entry */
1848 if (h_OldToNew(state, cb->hhead, &cb->hhead)) {
1853 /* restore the cb->tprev entry */
1854 if (cb_OldToNew(state, cb->tprev, &cb->tprev)) {
1859 /* restore the cb->tnext entry */
1860 if (cb_OldToNew(state, cb->tnext, &cb->tnext)) {
1865 /* restore the cb->hprev entry */
1866 if (cb_OldToNew(state, cb->hprev, &cb->hprev)) {
1871 /* restore the cb->hnext entry */
1872 if (cb_OldToNew(state, cb->hnext, &cb->hnext)) {
1879 /* restore the timeout queue head indices */
1880 for (i = 0; i < state->cb_timeout_hdr->records; i++) {
1881 if (cb_OldToNew(state, timeout[i], &timeout[i])) {
1887 /* restore the FE hash table queue heads */
1888 for (i = 0; i < state->cb_fehash_hdr->records; i++) {
1889 if (fe_OldToNew(state, HashTable[i], &HashTable[i])) {
1900 cb_stateVerify(struct fs_dump_state * state)
1904 if (cb_stateVerifyFEHash(state)) {
1908 if (cb_stateVerifyTimeoutQueues(state)) {
1916 cb_stateVerifyFEHash(struct fs_dump_state * state)
1919 struct FileEntry * fe;
1920 afs_uint32 fei, chain_len;
1922 for (i = 0; i < FEHASH_SIZE; i++) {
1924 for (fei = HashTable[i], fe = itofe(fei);
1926 fei = fe->fnext, fe = itofe(fei)) {
1927 if (fei > cbstuff.nblks) {
1928 ViceLog(0, ("cb_stateVerifyFEHash: error: index out of range (fei=%d)\n", fei));
1932 if (cb_stateVerifyFE(state, fe)) {
1935 if (chain_len > FS_STATE_FE_MAX_HASH_CHAIN_LEN) {
1936 ViceLog(0, ("cb_stateVerifyFEHash: error: hash chain %d length exceeds %d; assuming there's a loop\n",
1937 i, FS_STATE_FE_MAX_HASH_CHAIN_LEN));
1949 cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe)
1953 if ((fe->firstcb && !fe->ncbs) ||
1954 (!fe->firstcb && fe->ncbs)) {
1955 ViceLog(0, ("cb_stateVerifyFE: error: fe->firstcb does not agree with fe->ncbs (fei=%lu, fe->firstcb=%lu, fe->ncbs=%lu)\n",
1956 afs_printable_uint32_lu(fetoi(fe)),
1957 afs_printable_uint32_lu(fe->firstcb),
1958 afs_printable_uint32_lu(fe->ncbs)));
1961 if (cb_stateVerifyFCBList(state, fe)) {
1962 ViceLog(0, ("cb_stateVerifyFE: error: FCBList failed verification (fei=%lu)\n",
1963 afs_printable_uint32_lu(fetoi(fe))));
1971 cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe)
1974 afs_uint32 cbi, fei, chain_len = 0;
1975 struct CallBack * cb;
1979 for (cbi = fe->firstcb, cb = itocb(cbi);
1981 cbi = cb->cnext, cb = itocb(cbi)) {
1982 if (cbi > cbstuff.nblks) {
1983 ViceLog(0, ("cb_stateVerifyFCBList: error: list index out of range (cbi=%d, ncbs=%d)\n",
1984 cbi, cbstuff.nblks));
1988 if (cb->fhead != fei) {
1989 ViceLog(0, ("cb_stateVerifyFCBList: error: cb->fhead != fei (fei=%d, cb->fhead=%d)\n",
1993 if (chain_len > FS_STATE_FCB_MAX_LIST_LEN) {
1994 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (fei=%d); assuming there's a loop\n",
1995 FS_STATE_FCB_MAX_LIST_LEN, fei));
2002 if (fe->ncbs != chain_len) {
2003 ViceLog(0, ("cb_stateVerifyFCBList: error: list length mismatch (len=%d, fe->ncbs=%d)\n",
2004 chain_len, fe->ncbs));
2013 cb_stateVerifyHCBList(struct fs_dump_state * state, struct host * host)
2016 afs_uint32 hi, chain_len, cbi;
2017 struct CallBack *cb, *ncb;
2022 for (cbi = host->cblist, cb = itocb(cbi);
2024 cbi = cb->hnext, cb = ncb) {
2025 if (chain_len && (host->cblist == cbi)) {
2026 /* we've wrapped around the circular list, and everything looks ok */
2029 if (cb->hhead != hi) {
2030 ViceLog(0, ("cb_stateVerifyHCBList: error: incorrect cb->hhead (cbi=%d, h->index=%d, cb->hhead=%d)\n",
2031 cbi, hi, cb->hhead));
2034 if (!cb->hprev || !cb->hnext) {
2035 ViceLog(0, ("cb_stateVerifyHCBList: error: null index in circular list (cbi=%d, h->index=%d)\n",
2040 if ((cb->hprev > cbstuff.nblks) ||
2041 (cb->hnext > cbstuff.nblks)) {
2042 ViceLog(0, ("cb_stateVerifyHCBList: error: list index out of range (cbi=%d, h->index=%d, cb->hprev=%d, cb->hnext=%d, nCBs=%d)\n",
2043 cbi, hi, cb->hprev, cb->hnext, cbstuff.nblks));
2047 ncb = itocb(cb->hnext);
2048 if (cbi != ncb->hprev) {
2049 ViceLog(0, ("cb_stateVerifyHCBList: error: corrupt linked list (cbi=%d, h->index=%d)\n",
2054 if (chain_len > FS_STATE_HCB_MAX_LIST_LEN) {
2055 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (h->index=%d); assuming there's a loop\n",
2056 FS_STATE_HCB_MAX_LIST_LEN, hi));
2068 cb_stateVerifyTimeoutQueues(struct fs_dump_state * state)
2071 afs_uint32 cbi, chain_len;
2072 struct CallBack *cb, *ncb;
2074 for (i = 0; i < CB_NUM_TIMEOUT_QUEUES; i++) {
2076 for (cbi = timeout[i], cb = itocb(cbi);
2078 cbi = cb->tnext, cb = ncb) {
2079 if (chain_len && (cbi == timeout[i])) {
2080 /* we've wrapped around the circular list, and everything looks ok */
2083 if (cbi > cbstuff.nblks) {
2084 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: list index out of range (cbi=%d, tindex=%d)\n",
2089 if (itot(cb->thead) != &timeout[i]) {
2090 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: cb->thead points to wrong timeout queue (tindex=%d, cbi=%d, cb->thead=%d)\n",
2091 i, cbi, cb->thead));
2094 if (!cb->tprev || !cb->tnext) {
2095 ViceLog(0, ("cb_stateVerifyTimeoutQueues: null index in circular list (cbi=%d, tindex=%d)\n",
2100 if ((cb->tprev > cbstuff.nblks) ||
2101 (cb->tnext > cbstuff.nblks)) {
2102 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list index out of range (cbi=%d, tindex=%d, cb->tprev=%d, cb->tnext=%d, nCBs=%d)\n",
2103 cbi, i, cb->tprev, cb->tnext, cbstuff.nblks));
2107 ncb = itocb(cb->tnext);
2108 if (cbi != ncb->tprev) {
2109 ViceLog(0, ("cb_stateVerifyTimeoutQueues: corrupt linked list (cbi=%d, tindex=%d)\n",
2114 if (chain_len > FS_STATE_TCB_MAX_LIST_LEN) {
2115 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list length exceeds %d (tindex=%d); assuming there's a loop\n",
2116 FS_STATE_TCB_MAX_LIST_LEN, i));
2128 cb_stateSaveTimeouts(struct fs_dump_state * state)
2131 struct iovec iov[2];
2133 AssignInt64(state->eof_offset, &state->cb_hdr->timeout_offset);
2135 memset(state->cb_timeout_hdr, 0, sizeof(struct callback_state_fehash_header));
2136 state->cb_timeout_hdr->magic = CALLBACK_STATE_TIMEOUT_MAGIC;
2137 state->cb_timeout_hdr->records = CB_NUM_TIMEOUT_QUEUES;
2138 state->cb_timeout_hdr->len = sizeof(struct callback_state_timeout_header) +
2139 (state->cb_timeout_hdr->records * sizeof(afs_uint32));
2141 iov[0].iov_base = (char *)state->cb_timeout_hdr;
2142 iov[0].iov_len = sizeof(struct callback_state_timeout_header);
2143 iov[1].iov_base = (char *)timeout;
2144 iov[1].iov_len = sizeof(timeout);
2146 if (fs_stateSeek(state, &state->cb_hdr->timeout_offset)) {
2151 if (fs_stateWriteV(state, iov, 2)) {
2156 fs_stateIncEOF(state, state->cb_timeout_hdr->len);
2163 cb_stateRestoreTimeouts(struct fs_dump_state * state)
2167 if (fs_stateReadHeader(state, &state->cb_hdr->timeout_offset,
2168 state->cb_timeout_hdr,
2169 sizeof(struct callback_state_timeout_header))) {
2174 if (state->cb_timeout_hdr->magic != CALLBACK_STATE_TIMEOUT_MAGIC) {
2178 if (state->cb_timeout_hdr->records != CB_NUM_TIMEOUT_QUEUES) {
2183 len = state->cb_timeout_hdr->records * sizeof(afs_uint32);
2185 if (state->cb_timeout_hdr->len !=
2186 (sizeof(struct callback_state_timeout_header) + len)) {
2191 if (fs_stateRead(state, timeout, len)) {
2201 cb_stateSaveFEHash(struct fs_dump_state * state)
2204 struct iovec iov[2];
2206 AssignInt64(state->eof_offset, &state->cb_hdr->fehash_offset);
2208 memset(state->cb_fehash_hdr, 0, sizeof(struct callback_state_fehash_header));
2209 state->cb_fehash_hdr->magic = CALLBACK_STATE_FEHASH_MAGIC;
2210 state->cb_fehash_hdr->records = FEHASH_SIZE;
2211 state->cb_fehash_hdr->len = sizeof(struct callback_state_fehash_header) +
2212 (state->cb_fehash_hdr->records * sizeof(afs_uint32));
2214 iov[0].iov_base = (char *)state->cb_fehash_hdr;
2215 iov[0].iov_len = sizeof(struct callback_state_fehash_header);
2216 iov[1].iov_base = (char *)HashTable;
2217 iov[1].iov_len = sizeof(HashTable);
2219 if (fs_stateSeek(state, &state->cb_hdr->fehash_offset)) {
2224 if (fs_stateWriteV(state, iov, 2)) {
2229 fs_stateIncEOF(state, state->cb_fehash_hdr->len);
2236 cb_stateRestoreFEHash(struct fs_dump_state * state)
2240 if (fs_stateReadHeader(state, &state->cb_hdr->fehash_offset,
2241 state->cb_fehash_hdr,
2242 sizeof(struct callback_state_fehash_header))) {
2247 if (state->cb_fehash_hdr->magic != CALLBACK_STATE_FEHASH_MAGIC) {
2251 if (state->cb_fehash_hdr->records != FEHASH_SIZE) {
2256 len = state->cb_fehash_hdr->records * sizeof(afs_uint32);
2258 if (state->cb_fehash_hdr->len !=
2259 (sizeof(struct callback_state_fehash_header) + len)) {
2264 if (fs_stateRead(state, HashTable, len)) {
2274 cb_stateSaveFEs(struct fs_dump_state * state)
2278 struct FileEntry *fe;
2280 AssignInt64(state->eof_offset, &state->cb_hdr->fe_offset);
2282 for (hash = 0; hash < FEHASH_SIZE ; hash++) {
2283 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
2285 if (cb_stateSaveFE(state, fe)) {
2297 cb_stateRestoreFEs(struct fs_dump_state * state)
2299 int count, nFEs, ret = 0;
2301 nFEs = state->cb_hdr->nFEs;
2303 for (count = 0; count < nFEs; count++) {
2304 if (cb_stateRestoreFE(state)) {
2315 cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe)
2317 int ret = 0, iovcnt, cbi, written = 0;
2319 struct callback_state_entry_header hdr;
2320 struct FEDiskEntry fedsk;
2321 struct CBDiskEntry cbdsk[16];
2322 struct iovec iov[16];
2323 struct CallBack *cb;
2326 if (fei > state->cb_hdr->fe_max) {
2327 state->cb_hdr->fe_max = fei;
2330 memset(&hdr, 0, sizeof(struct callback_state_entry_header));
2332 if (cb_stateFEToDiskEntry(fe, &fedsk)) {
2337 iov[0].iov_base = (char *)&hdr;
2338 iov[0].iov_len = sizeof(hdr);
2339 iov[1].iov_base = (char *)&fedsk;
2340 iov[1].iov_len = sizeof(struct FEDiskEntry);
2343 for (cbi = fe->firstcb, cb = itocb(cbi);
2345 cbi = cb->cnext, cb = itocb(cbi), hdr.nCBs++) {
2346 if (cbi > state->cb_hdr->cb_max) {
2347 state->cb_hdr->cb_max = cbi;
2349 if (cb_stateCBToDiskEntry(cb, &cbdsk[iovcnt])) {
2353 cbdsk[iovcnt].index = cbi;
2354 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2355 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2357 if ((iovcnt == 16) || (!cb->cnext)) {
2358 if (fs_stateWriteV(state, iov, iovcnt)) {
2367 hdr.magic = CALLBACK_STATE_ENTRY_MAGIC;
2368 hdr.len = sizeof(hdr) + sizeof(struct FEDiskEntry) +
2369 (hdr.nCBs * sizeof(struct CBDiskEntry));
2372 if (fs_stateWriteV(state, iov, iovcnt)) {
2377 if (fs_stateWriteHeader(state, &state->eof_offset, &hdr, sizeof(hdr))) {
2383 fs_stateIncEOF(state, hdr.len);
2386 if (fs_stateSeek(state, &state->eof_offset)) {
2392 state->cb_hdr->nFEs++;
2393 state->cb_hdr->nCBs += hdr.nCBs;
2400 cb_stateRestoreFE(struct fs_dump_state * state)
2402 int ret = 0, iovcnt, nCBs;
2403 struct callback_state_entry_header hdr;
2404 struct FEDiskEntry fedsk;
2405 struct CBDiskEntry cbdsk[16];
2406 struct iovec iov[16];
2407 struct FileEntry * fe;
2409 iov[0].iov_base = (char *)&hdr;
2410 iov[0].iov_len = sizeof(hdr);
2411 iov[1].iov_base = (char *)&fedsk;
2412 iov[1].iov_len = sizeof(fedsk);
2415 if (fs_stateReadV(state, iov, iovcnt)) {
2420 if (hdr.magic != CALLBACK_STATE_ENTRY_MAGIC) {
2427 ViceLog(0, ("cb_stateRestoreFE: ran out of free FileEntry structures\n"));
2432 if (cb_stateDiskEntryToFE(state, &fedsk, fe)) {
2438 for (iovcnt = 0, nCBs = 0;
2441 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2442 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2444 if ((iovcnt == 16) || (nCBs == hdr.nCBs - 1)) {
2445 if (fs_stateReadV(state, iov, iovcnt)) {
2449 if (cb_stateRestoreCBs(state, fe, iov, iovcnt)) {
2463 cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
2464 struct iovec * iov, int niovecs)
2467 struct CallBack * cb;
2468 struct CBDiskEntry * cbdsk;
2470 for (idx = 0; idx < niovecs; idx++) {
2471 cbdsk = (struct CBDiskEntry *) iov[idx].iov_base;
2473 if (cbdsk->cb.hhead < state->h_map.len &&
2474 state->h_map.entries[cbdsk->cb.hhead].valid == FS_STATE_IDX_SKIPPED) {
2478 if ((cb = GetCB()) == NULL) {
2479 ViceLog(0, ("cb_stateRestoreCBs: ran out of free CallBack structures\n"));
2483 if (cb_stateDiskEntryToCB(state, cbdsk, cb)) {
2484 ViceLog(0, ("cb_stateRestoreCBs: corrupt CallBack disk entry\n"));
2496 cb_stateFillHeader(struct callback_state_header * hdr)
2498 hdr->stamp.magic = CALLBACK_STATE_MAGIC;
2499 hdr->stamp.version = CALLBACK_STATE_VERSION;
2500 hdr->tfirst = tfirst;
2505 cb_stateCheckHeader(struct callback_state_header * hdr)
2509 if (hdr->stamp.magic != CALLBACK_STATE_MAGIC) {
2511 } else if (hdr->stamp.version != CALLBACK_STATE_VERSION) {
2513 } else if ((hdr->nFEs > cbstuff.nblks) || (hdr->nCBs > cbstuff.nblks)) {
2514 ViceLog(0, ("cb_stateCheckHeader: saved callback state larger than callback memory allocation\n"));
2520 /* disk entry conversion routines */
2522 cb_stateFEToDiskEntry(struct FileEntry * in, struct FEDiskEntry * out)
2524 memcpy(&out->fe, in, sizeof(struct FileEntry));
2525 out->index = fetoi(in);
2530 cb_stateDiskEntryToFE(struct fs_dump_state * state,
2531 struct FEDiskEntry * in, struct FileEntry * out)
2535 memcpy(out, &in->fe, sizeof(struct FileEntry));
2537 /* setup FE map entry */
2538 if (!in->index || (in->index >= state->fe_map.len)) {
2539 ViceLog(0, ("cb_stateDiskEntryToFE: index (%d) out of range",
2544 state->fe_map.entries[in->index].valid = FS_STATE_IDX_VALID;
2545 state->fe_map.entries[in->index].old_idx = in->index;
2546 state->fe_map.entries[in->index].new_idx = fetoi(out);
2553 cb_stateCBToDiskEntry(struct CallBack * in, struct CBDiskEntry * out)
2555 memcpy(&out->cb, in, sizeof(struct CallBack));
2556 out->index = cbtoi(in);
2561 cb_stateDiskEntryToCB(struct fs_dump_state * state,
2562 struct CBDiskEntry * in, struct CallBack * out)
2566 memcpy(out, &in->cb, sizeof(struct CallBack));
2568 /* setup CB map entry */
2569 if (!in->index || (in->index >= state->cb_map.len)) {
2570 ViceLog(0, ("cb_stateDiskEntryToCB: index (%d) out of range\n",
2575 state->cb_map.entries[in->index].valid = FS_STATE_IDX_VALID;
2576 state->cb_map.entries[in->index].old_idx = in->index;
2577 state->cb_map.entries[in->index].new_idx = cbtoi(out);
2583 /* index map routines */
2585 cb_stateAllocMap(struct fs_dump_state * state)
2587 state->fe_map.len = state->cb_hdr->fe_max + 1;
2588 state->cb_map.len = state->cb_hdr->cb_max + 1;
2589 state->fe_map.entries = (struct idx_map_entry_t *)
2590 calloc(state->fe_map.len, sizeof(struct idx_map_entry_t));
2591 state->cb_map.entries = (struct idx_map_entry_t *)
2592 calloc(state->cb_map.len, sizeof(struct idx_map_entry_t));
2593 return ((state->fe_map.entries != NULL) && (state->cb_map.entries != NULL)) ? 0 : 1;
2597 fe_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2601 /* FEs use a one-based indexing system, so old==0 implies no mapping */
2607 if (old >= state->fe_map.len) {
2608 ViceLog(0, ("fe_OldToNew: index %d is out of range\n", old));
2610 } else if (state->fe_map.entries[old].valid != FS_STATE_IDX_VALID ||
2611 state->fe_map.entries[old].old_idx != old) { /* sanity check */
2612 ViceLog(0, ("fe_OldToNew: index %d points to an invalid FileEntry record\n", old));
2615 *new = state->fe_map.entries[old].new_idx;
2623 cb_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2627 /* CBs use a one-based indexing system, so old==0 implies no mapping */
2633 if (old >= state->cb_map.len) {
2634 ViceLog(0, ("cb_OldToNew: index %d is out of range\n", old));
2636 } else if (state->cb_map.entries[old].valid != FS_STATE_IDX_VALID ||
2637 state->cb_map.entries[old].old_idx != old) { /* sanity check */
2638 ViceLog(0, ("cb_OldToNew: index %d points to an invalid CallBack record\n", old));
2641 *new = state->cb_map.entries[old].new_idx;
2647 #endif /* AFS_DEMAND_ATTACH_FS */
2649 #define DumpBytes(fd,buf,req) if (write(fd, buf, req) < 0) ; /* don't care */
2652 DumpCallBackState_r(void)
2655 afs_uint32 magic = MAGICV2, now = (afs_int32) time(NULL), freelisthead;
2657 oflag = O_WRONLY | O_CREAT | O_TRUNC;
2661 fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, oflag, 0666);
2664 ("Couldn't create callback dump file %s\n",
2665 AFSDIR_SERVER_CBKDUMP_FILEPATH));
2669 * Collect but ignoring the return value of write(2) here,
2670 * to avoid compiler warnings on some platforms.
2672 DumpBytes(fd, &magic, sizeof(magic));
2673 DumpBytes(fd, &now, sizeof(now));
2674 DumpBytes(fd, &cbstuff, sizeof(cbstuff));
2675 DumpBytes(fd, TimeOuts, sizeof(TimeOuts));
2676 DumpBytes(fd, timeout, sizeof(timeout));
2677 DumpBytes(fd, &tfirst, sizeof(tfirst));
2678 freelisthead = cbtoi((struct CallBack *)CBfree);
2679 DumpBytes(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2680 freelisthead = fetoi((struct FileEntry *)FEfree);
2681 DumpBytes(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2682 DumpBytes(fd, HashTable, sizeof(HashTable));
2683 DumpBytes(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2684 DumpBytes(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2691 DumpCallBackState(void) {
2695 rc = DumpCallBackState_r();
2701 #endif /* !INTERPRET_DUMP */
2703 #ifdef INTERPRET_DUMP
2706 ReadBytes(int fd, void *buf, size_t req)
2710 count = read(fd, buf, req);
2714 } else if (count != req) {
2715 fprintf(stderr, "read: premature EOF (expected %lu, got %lu)\n",
2716 (unsigned long)req, (unsigned long)count);
2721 /* This is only compiled in for the callback analyzer program */
2722 /* Returns the time of the dump */
2724 ReadDump(char *file, int timebits)
2727 afs_uint32 magic, freelisthead;
2735 fd = open(file, oflag);
2737 fprintf(stderr, "Couldn't read dump file %s\n", file);
2740 ReadBytes(fd, &magic, sizeof(magic));
2741 if (magic == MAGICV2) {
2744 if (magic != MAGIC) {
2746 "Magic number of %s is invalid. You might be trying to\n",
2749 "run this program on a machine type with a different byte ordering.\n");
2753 if (timebits == 64) {
2754 ReadBytes(fd, &now64, sizeof(afs_int64));
2755 now = (afs_int32) now64;
2757 ReadBytes(fd, &now, sizeof(afs_int32));
2759 ReadBytes(fd, &cbstuff, sizeof(cbstuff));
2760 ReadBytes(fd, TimeOuts, sizeof(TimeOuts));
2761 ReadBytes(fd, timeout, sizeof(timeout));
2762 ReadBytes(fd, &tfirst, sizeof(tfirst));
2763 ReadBytes(fd, &freelisthead, sizeof(freelisthead));
2764 CB = ((struct CallBack
2765 *)(calloc(cbstuff.nblks, sizeof(struct CallBack)))) - 1;
2766 FE = ((struct FileEntry
2767 *)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
2768 CBfree = (struct CallBack *)itocb(freelisthead);
2769 ReadBytes(fd, &freelisthead, sizeof(freelisthead));
2770 FEfree = (struct FileEntry *)itofe(freelisthead);
2771 ReadBytes(fd, HashTable, sizeof(HashTable));
2772 ReadBytes(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2773 ReadBytes(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2775 perror("Error reading dumpfile");
2782 #include "AFS_component_version_number.h"
2784 #include "AFS_component_version_number.c"
2787 static afs_uint32 *cbTrack;
2790 main(int argc, char **argv)
2792 int err = 0, cbi = 0, stats = 0, noptions = 0, all = 0, vol = 0, raw = 0;
2794 struct FileEntry *fe;
2795 struct CallBack *cb;
2799 memset(&fid, 0, sizeof(fid));
2802 while (argc && **argv == '-') {
2805 if (!strcmp(*argv, "-host")) {
2811 cbi = atoi(*++argv);
2812 } else if (!strcmp(*argv, "-fid")) {
2818 fid.Volume = atoi(*++argv);
2819 fid.Vnode = atoi(*++argv);
2820 fid.Unique = atoi(*++argv);
2821 } else if (!strcmp(*argv, "-time")) {
2822 fprintf(stderr, "-time not supported\n");
2824 } else if (!strcmp(*argv, "-stats")) {
2826 } else if (!strcmp(*argv, "-all")) {
2828 } else if (!strcmp(*argv, "-raw")) {
2830 } else if (!strcmp(*argv, "-timebits")) {
2836 timebits = atoi(*++argv);
2837 if ((timebits != 32)
2841 } else if (!strcmp(*argv, "-volume")) {
2847 vol = atoi(*++argv);
2852 if (err || argc != 1) {
2854 "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] [-timebits 32"
2856 "] callbackdumpfile\n");
2858 "[cbid is shown for each host in the hosts.dump file]\n");
2861 now = ReadDump(*argv, timebits);
2862 if (stats || noptions == 0) {
2863 time_t uxtfirst = UXtime(tfirst), tnow = now;
2864 printf("The time of the dump was %u %s", (unsigned int) now, ctime(&tnow));
2865 printf("The last time cleanup ran was %u %s", (unsigned int) uxtfirst,
2867 PrintCallBackStats();
2870 cbTrack = calloc(cbstuff.nblks, sizeof(cbTrack[0]));
2875 struct CallBack *cb;
2876 struct FileEntry *fe;
2878 for (hash = 0; hash < FEHASH_SIZE; hash++) {
2879 for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
2880 if (!vol || (fe->volid == vol)) {
2881 afs_uint32 fe_i = fetoi(fe);
2883 for (cb = itocb(fe->firstcb); cb; cb = itocb(cb->cnext)) {
2884 afs_uint32 cb_i = cbtoi(cb);
2886 if (cb_i > cbstuff.nblks) {
2887 printf("CB index out of range (%u > %d), stopped for this FE\n",
2888 cb_i, cbstuff.nblks);
2892 if (cbTrack[cb_i]) {
2893 printf("CB entry already claimed for FE[%u] (this is FE[%u]), stopped\n",
2894 cbTrack[cb_i], fe_i);
2897 cbTrack[cb_i] = fe_i;
2909 afs_uint32 cfirst = cbi;
2914 } while (cbi != cfirst);
2919 printf("No callback entries for %u.%u\n", fid.Volume, fid.Vnode);
2922 cb = itocb(fe->firstcb);
2925 cb = itocb(cb->cnext);
2930 for (i = 1; i < cbstuff.nblks; i++) {
2931 p = (afs_int32 *) & FE[i];
2932 printf("%d:%12x%12x%12x%12x\n", i, p[0], p[1], p[2], p[3]);
2941 PrintCB(struct CallBack *cb, afs_uint32 now)
2943 struct FileEntry *fe = itofe(cb->fhead);
2944 time_t expires = TIndexToTime(cb->thead);
2949 printf("vol=%" AFS_VOLID_FMT " vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %lu secs at %s",
2950 afs_printable_VolumeId_lu(fe->volid), fe->vnode, fe->ncbs,
2951 cb->hhead, cb->status, fe->status, expires - now, ctime(&expires));
2956 #if !defined(INTERPRET_DUMP)
2958 ** try breaking calbacks on afidp from host. Use multi_rx.
2959 ** return 0 on success, non-zero on failure
2962 MultiBreakCallBackAlternateAddress(struct host *host, struct AFSCBFids *afidp)
2966 retVal = MultiBreakCallBackAlternateAddress_r(host, afidp);
2972 MultiBreakCallBackAlternateAddress_r(struct host *host,
2973 struct AFSCBFids *afidp)
2976 struct rx_connection **conns;
2977 struct rx_connection *connSuccess = 0;
2978 struct AddrPort *interfaces;
2979 static struct rx_securityClass *sc = 0;
2980 static struct AFSCBs tc = { 0, 0 };
2983 /* nothing more can be done */
2984 if (!host->interface)
2985 return 1; /* failure */
2987 /* the only address is the primary interface */
2988 if (host->interface->numberOfInterfaces <= 1)
2989 return 1; /* failure */
2991 /* initialise a security object only once */
2993 sc = rxnull_NewClientSecurityObject();
2995 i = host->interface->numberOfInterfaces;
2996 interfaces = calloc(i, sizeof(struct AddrPort));
2997 conns = calloc(i, sizeof(struct rx_connection *));
2998 if (!interfaces || !conns) {
2999 ViceLogThenPanic(0, ("Failed malloc in "
3000 "MultiBreakCallBackAlternateAddress_r\n"));
3003 /* initialize alternate rx connections */
3004 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3005 /* this is the current primary address */
3006 if (host->host == host->interface->interface[i].addr &&
3007 host->port == host->interface->interface[i].port)
3010 interfaces[j] = host->interface->interface[i];
3012 rx_NewConnection(interfaces[j].addr,
3013 interfaces[j].port, 1, sc, 0);
3014 rx_SetConnDeadTime(conns[j], 2);
3015 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3019 opr_Assert(j); /* at least one alternate address */
3021 ("Starting multibreakcall back on all addr for host %p (%s:%d)\n",
3022 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
3024 multi_Rx(conns, j) {
3025 multi_RXAFSCB_CallBack(afidp, &tc);
3029 if (host->callback_rxcon)
3030 rx_DestroyConnection(host->callback_rxcon);
3031 host->callback_rxcon = conns[multi_i];
3032 /* add then remove */
3033 addInterfaceAddr_r(host, interfaces[multi_i].addr,
3034 interfaces[multi_i].port);
3035 removeInterfaceAddr_r(host, host->host, host->port);
3036 host->host = interfaces[multi_i].addr;
3037 host->port = interfaces[multi_i].port;
3038 connSuccess = conns[multi_i];
3039 rx_SetConnDeadTime(host->callback_rxcon, 50);
3040 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3042 ("multibreakcall success with addr %s:%d\n",
3043 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3044 ntohs(interfaces[multi_i].port)));
3051 /* Destroy all connections except the one on which we succeeded */
3052 for (i = 0; i < j; i++)
3053 if (conns[i] != connSuccess)
3054 rx_DestroyConnection(conns[i]);
3060 return 0; /* success */
3062 return 1; /* failure */
3067 ** try multi_RX probes to host.
3068 ** return 0 on success, non-0 on failure
3071 MultiProbeAlternateAddress_r(struct host *host)
3074 struct rx_connection **conns;
3075 struct rx_connection *connSuccess = 0;
3076 struct AddrPort *interfaces;
3077 static struct rx_securityClass *sc = 0;
3080 /* nothing more can be done */
3081 if (!host->interface)
3082 return 1; /* failure */
3084 /* the only address is the primary interface */
3085 if (host->interface->numberOfInterfaces <= 1)
3086 return 1; /* failure */
3088 /* initialise a security object only once */
3090 sc = rxnull_NewClientSecurityObject();
3092 i = host->interface->numberOfInterfaces;
3093 interfaces = calloc(i, sizeof(struct AddrPort));
3094 conns = calloc(i, sizeof(struct rx_connection *));
3095 if (!interfaces || !conns) {
3096 ViceLogThenPanic(0, ("Failed malloc in "
3097 "MultiProbeAlternateAddress_r\n"));
3100 /* initialize alternate rx connections */
3101 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3102 /* this is the current primary address */
3103 if (host->host == host->interface->interface[i].addr &&
3104 host->port == host->interface->interface[i].port)
3107 interfaces[j] = host->interface->interface[i];
3109 rx_NewConnection(interfaces[j].addr,
3110 interfaces[j].port, 1, sc, 0);
3111 rx_SetConnDeadTime(conns[j], 2);
3112 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3116 opr_Assert(j); /* at least one alternate address */
3118 ("Starting multiprobe on all addr for host %p (%s:%d)\n",
3119 host, afs_inet_ntoa_r(host->host, hoststr),
3120 ntohs(host->port)));
3122 multi_Rx(conns, j) {
3123 multi_RXAFSCB_ProbeUuid(&host->interface->uuid);
3127 if (host->callback_rxcon)
3128 rx_DestroyConnection(host->callback_rxcon);
3129 host->callback_rxcon = conns[multi_i];
3130 /* add then remove */
3131 addInterfaceAddr_r(host, interfaces[multi_i].addr,
3132 interfaces[multi_i].port);
3133 removeInterfaceAddr_r(host, host->host, host->port);
3134 host->host = interfaces[multi_i].addr;
3135 host->port = interfaces[multi_i].port;
3136 connSuccess = conns[multi_i];
3137 rx_SetConnDeadTime(host->callback_rxcon, 50);
3138 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3140 ("multiprobe success with addr %s:%d\n",
3141 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3142 ntohs(interfaces[multi_i].port)));
3147 ("multiprobe failure with addr %s:%d\n",
3148 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3149 ntohs(interfaces[multi_i].port)));
3151 /* This is less than desirable but its the best we can do.
3152 * The AFS Cache Manager will return either 0 for a Uuid
3153 * match and a 1 for a non-match. If the error is 1 we
3154 * therefore know that our mapping of IP address to Uuid
3155 * is wrong. We should attempt to find the correct
3156 * Uuid and fix the host tables.
3158 if (multi_error == 1) {
3159 /* remove the current alternate address from this host */
3161 removeInterfaceAddr_r(host, interfaces[multi_i].addr, interfaces[multi_i].port);
3165 #ifdef AFS_DEMAND_ATTACH_FS
3166 /* try to bail ASAP if the fileserver is shutting down */
3168 if (fs_state.mode == FS_MODE_SHUTDOWN) {
3177 /* Destroy all connections except the one on which we succeeded */
3178 for (i = 0; i < j; i++)
3179 if (conns[i] != connSuccess)
3180 rx_DestroyConnection(conns[i]);
3186 return 0; /* success */
3188 return 1; /* failure */
3191 #endif /* !defined(INTERPRET_DUMP) */