/*
* Copyright 2000, International Business Machines Corporation and others.
* All Rights Reserved.
- *
+ *
* This software has been released under the terms of the IBM Public
* License. For details, see the LICENSE file in the top-level source
* directory or online at http://www.openafs.org/dl/license10.html
+ *
+ * Portions Copyright (c) 2006 Sine Nomine Associates
*/
/*
* NEW callback package callback.c (replaces vicecb.c)
* Updated call back routines, NOW with:
- *
+ *
* Faster DeleteVenus (Now called DeleteAllCallBacks)
* Call back breaking for volumes
* Adaptive timeouts on call backs
* nblocks must be < 65536
* Space used is nblocks*16 bytes
* Note that space will be reclaimed by breaking callbacks of old hosts
- *
+ *
* time = AddCallBack(host, fid)
* Add a call back.
* Returns the expiration time at the workstation.
- *
+ *
* BreakCallBack(host, fid)
* Break all call backs for fid, except for the specified host.
* Delete all of them.
- *
- * BreakVolumeCallBacks(volume)
+ *
+ * BreakVolumeCallBacksLater(volume)
* Break all call backs on volume, using single call to each host
* Delete all the call backs.
- *
+ *
* DeleteCallBack(host,fid)
* Delete (do not break) single call back for fid.
- *
+ *
* DeleteFileCallBacks(fid)
* Delete (do not break) all call backs for fid.
*
* DeleteAllCallBacks(host)
* Delete (do not break) all call backs for host.
- *
+ *
* CleanupTimedOutCallBacks()
* Delete all timed out call back entries
* Must be called periodically by file server.
- *
+ *
* BreakDelayedCallBacks(host)
* Break all delayed call backs for host.
* Returns 1: one or more failed, 0: success.
- *
+ *
* PrintCallBackStats()
* Print statistics about call backs to stdout.
- *
+ *
* DumpCallBacks() ---wishful thinking---
* Dump call back state to /tmp/callback.state.
* This is separately interpretable by the program pcb.
#include <afsconfig.h>
#include <afs/param.h>
+#include <afs/stds.h>
-RCSID
- ("$Header$");
+#include <roken.h>
-#include <stdio.h>
-#include <stdlib.h> /* for malloc() */
-#include <time.h> /* ANSI standard location for time stuff */
-#ifdef AFS_NT40_ENV
-#include <fcntl.h>
-#include <io.h>
-#else
-#include <sys/time.h>
+#ifdef HAVE_SYS_FILE_H
#include <sys/file.h>
#endif
-#ifdef HAVE_STRING_H
-#include <string.h>
-#else
-#ifdef HAVE_STRINGS_H
-#include <strings.h>
-#endif
-#endif
-#include <afs/assert.h>
-
-#include <afs/stds.h>
+#include <afs/opr.h>
+#include <opr/lock.h>
#include <afs/nfs.h> /* yuck. This is an abomination. */
-#include <lwp.h>
#include <rx/rx.h>
-#include <afscbint.h>
+#include <rx/rx_queue.h>
+#include <afs/afscbint.h>
#include <afs/afsutil.h>
-#include <lock.h>
#include <afs/ihandle.h>
+#include <afs/partition.h>
#include <afs/vnode.h>
#include <afs/volume.h>
#include "viced_prototypes.h"
#include <afs/ptclient.h> /* need definition of prlist for host.h */
#include "host.h"
+#include "callback.h"
+#ifdef AFS_DEMAND_ATTACH_FS
+#include "serialize_state.h"
+#endif /* AFS_DEMAND_ATTACH_FS */
+
extern afsUUID FS_HostUUID;
extern int hostCount;
-int ShowProblems = 1;
-
-/* Maximum number of call backs to break at once, single fid */
-/* There is some debate as to just how large this value should be */
-/* Ideally, it would be very very large, but I am afraid that the */
-/* cache managers will all send in their responses simultaneously, */
-/* thereby swamping the file server. As a result, something like */
-/* 10 or 15 might be a better bet. */
-#define MAX_CB_HOSTS 10
-/* max time to break a callback, otherwise client is dead or net is hosed */
-#define MAXCBT 25
-
-#define u_byte unsigned char
+#ifndef INTERPRET_DUMP
+static int ShowProblems = 1;
+#endif
struct cbcounters cbstuff;
-struct cbstruct {
- struct host *hp;
- afs_uint32 thead;
-};
-
-struct FileEntry {
- afs_uint32 vnode;
- afs_uint32 unique;
- afs_uint32 volid;
- afs_uint32 fnext;
- afs_uint32 ncbs;
- afs_uint32 firstcb;
- afs_uint32 status;
- afs_uint32 spare;
-} *FE; /* Don't use FE[0] */
-#define FE_LATER 0x1
-
-struct CallBack {
- afs_uint32 cnext; /* Next call back entry */
- afs_uint32 fhead; /* Head of this call back chain */
- u_byte thead; /* Head of timeout chain */
- u_byte status; /* Call back status; see definitions, below */
- afs_uint32 hhead; /* Head of host table chain */
- afs_uint32 tprev, tnext; /* Timeout chain */
- afs_uint32 hprev, hnext; /* Chain from host table */
- unsigned short spare; /* make it a multiple of 32 bits. */
-} *CB; /* Don't use CB[0] */
-
-/* status values for status field of CallBack structure */
-#define CB_NORMAL 1 /* Normal call back */
-#define CB_DELAYED 2 /* Delayed call back due to rpc problems.
- * The call back entry will be added back to the
- * host list at the END of the list, so that
- * searching backwards in the list will find all
- * the (consecutive)host. delayed call back entries */
-#define CB_VOLUME 3 /* Callback for a volume */
-#define CB_BULK 4 /* Normal callbacks, handed out from FetchBulkStatus */
-
-/* call back indices to pointers, and vice-versa */
-#define itocb(i) ((i)?CB+(i):0)
-#define cbtoi(cbp) (!(cbp)?0:(cbp)-CB)
-
-/* file entry indices to pointers, and vice-versa */
-#define itofe(i) ((i)?FE+(i):0)
-#define fetoi(fep) (!(fep)?0:(fep)-FE)
-
-/* Timeouts: there are 128 possible timeout values in effect at any
- * given time. Each timeout represents timeouts in an interval of 128
- * seconds. So the maximum timeout for a call back is 128*128=16384
- * seconds, or 4 1/2 hours. The timeout cleanup stuff is called only
- * if space runs out or by the file server every 5 minutes. This 5
- * minute slack should be allowed for--so a maximum time of 4 hours
- * is safer.
- *
- * Timeouts must be chosen to correspond to an exact multiple
- * of 128, because all times are truncated to a 128 multiple, and
- * timed out if the current truncated time is <= to the truncated time
- * corresponding to the timeout queue.
- */
+static struct FileEntry * FE = NULL; /* don't use FE[0] */
+static struct CallBack * CB = NULL; /* don't use CB[0] */
-/* Unix time to Call Back time, and vice-versa. Call back time is
- in units of 128 seconds, corresponding to time queues. */
-#define CBtime(uxtime) ((uxtime)>>7)
-#define UXtime(cbtime) ((cbtime)<<7)
+static struct CallBack * CBfree = NULL;
+static struct FileEntry * FEfree = NULL;
-/* Given a Unix time, compute the closest Unix time that corresponds to
- a time queue, rounding up */
-#define TimeCeiling(uxtime) (((uxtime)+127)&~127)
/* Time to live for call backs depends upon number of users of the file.
* TimeOuts is indexed by this number/8 (using TimeOut macro). Times
}; /* Anything more: MinTimeOut */
/* minimum time given for a call back */
+#ifndef INTERPRET_DUMP
static int MinTimeOut = (7 * 60);
-
-#define TimeOutCutoff ((sizeof(TimeOuts)/sizeof(TimeOuts[0]))*8)
-#define TimeOut(nusers) ((nusers)>=TimeOutCutoff? MinTimeOut: TimeOuts[(nusers)>>3])
-
-/* time out at server is 3 minutes more than ws */
-#define ServerBias (3*60)
+#endif
/* Heads of CB queues; a timeout index is 1+index into this array */
-static afs_uint32 timeout[128];
-
-/* Convert cbtime to timeout queue index */
-#define TIndex(cbtime) (((cbtime)&127)+1)
-
-/* Convert cbtime to pointer to timeout queue head */
-#define THead(cbtime) (&timeout[TIndex(cbtime)-1])
+static afs_uint32 timeout[CB_NUM_TIMEOUT_QUEUES];
static afs_int32 tfirst; /* cbtime of oldest unexpired call back time queue */
-/* Normalize index into timeout array so that two such indices will be
- ordered correctly, so that they can be compared to see which times
- sooner, or so that the difference in time out times between them
- can be computed. */
-#define TNorm(index) ((index)<TIndex(tfirst)?(index)+128:(index))
-
-/* This converts a timeout index into the actual time it will expire */
-#define TIndexToTime(index) (UXtime(TNorm(index) - TIndex(tfirst) + tfirst))
-
-
-/* Convert pointer to timeout queue head to index, and vice versa */
-#define ttoi(t) ((t-timeout)+1)
-#define itot(i) ((timeout)+(i-1))
/* 16 byte object get/free routines */
struct object {
struct object *next;
};
-struct VCBParams {
- struct cbstruct cba[MAX_CB_HOSTS]; /* re-entrant storage */
- unsigned int ncbas;
- afs_uint32 thead; /* head of timeout queue for youngest callback */
- struct AFSFid *fid;
-};
-
-struct CallBack *CBfree = 0;
-struct FileEntry *FEfree = 0;
-
/* Prototypes for static routines */
-static struct FileEntry *FindFE(register AFSFid * fid);
-static struct CallBack *iGetCB(register int *nused);
-static int iFreeCB(register struct CallBack *cb, register int *nused);
-static struct FileEntry *iGetFE(register int *nused);
-static int iFreeFE(register struct FileEntry *fe, register int *nused);
-static int TAdd(register struct CallBack *cb, register afs_uint32 * thead);
-static int TDel(register struct CallBack *cb);
-static int HAdd(register struct CallBack *cb, register struct host *host);
-static int HDel(register struct CallBack *cb);
+static struct FileEntry *FindFE(AFSFid * fid);
+
+#ifndef INTERPRET_DUMP
+static struct CallBack *iGetCB(int *nused);
+static int iFreeCB(struct CallBack *cb, int *nused);
+static struct FileEntry *iGetFE(int *nused);
+static int iFreeFE(struct FileEntry *fe, int *nused);
+static int TAdd(struct CallBack *cb, afs_uint32 * thead);
+static int TDel(struct CallBack *cb);
+static int HAdd(struct CallBack *cb, struct host *host);
+static int HDel(struct CallBack *cb);
static int CDel(struct CallBack *cb, int deletefe);
-static int CDelPtr(register struct FileEntry *fe, register afs_uint32 * cbp,
+static int CDelPtr(struct FileEntry *fe, afs_uint32 * cbp,
int deletefe);
static afs_uint32 *FindCBPtr(struct FileEntry *fe, struct host *host);
-static int FDel(register struct FileEntry *fe);
+static int FDel(struct FileEntry *fe);
static int AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead,
int type, int locked);
static void MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
- struct AFSCBFids *afidp, struct host *xhost);
-static int MultiBreakVolumeCallBack_r(struct host *host, int isheld,
+ struct AFSCBFids *afidp);
+static int MultiBreakVolumeCallBack_r(struct host *host,
struct VCBParams *parms, int deletefe);
-static int MultiBreakVolumeCallBack(struct host *host, int isheld,
- struct VCBParams *parms);
-static int MultiBreakVolumeLaterCallBack(struct host *host, int isheld,
- struct VCBParams *parms);
-static int lih_r(register struct host *host, register int held,
- register struct host *hostp);
+static int MultiBreakVolumeLaterCallBack(struct host *host, void *rock);
static int GetSomeSpace_r(struct host *hostp, int locked);
static int ClearHostCallbacks_r(struct host *hp, int locked);
+static int DumpCallBackState_r(void);
+#endif
#define GetCB() ((struct CallBack *)iGetCB(&cbstuff.nCBs))
#define GetFE() ((struct FileEntry *)iGetFE(&cbstuff.nFEs))
#define FreeCB(cb) iFreeCB((struct CallBack *)cb, &cbstuff.nCBs)
#define FreeFE(fe) iFreeFE((struct FileEntry *)fe, &cbstuff.nFEs)
+
/* Other protos - move out sometime */
-extern void ShutDown();
+void PrintCB(struct CallBack *cb, afs_uint32 now);
-#define VHASH 512 /* Power of 2 */
-static afs_uint32 HashTable[VHASH]; /* File entry hash table */
-#define VHash(volume, unique) (((volume)+(unique))&(VHASH-1))
+static afs_uint32 HashTable[FEHASH_SIZE]; /* File entry hash table */
static struct FileEntry *
-FindFE(register AFSFid * fid)
+FindFE(AFSFid * fid)
{
int hash;
- register int fei;
- register struct FileEntry *fe;
+ int fei;
+ struct FileEntry *fe;
- hash = VHash(fid->Volume, fid->Unique);
+ hash = FEHash(fid->Volume, fid->Unique);
for (fei = HashTable[hash]; fei; fei = fe->fnext) {
fe = itofe(fei);
if (fe->volid == fid->Volume && fe->unique == fid->Unique
- && fe->vnode == fid->Vnode)
+ && fe->vnode == fid->Vnode && (fe->status & FE_LATER) != FE_LATER)
return fe;
}
return 0;
#ifndef INTERPRET_DUMP
static struct CallBack *
-iGetCB(register int *nused)
+iGetCB(int *nused)
{
- register struct CallBack *ret;
+ struct CallBack *ret;
if ((ret = CBfree)) {
CBfree = (struct CallBack *)(((struct object *)ret)->next);
}
static int
-iFreeCB(register struct CallBack *cb, register int *nused)
+iFreeCB(struct CallBack *cb, int *nused)
{
((struct object *)cb)->next = (struct object *)CBfree;
CBfree = cb;
}
static struct FileEntry *
-iGetFE(register int *nused)
+iGetFE(int *nused)
{
- register struct FileEntry *ret;
+ struct FileEntry *ret;
if ((ret = FEfree)) {
FEfree = (struct FileEntry *)(((struct object *)ret)->next);
}
static int
-iFreeFE(register struct FileEntry *fe, register int *nused)
+iFreeFE(struct FileEntry *fe, int *nused)
{
((struct object *)fe)->next = (struct object *)FEfree;
FEfree = fe;
/* Add cb to end of specified timeout list */
static int
-TAdd(register struct CallBack *cb, register afs_uint32 * thead)
+TAdd(struct CallBack *cb, afs_uint32 * thead)
{
if (!*thead) {
(*thead) = cb->tnext = cb->tprev = cbtoi(cb);
} else {
- register struct CallBack *thp = itocb(*thead);
+ struct CallBack *thp = itocb(*thead);
cb->tprev = thp->tprev;
cb->tnext = *thead;
/* Delete call back entry from timeout list */
static int
-TDel(register struct CallBack *cb)
+TDel(struct CallBack *cb)
{
- register afs_uint32 *thead = itot(cb->thead);
+ afs_uint32 *thead = itot(cb->thead);
if (*thead == cbtoi(cb))
*thead = (*thead == cb->tnext ? 0 : cb->tnext);
/* Add cb to end of specified host list */
static int
-HAdd(register struct CallBack *cb, register struct host *host)
+HAdd(struct CallBack *cb, struct host *host)
{
cb->hhead = h_htoi(host);
if (!host->cblist) {
host->cblist = cb->hnext = cb->hprev = cbtoi(cb);
} else {
- register struct CallBack *hhp = itocb(host->cblist);
+ struct CallBack *fcb = itocb(host->cblist);
- cb->hprev = hhp->hprev;
- cb->hnext = host->cblist;
- hhp->hprev = (itocb(hhp->hprev)->hnext = cbtoi(cb));
+ cb->hprev = fcb->hprev;
+ cb->hnext = cbtoi(fcb);
+ fcb->hprev = (itocb(fcb->hprev)->hnext = cbtoi(cb));
}
return 0;
}
/* Delete call back entry from host list */
static int
-HDel(register struct CallBack *cb)
+HDel(struct CallBack *cb)
{
- register afs_uint32 *hhead = &h_itoh(cb->hhead)->cblist;
+ afs_uint32 *hhead = &h_itoh(cb->hhead)->cblist;
if (*hhead == cbtoi(cb))
*hhead = (*hhead == cb->hnext ? 0 : cb->hnext);
{
int cbi = cbtoi(cb);
struct FileEntry *fe = itofe(cb->fhead);
- register afs_uint32 *cbp;
- register int safety;
+ afs_uint32 *cbp;
+ int safety;
for (safety = 0, cbp = &fe->firstcb; *cbp && *cbp != cbi;
cbp = &itocb(*cbp)->cnext, safety++) {
if (safety > cbstuff.nblks + 10) {
- assert(0);
- ViceLog(0,
- ("CDel: Internal Error -- shutting down: wanted %d from %d, now at %d\n",
- cbi, fe->firstcb, *cbp));
- DumpCallBackState();
+ ViceLogThenPanic(0, ("CDel: Internal Error -- shutting down: "
+ "wanted %d from %d, now at %d\n",
+ cbi, fe->firstcb, *cbp));
+ DumpCallBackState_r();
ShutDownAndCore(PANIC);
}
}
/* N.B. This one also deletes the CB, and also possibly parent FE, so
* make sure that it is not on any other list before calling this
* routine */
-int Ccdelpt = 0, CcdelB = 0;
+static int Ccdelpt = 0, CcdelB = 0;
static int
-CDelPtr(register struct FileEntry *fe, register afs_uint32 * cbp,
+CDelPtr(struct FileEntry *fe, afs_uint32 * cbp,
int deletefe)
{
- register struct CallBack *cb;
+ struct CallBack *cb;
if (!*cbp)
return 0;
CcdelB++;
*cbp = cb->cnext;
FreeCB(cb);
- if (deletefe && (--fe->ncbs == 0))
+ if ((--fe->ncbs == 0) && deletefe)
FDel(fe);
return 0;
}
static afs_uint32 *
FindCBPtr(struct FileEntry *fe, struct host *host)
{
- register afs_uint32 hostindex = h_htoi(host);
- register struct CallBack *cb;
- register afs_uint32 *cbp;
- register int safety;
+ afs_uint32 hostindex = h_htoi(host);
+ struct CallBack *cb;
+ afs_uint32 *cbp;
+ int safety;
for (safety = 0, cbp = &fe->firstcb; *cbp; cbp = &cb->cnext, safety++) {
if (safety > cbstuff.nblks) {
ViceLog(0, ("FindCBPtr: Internal Error -- shutting down.\n"));
- DumpCallBackState();
+ DumpCallBackState_r();
ShutDownAndCore(PANIC);
}
cb = itocb(*cbp);
/* Delete file entry from hash table */
static int
-FDel(register struct FileEntry *fe)
+FDel(struct FileEntry *fe)
{
- register int fei = fetoi(fe);
- register afs_uint32 *p = &HashTable[VHash(fe->volid, fe->unique)];
+ int fei = fetoi(fe);
+ afs_uint32 *p = &HashTable[FEHash(fe->volid, fe->unique)];
while (*p && *p != fei)
p = &itofe(*p)->fnext;
- assert(*p);
+ opr_Assert(*p);
*p = fe->fnext;
FreeFE(fe);
return 0;
}
+/* initialize the callback package */
int
InitCallBack(int nblks)
{
+ opr_Assert(nblks > 0);
+
H_LOCK;
- tfirst = CBtime(FT_ApproxTime());
+ tfirst = CBtime(time(NULL));
/* N.B. The "-1", below, is because
* FE[0] and CB[0] are not used--and not allocated */
- FE = ((struct FileEntry *)(calloc(nblks, sizeof(struct FileEntry)))) - 1;
+ FE = calloc(nblks, sizeof(struct FileEntry));
if (!FE) {
- ViceLog(0, ("Failed malloc in InitCallBack\n"));
- assert(0);
+ ViceLogThenPanic(0, ("Failed malloc in InitCallBack\n"));
}
+ FE--; /* FE[0] is supposed to point to junk */
cbstuff.nFEs = nblks;
while (cbstuff.nFEs)
FreeFE(&FE[cbstuff.nFEs]); /* This is correct */
- CB = ((struct CallBack *)(calloc(nblks, sizeof(struct CallBack)))) - 1;
+ CB = calloc(nblks, sizeof(struct CallBack));
if (!CB) {
- ViceLog(0, ("Failed malloc in InitCallBack\n"));
- assert(0);
+ ViceLogThenPanic(0, ("Failed malloc in InitCallBack\n"));
}
+ CB--; /* CB[0] is supposed to point to junk */
cbstuff.nCBs = nblks;
while (cbstuff.nCBs)
FreeCB(&CB[cbstuff.nCBs]); /* This is correct */
XCallBackBulk_r(struct host * ahost, struct AFSFid * fids, afs_int32 nfids)
{
struct AFSCallBack tcbs[AFSCBMAX];
- register int i;
+ int i;
struct AFSCBFids tf;
struct AFSCBs tc;
int code;
int j;
+ struct rx_connection *cb_conn = NULL;
-#ifdef ADAPT_MTU
rx_SetConnDeadTime(ahost->callback_rxcon, 4);
rx_SetConnHardDeadTime(ahost->callback_rxcon, AFS_HARDDEADTIME);
-#endif
code = 0;
j = 0;
tc.AFSCBs_len = i;
tc.AFSCBs_val = tcbs;
+ cb_conn = ahost->callback_rxcon;
+ rx_GetConnection(cb_conn);
H_UNLOCK;
- code |= RXAFSCB_CallBack(ahost->callback_rxcon, &tf, &tc);
+ code |= RXAFSCB_CallBack(cb_conn, &tf, &tc);
+ rx_PutConnection(cb_conn);
+ cb_conn = NULL;
H_LOCK;
}
return code;
}
-/* the locked flag tells us if the host entry has already been locked
+/* the locked flag tells us if the host entry has already been locked
* by our parent. I don't think anybody actually calls us with the
* host locked, but here's how to make that work: GetSomeSpace has to
* change so that it doesn't attempt to lock any hosts < "host". That
AddCallBack1(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
int locked)
{
- int retVal;
+ int retVal = 0;
H_LOCK;
if (!locked) {
h_Lock_r(host);
}
- retVal = AddCallBack1_r(host, fid, thead, type, 1);
+ if (!(host->hostFlags & HOSTDELETED))
+ retVal = AddCallBack1_r(host, fid, thead, type, 1);
if (!locked) {
h_Unlock_r(host);
struct FileEntry *fe;
struct CallBack *cb = 0, *lastcb = 0;
struct FileEntry *newfe = 0;
- afs_uint32 time_out;
+ afs_uint32 time_out = 0;
afs_uint32 *Thead = thead;
struct CallBack *newcb = 0;
int safety;
+ cbstuff.AddCallBacks++;
+
host->Console |= 2;
/* allocate these guys first, since we can't call the allocator with
- * the host structure locked -- or we might deadlock. However, we have
+ * the host structure locked -- or we might deadlock. However, we have
* to avoid races with FindFE... */
while (!(newcb = GetCB())) {
GetSomeSpace_r(host, locked);
if (!locked) {
h_Lock_r(host); /* this can yield, so do it before we get any */
/* fragile info */
+ if (host->hostFlags & HOSTDELETED) {
+ host->Console &= ~2;
+ h_Unlock_r(host);
+ return 0;
+ }
}
fe = FindFE(fid);
if (type == CB_NORMAL) {
time_out =
- TimeCeiling(FT_ApproxTime() + TimeOut(fe ? fe->ncbs : 0) +
+ TimeCeiling(time(NULL) + TimeOut(fe ? fe->ncbs : 0) +
ServerBias);
Thead = THead(CBtime(time_out));
} else if (type == CB_VOLUME) {
- time_out = TimeCeiling((60 * 120 + FT_ApproxTime()) + ServerBias);
+ time_out = TimeCeiling((60 * 120 + time(NULL)) + ServerBias);
Thead = THead(CBtime(time_out));
} else if (type == CB_BULK) {
/* bulk status can get so many callbacks all at once, and most of them
* are probably not for things that will be used for long.
*/
time_out =
- TimeCeiling(FT_ApproxTime() + ServerBias +
+ TimeCeiling(time(NULL) + ServerBias +
TimeOut(22 + (fe ? fe->ncbs : 0)));
Thead = THead(CBtime(time_out));
}
host->Console &= ~2;
if (!fe) {
- register afs_uint32 hash;
+ afs_uint32 hash;
fe = newfe;
newfe = NULL;
fe->vnode = fid->Vnode;
fe->unique = fid->Unique;
fe->ncbs = 0;
- hash = VHash(fid->Volume, fid->Unique);
+ fe->status = 0;
+ hash = FEHash(fid->Volume, fid->Unique);
fe->fnext = HashTable[hash];
HashTable[hash] = fetoi(fe);
}
lastcb = cb, cb = itocb(cb->cnext), safety++) {
if (safety > cbstuff.nblks) {
ViceLog(0, ("AddCallBack1: Internal Error -- shutting down.\n"));
- DumpCallBackState();
+ DumpCallBackState_r();
ShutDownAndCore(PANIC);
}
if (cb->hhead == h_htoi(host))
TDel(cb);
TAdd(cb, Thead);
}
+ if (newfe == NULL) { /* we are using the new FE */
+ fe->firstcb = cbtoi(cb);
+ fe->ncbs++;
+ cb->fhead = fetoi(fe);
+ }
} else {
cb = newcb;
newcb = NULL;
return 0;
}
-/* Take an array full of hosts, all held. Break callbacks to them, and
- * release the holds once you're done, except don't release xhost. xhost
- * may be NULL. Currently only works for a single Fid in afidp array.
+static int
+CompareCBA(const void *e1, const void *e2)
+{
+ const struct cbstruct *cba1 = (const struct cbstruct *)e1;
+ const struct cbstruct *cba2 = (const struct cbstruct *)e2;
+ return ((cba1->hp)->index - (cba2->hp)->index);
+}
+
+/* Take an array full of hosts, all held. Break callbacks to them, and
+ * release the holds once you're done.
+ * Currently only works for a single Fid in afidp array.
* If you want to make this work with multiple fids, you need to fix
* the error handling. One approach would be to force a reset if a
* multi-fid call fails, or you could add delayed callbacks for each
* fid. You probably also need to sort and remove duplicate hosts.
- * When this is called from the BreakVolumeCallBacks path, it does NOT
- * force a reset if the RPC fails, it just marks the host down and tries
+ * When this is called from the BreakVolumeCallBacks path, it does NOT
+ * force a reset if the RPC fails, it just marks the host down and tries
* to create a delayed callback. */
/* N.B. be sure that code works when ncbas == 0 */
/* N.B. requires all the cba[*].hp pointers to be valid... */
-/* This routine does not hold a lock on the host for the duration of
+/* This routine does not hold a lock on the host for the duration of
* the BreakCallBack RPC, which is a significant deviation from tradition.
* It _does_ get a lock on the host before setting VenusDown = 1,
* which is sufficient only if VenusDown = 0 only happens when the
* wherever that is done. */
static void
MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
- struct AFSCBFids *afidp, struct host *xhost)
+ struct AFSCBFids *afidp)
{
int i, j;
struct rx_connection *conns[MAX_CB_HOSTS];
static struct AFSCBs tc = { 0, 0 };
+ int multi_to_cba_map[MAX_CB_HOSTS];
- assert(ncbas <= MAX_CB_HOSTS);
+ opr_Assert(ncbas <= MAX_CB_HOSTS);
+
+ /*
+ * When we issue a multi_Rx callback break, we must rx_NewCall a call for
+ * each host before we do anything. If there are no call channels
+ * available on the conn, we must wait for one of the existing calls to
+ * finish. If another thread is breaking callbacks at the same time, it is
+ * possible for us to be waiting on NewCall for one of their multi_Rx
+ * CallBack calls to finish, but they are waiting on NewCall for one of
+ * our calls to finish. So we deadlock.
+ *
+ * This can be thought of as similar to obtaining multiple locks at the
+ * same time. So if we establish an ordering, the possibility of deadlock
+ * goes away. Here we provide such an ordering, by sorting our CBAs
+ * according to CompareCBA.
+ */
+ qsort(cba, ncbas, sizeof(struct cbstruct), CompareCBA);
/* set up conns for multi-call */
for (i = 0, j = 0; i < ncbas; i++) {
if (!thishost || (thishost->hostFlags & HOSTDELETED)) {
continue;
}
+ rx_GetConnection(thishost->callback_rxcon);
+ multi_to_cba_map[j] = i;
conns[j++] = thishost->callback_rxcon;
-#ifdef ADAPT_MTU
rx_SetConnDeadTime(thishost->callback_rxcon, 4);
rx_SetConnHardDeadTime(thishost->callback_rxcon, AFS_HARDDEADTIME);
-#endif
}
if (j) { /* who knows what multi would do with 0 conns? */
struct host *hp;
char hoststr[16];
- idx = 0;
- /* If there's an error, we have to hunt for the right host.
- * The conns array _should_ correspond one-to-one to the cba
- * array, except in some rare cases it might be missing one
- * or more elements. So the optimistic case is almost
- * always right. At worst, it's the starting point for the
- * hunt. */
- for (hp = 0, i = multi_i; i < j; i++) {
- hp = cba[i].hp; /* optimistic, but usually right */
- if (!hp) {
- break;
- }
- if (conns[multi_i] == hp->callback_rxcon) {
- idx = cba[i].thead;
- break;
- }
- }
+ i = multi_to_cba_map[multi_i];
+ hp = cba[i].hp;
+ idx = cba[i].thead;
- if (!hp) {
+ if (!hp || !idx) {
ViceLog(0,
- ("BCB: INTERNAL ERROR: hp=%x, cba=%x\n", hp,
- cba));
+ ("BCB: INTERNAL ERROR: hp=%p, cba=%p, thead=%u\n",
+ hp, cba, idx));
} else {
- /*
+ /*
** try breaking callbacks on alternate interface addresses
*/
if (MultiBreakCallBackAlternateAddress(hp, afidp)) {
if (ShowProblems) {
ViceLog(7,
- ("BCB: Failed on file %u.%u.%u, host %s:%d is down\n",
+ ("BCB: Failed on file %u.%u.%u, "
+ "Host %p (%s:%d) is down\n",
afidp->AFSCBFids_val->Volume,
afidp->AFSCBFids_val->Vnode,
afidp->AFSCBFids_val->Unique,
+ hp,
afs_inet_ntoa_r(hp->host, hoststr),
ntohs(hp->port)));
}
H_LOCK;
h_Lock_r(hp);
- hp->hostFlags |= VENUSDOWN;
- /**
- * We always go into AddCallBack1_r with the host locked
- */
- AddCallBack1_r(hp, afidp->AFSCBFids_val, itot(idx),
- CB_DELAYED, 1);
+ if (!(hp->hostFlags & HOSTDELETED)) {
+ hp->hostFlags |= VENUSDOWN;
+ /**
+ * We always go into AddCallBack1_r with the host locked
+ */
+ AddCallBack1_r(hp, afidp->AFSCBFids_val, itot(idx),
+ CB_DELAYED, 1);
+ }
h_Unlock_r(hp);
H_UNLOCK;
}
for (i = 0; i < ncbas; i++) {
struct host *hp;
hp = cba[i].hp;
- if (hp && xhost != hp)
+ if (hp) {
h_Release_r(hp);
+ }
+ }
+
+ /* H_UNLOCK around this so h_FreeConnection does not deadlock.
+ h_FreeConnection should *never* be called on a callback connection,
+ but on 10/27/04 a deadlock occurred where it was, when we know why,
+ this should be reverted. -- shadow */
+ H_UNLOCK;
+ for (i = 0; i < j; i++) {
+ rx_PutConnection(conns[i]);
}
+ H_LOCK;
return;
}
* Break all call backs for fid, except for the specified host (unless flag
* is true, in which case all get a callback message. Assumption: the specified
* host is h_Held, by the caller; the others aren't.
- * Specified host may be bogus, that's ok. This used to check to see if the
- * host was down in two places, once right after the host was h_held, and
+ * Specified host may be bogus, that's ok. This used to check to see if the
+ * host was down in two places, once right after the host was h_held, and
* again after it was locked. That race condition is incredibly rare and
- * relatively harmless even when it does occur, so we don't check for it now.
+ * relatively harmless even when it does occur, so we don't check for it now.
*/
/* if flag is true, send a break callback msg to "host", too */
int
int hostindex;
char hoststr[16];
- ViceLog(7,
- ("BCB: BreakCallBack(all but %s:%d, (%u,%u,%u))\n",
- afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
- fid->Volume, fid->Vnode, fid->Unique));
+ if (xhost)
+ ViceLog(7,
+ ("BCB: BreakCallBack(Host %p all but %s:%d, (%u,%u,%u))\n",
+ xhost, afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
+ fid->Volume, fid->Vnode, fid->Unique));
+ else
+ ViceLog(7,
+ ("BCB: BreakCallBack(No Host, (%u,%u,%u))\n",
+ fid->Volume, fid->Vnode, fid->Unique));
H_LOCK;
cbstuff.BreakCallBacks++;
if (!fe) {
goto done;
}
- hostindex = h_htoi(xhost);
+ hostindex = xhost ? h_htoi(xhost) : 0;
cb = itocb(fe->firstcb);
if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
/* the most common case is what follows the || */
ViceLog(0, ("BCB: BOGUS! cb->hhead is NULL!\n"));
} else if (thishost->hostFlags & VENUSDOWN) {
ViceLog(7,
- ("BCB: %s:%d is down; delaying break call back\n",
- afs_inet_ntoa_r(thishost->host, hoststr),
+ ("BCB: %p (%s:%d) is down; delaying break call back\n",
+ thishost, afs_inet_ntoa_r(thishost->host, hoststr),
ntohs(thishost->port)));
cb->status = CB_DELAYED;
} else {
- h_Hold_r(thishost);
- cba[ncbas].hp = thishost;
- cba[ncbas].thead = cb->thead;
- ncbas++;
+ if (!(thishost->hostFlags & HOSTDELETED)) {
+ h_Hold_r(thishost);
+ cba[ncbas].hp = thishost;
+ cba[ncbas].thead = cb->thead;
+ ncbas++;
+ }
TDel(cb);
HDel(cb);
- CDel(cb, 1); /* Usually first; so this delete
+ CDel(cb, 1); /* Usually first; so this delete
* is reasonably inexpensive */
}
}
}
if (ncbas) {
- MultiBreakCallBack_r(cba, ncbas, &tf, xhost);
+ MultiBreakCallBack_r(cba, ncbas, &tf);
/* we need to to all these initializations again because MultiBreakCallBack may block */
fe = FindFE(fid);
int
DeleteCallBack(struct host *host, AFSFid * fid)
{
- register struct FileEntry *fe;
- register afs_uint32 *pcb;
+ struct FileEntry *fe;
+ afs_uint32 *pcb;
char hoststr[16];
+ H_LOCK;
cbstuff.DeleteCallBacks++;
- H_LOCK;
h_Lock_r(host);
+ /* do not care if the host has been HOSTDELETED */
fe = FindFE(fid);
if (!fe) {
h_Unlock_r(host);
pcb = FindCBPtr(fe, host);
if (!*pcb) {
ViceLog(8,
- ("DCB: No call back for host %s:%d, (%u, %u, %u)\n",
- afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
+ ("DCB: No call back for host %p (%s:%d), (%u, %u, %u)\n",
+ host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
fid->Volume, fid->Vnode, fid->Unique));
h_Unlock_r(host);
H_UNLOCK;
int
DeleteFileCallBacks(AFSFid * fid)
{
- register struct FileEntry *fe;
- register struct CallBack *cb;
- register afs_uint32 cbi;
- register int n;
+ struct FileEntry *fe;
+ struct CallBack *cb;
+ afs_uint32 cbi;
+ int n;
H_LOCK;
cbstuff.DeleteFiles++;
TDel(cb);
HDel(cb);
FreeCB(cb);
+ fe->ncbs--;
}
FDel(fe);
H_UNLOCK;
int
DeleteAllCallBacks_r(struct host *host, int deletefe)
{
- register struct CallBack *cb;
- register int cbi, first;
+ struct CallBack *cb;
+ int cbi, first;
cbstuff.DeleteAllCallBacks++;
cbi = first = host->cblist;
BreakDelayedCallBacks_r(struct host *host)
{
struct AFSFid fids[AFSCBMAX];
- u_byte thead[AFSCBMAX]; /* This should match thead in struct Callback */
int cbi, first, nfids;
struct CallBack *cb;
int code;
char hoststr[16];
+ struct rx_connection *cb_conn;
cbstuff.nbreakers++;
if (!(host->hostFlags & RESETDONE) && !(host->hostFlags & HOSTDELETED)) {
- host->hostFlags &= ~ALTADDR; /* alterrnate addresses are invalid */
+ host->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
+ cb_conn = host->callback_rxcon;
+ rx_GetConnection(cb_conn);
if (host->interface) {
H_UNLOCK;
code =
- RXAFSCB_InitCallBackState3(host->callback_rxcon,
- &FS_HostUUID);
- H_LOCK;
+ RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
} else {
H_UNLOCK;
- code = RXAFSCB_InitCallBackState(host->callback_rxcon);
- H_LOCK;
+ code = RXAFSCB_InitCallBackState(cb_conn);
}
+ rx_PutConnection(cb_conn);
+ cb_conn = NULL;
+ H_LOCK;
host->hostFlags |= ALTADDR; /* alternate addresses are valid */
if (code) {
if (ShowProblems) {
ViceLog(0,
- ("CB: Call back connect back failed (in break delayed) for %s:%d\n",
- afs_inet_ntoa_r(host->host, hoststr),
+ ("CB: Call back connect back failed (in break delayed) "
+ "for Host %p (%s:%d)\n",
+ host, afs_inet_ntoa_r(host->host, hoststr),
ntohs(host->port)));
}
host->hostFlags |= VENUSDOWN;
} else {
ViceLog(25,
- ("InitCallBackState success on %s\n",
- afs_inet_ntoa_r(host->host, hoststr)));
+ ("InitCallBackState success on %p (%s:%d)\n",
+ host, afs_inet_ntoa_r(host->host, hoststr),
+ ntohs(host->port)));
/* reset was done successfully */
host->hostFlags |= RESETDONE;
host->hostFlags &= ~VENUSDOWN;
cb = itocb(cbi);
cbi = cb->hnext;
if (cb->status == CB_DELAYED) {
- register struct FileEntry *fe = itofe(cb->fhead);
- thead[nfids] = cb->thead;
+ struct FileEntry *fe = itofe(cb->fhead);
fids[nfids].Volume = fe->volid;
fids[nfids].Vnode = fe->vnode;
fids[nfids].Unique = fe->unique;
int i;
if (ShowProblems) {
ViceLog(0,
- ("CB: XCallBackBulk failed, host=%s:%d; callback list follows:\n",
- afs_inet_ntoa_r(host->host, hoststr),
+ ("CB: XCallBackBulk failed, Host %p (%s:%d); "
+ "callback list follows:\n",
+ host, afs_inet_ntoa_r(host->host, hoststr),
ntohs(host->port)));
}
for (i = 0; i < nfids; i++) {
if (ShowProblems) {
ViceLog(0,
- ("CB: Host %s:%d, file %u.%u.%u (part of bulk callback)\n",
- afs_inet_ntoa_r(host->host, hoststr),
+ ("CB: Host %p (%s:%d), file %u.%u.%u "
+ "(part of bulk callback)\n",
+ host, afs_inet_ntoa_r(host->host, hoststr),
ntohs(host->port), fids[i].Volume,
fids[i].Vnode, fids[i].Unique));
}
cbstuff.nbreakers--;
/* If we succeeded it's always ok to unset HFE_LATER */
- if (!host->hostFlags & VENUSDOWN)
+ if (!(host->hostFlags & VENUSDOWN))
host->hostFlags &= ~HFE_LATER;
return (host->hostFlags & VENUSDOWN);
}
-/*
-** isheld is 0 if the host is held in h_Enumerate
-** isheld is 1 if the host is held in BreakVolumeCallBacks
-*/
static int
-MultiBreakVolumeCallBack_r(struct host *host, int isheld,
+MultiBreakVolumeCallBack_r(struct host *host,
struct VCBParams *parms, int deletefe)
{
char hoststr[16];
- if (!isheld)
- return isheld; /* host is held only by h_Enumerate, do nothing */
if (host->hostFlags & HOSTDELETED)
- return 0; /* host is deleted, release hold */
+ return 0;
+
+ if (!(host->hostFlags & HCBREAK))
+ return 0; /* host is not flagged to notify */
if (host->hostFlags & VENUSDOWN) {
h_Lock_r(host);
- if (host->hostFlags & HOSTDELETED) {
- h_Unlock_r(host);
- return 0; /* Release hold */
- }
- ViceLog(8,
- ("BVCB: volume call back for host %s:%d failed\n",
- afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
+ /* Do not care if the host is now HOSTDELETED */
if (ShowProblems) {
ViceLog(0,
- ("CB: volume callback for host %s:%d failed\n",
- afs_inet_ntoa_r(host->host, hoststr),
+ ("BVCB: volume callback for Host %p (%s:%d) failed\n",
+ host, afs_inet_ntoa_r(host->host, hoststr),
ntohs(host->port)));
}
- DeleteAllCallBacks_r(host, deletefe); /* Delete all callback state
- * rather than attempting to
+ DeleteAllCallBacks_r(host, deletefe); /* Delete all callback state
+ * rather than attempting to
* selectively remember to
* delete the volume callbacks
* later */
- host->hostFlags &= ~RESETDONE; /* Do InitCallBackState when host returns */
+ host->hostFlags &= ~(RESETDONE|HCBREAK); /* Do InitCallBackState when host returns */
h_Unlock_r(host);
- return 0; /* release hold */
+ return 0;
}
- assert(parms->ncbas <= MAX_CB_HOSTS);
+ opr_Assert(parms->ncbas <= MAX_CB_HOSTS);
/* Do not call MultiBreakCallBack on the current host structure
** because it would prematurely release the hold on the host
tf.AFSCBFids_val = parms->fid;
/* this releases all the hosts */
- MultiBreakCallBack_r(parms->cba, parms->ncbas, &tf, 0 /* xhost */ );
+ MultiBreakCallBack_r(parms->cba, parms->ncbas, &tf);
parms->ncbas = 0;
}
parms->cba[parms->ncbas].hp = host;
parms->cba[(parms->ncbas)++].thead = parms->thead;
- return 1; /* DON'T release hold, because we still need it. */
-}
+ host->hostFlags &= ~HCBREAK;
-/*
-** isheld is 0 if the host is held in h_Enumerate
-** isheld is 1 if the host is held in BreakVolumeCallBacks
-*/
-static int
-MultiBreakVolumeCallBack(struct host *host, int isheld,
- struct VCBParams *parms)
-{
- int retval;
- H_LOCK;
- retval = MultiBreakVolumeCallBack_r(host, isheld, parms, 1);
- H_UNLOCK;
- return retval;
+ /* we have more work to do on this host, so make sure we keep a reference
+ * to it */
+ h_Hold_r(host);
+
+ return 0;
}
-/*
-** isheld is 0 if the host is held in h_Enumerate
-** isheld is 1 if the host is held in BreakVolumeCallBacks
-*/
static int
-MultiBreakVolumeLaterCallBack(struct host *host, int isheld,
- struct VCBParams *parms)
+MultiBreakVolumeLaterCallBack(struct host *host, void *rock)
{
+ struct VCBParams *parms = (struct VCBParams *)rock;
int retval;
H_LOCK;
- retval = MultiBreakVolumeCallBack_r(host, isheld, parms, 0);
+ retval = MultiBreakVolumeCallBack_r(host, parms, 0);
H_UNLOCK;
return retval;
}
* this function is executing. It is just a temporary state, however,
* since the callback will be broken later by this same function.
*
- * Now uses multi-RX for CallBack RPC. Note that the
- * multiBreakCallBacks routine does not force a reset if the RPC
- * fails, unlike the previous version of this routine, but does create
- * a delayed callback. Resets will be forced if the host is
- * determined to be down before the RPC is executed.
+ * Now uses multi-RX for CallBack RPC in a different thread,
+ * only marking them here.
*/
-int
-BreakVolumeCallBacks(afs_uint32 volume)
-{
- struct AFSFid fid;
- int hash;
- afs_uint32 *feip;
- struct CallBack *cb;
- struct FileEntry *fe;
- struct host *host;
- struct VCBParams henumParms;
- afs_uint32 tthead = 0; /* zero is illegal value */
-
- H_LOCK;
- fid.Volume = volume, fid.Vnode = fid.Unique = 0;
- for (hash = 0; hash < VHASH; hash++) {
- for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
- if (fe->volid == volume) {
- register struct CallBack *cbnext;
- for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
- host = h_itoh(cb->hhead);
- h_Hold_r(host);
- cbnext = itocb(cb->cnext);
- if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
- tthead = cb->thead;
- }
- TDel(cb);
- HDel(cb);
- FreeCB(cb);
- /* leave hold for MultiBreakVolumeCallBack to clear */
- }
- *feip = fe->fnext;
- FreeFE(fe);
- } else {
- feip = &fe->fnext;
- }
- }
- }
-
- if (!tthead) {
- /* didn't find any callbacks, so return right away. */
- H_UNLOCK;
- return 0;
- }
- henumParms.ncbas = 0;
- henumParms.fid = &fid;
- henumParms.thead = tthead;
- H_UNLOCK;
- h_Enumerate(MultiBreakVolumeCallBack, (char *)&henumParms);
- H_LOCK;
- if (henumParms.ncbas) { /* do left-overs */
- struct AFSCBFids tf;
- tf.AFSCBFids_len = 1;
- tf.AFSCBFids_val = &fid;
-
- MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf, 0);
-
- henumParms.ncbas = 0;
- }
- H_UNLOCK;
- return 0;
-}
-
-#ifdef AFS_PTHREAD_ENV
extern pthread_cond_t fsync_cond;
-#else
-extern char fsync_wait[];
-#endif
int
-BreakVolumeCallBacksLater(afs_uint32 volume)
+BreakVolumeCallBacksLater(VolumeId volume)
{
int hash;
- afs_int32 *feip;
+ afs_uint32 *feip;
struct FileEntry *fe;
struct CallBack *cb;
struct host *host;
int found = 0;
- ViceLog(25, ("Setting later on volume %u\n", volume));
+ ViceLog(25, ("Setting later on volume %" AFS_VOLID_FMT "\n",
+ afs_printable_VolumeId_lu(volume)));
H_LOCK;
- for (hash = 0; hash < VHASH; hash++) {
- for (feip = &HashTable[hash]; fe = itofe(*feip);) {
+ for (hash = 0; hash < FEHASH_SIZE; hash++) {
+ for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
if (fe->volid == volume) {
- register struct CallBack *cbnext;
+ struct CallBack *cbnext;
for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
host = h_itoh(cb->hhead);
host->hostFlags |= HFE_LATER;
FSYNC_LOCK;
fe->status |= FE_LATER;
FSYNC_UNLOCK;
- found++;
+ found = 1;
}
feip = &fe->fnext;
}
}
ViceLog(25, ("Fsync thread wakeup\n"));
-#ifdef AFS_PTHREAD_ENV
- assert(pthread_cond_broadcast(&fsync_cond) == 0);
-#else
- LWP_NoYieldSignal(fsync_wait);
-#endif
+ FSYNC_LOCK;
+ opr_cv_broadcast(&fsync_cond);
+ FSYNC_UNLOCK;
return 0;
}
{
struct AFSFid fid;
int hash;
- afs_int32 *feip;
+ afs_uint32 *feip;
struct CallBack *cb;
struct FileEntry *fe = NULL;
struct FileEntry *myfe = NULL;
struct host *host;
struct VCBParams henumParms;
unsigned short tthead = 0; /* zero is illegal value */
+ char hoststr[16];
/* Unchain first */
ViceLog(25, ("Looking for FileEntries to unchain\n"));
/* Pick the first volume we see to clean up */
fid.Volume = fid.Vnode = fid.Unique = 0;
- for (hash = 0; hash < VHASH; hash++) {
- for (feip = &HashTable[hash]; fe = itofe(*feip);) {
+ for (hash = 0; hash < FEHASH_SIZE; hash++) {
+ for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
if (fe && (fe->status & FE_LATER)
&& (fid.Volume == 0 || fid.Volume == fe->volid)) {
/* Ugly, but used to avoid left side casting */
struct object *tmpfe;
ViceLog(125,
- ("Unchaining for %u:%u:%u\n", fe->vnode, fe->unique,
- fe->volid));
+ ("Unchaining for %u:%u:%" AFS_VOLID_FMT "\n", fe->vnode,
+ fe->unique, afs_printable_VolumeId_lu(fe->volid)));
fid.Volume = fe->volid;
*feip = fe->fnext;
+ fe->status &= ~FE_LATER; /* not strictly needed */
/* Works since volid is deeper than the largest pointer */
tmpfe = (struct object *)fe;
tmpfe->next = (struct object *)myfe;
/* loop over FEs from myfe and free/break */
tthead = 0;
for (fe = myfe; fe;) {
- register struct CallBack *cbnext;
+ struct CallBack *cbnext;
for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
- host = h_itoh(cb->hhead);
- h_Hold_r(host);
cbnext = itocb(cb->cnext);
- if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
- tthead = cb->thead;
+ host = h_itoh(cb->hhead);
+ if (cb->status == CB_DELAYED) {
+ if (!(host->hostFlags & HOSTDELETED)) {
+ /* mark this host for notification */
+ host->hostFlags |= HCBREAK;
+ if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
+ tthead = cb->thead;
+ }
+ }
+ TDel(cb);
+ HDel(cb);
+ CDel(cb, 0); /* Don't let CDel clean up the fe */
+ /* leave flag for MultiBreakVolumeCallBack to clear */
+ } else {
+ ViceLog(125,
+ ("Found host %p (%s:%d) non-DELAYED cb for %u:%u:%" AFS_VOLID_FMT "\n",
+ host, afs_inet_ntoa_r(host->host, hoststr),
+ ntohs(host->port), fe->vnode, fe->unique,
+ afs_printable_VolumeId_lu(fe->volid)));
}
- TDel(cb);
- HDel(cb);
- CDel(cb, 0); /* Don't let CDel clean up the fe */
- /* leave hold for MultiBreakVolumeCallBack to clear */
}
myfe = fe;
fe = (struct FileEntry *)((struct object *)fe)->next;
tf.AFSCBFids_len = 1;
tf.AFSCBFids_val = &fid;
- MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf, 0);
+ MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf);
henumParms.ncbas = 0;
}
}
int
CleanupTimedOutCallBacks_r(void)
{
- afs_uint32 now = CBtime(FT_ApproxTime());
- register afs_uint32 *thead;
- register struct CallBack *cb;
- register int ntimedout = 0;
+ afs_uint32 now = CBtime(time(NULL));
+ afs_uint32 *thead;
+ struct CallBack *cb;
+ int ntimedout = 0;
char hoststr[16];
while (tfirst <= now) {
- register int cbi;
+ int cbi;
cbi = *(thead = THead(tfirst));
if (cbi) {
do {
cb = itocb(cbi);
cbi = cb->tnext;
ViceLog(8,
- ("CCB: deleting timed out call back %s:%d, (%u,%u,%u)\n",
- afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
- h_itoh(cb->hhead)->port, itofe(cb->fhead)->volid,
+ ("CCB: deleting timed out call back %x (%s:%d), (%" AFS_VOLID_FMT ",%u,%u)\n",
+ h_itoh(cb->hhead)->host,
+ afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
+ h_itoh(cb->hhead)->port,
+ afs_printable_VolumeId_lu(itofe(cb->fhead)->volid),
itofe(cb->fhead)->vnode, itofe(cb->fhead)->unique));
HDel(cb);
CDel(cb, 1);
ntimedout++;
if (ntimedout > cbstuff.nblks) {
ViceLog(0, ("CCB: Internal Error -- shutting down...\n"));
- DumpCallBackState();
+ DumpCallBackState_r();
ShutDownAndCore(PANIC);
}
} while (cbi != *thead);
return (ntimedout > 0);
}
-static struct host *lih_host;
-static int lih_host_held = 0;
+/**
+ * parameters to pass to lih*_r from h_Enumerate_r when trying to find a host
+ * from which to clear callbacks.
+ */
+struct lih_params {
+ /**
+ * Points to the least interesting host found; try to clear callbacks on
+ * this host after h_Enumerate_r(lih*_r)'ing.
+ */
+ struct host *lih;
+
+ /**
+ * The last host we got from lih*_r, but we couldn't clear its callbacks
+ * for some reason. Choose the next-best host after this one (with the
+ * current lih*_r, this means to only select hosts that have an ActiveCall
+ * newer than lastlih).
+ */
+ struct host *lastlih;
+};
+
+/* Value of host->refCount that allows us to reliably infer that
+ * host may be held by some other thread */
+#define OTHER_MUSTHOLD_LIH 2
+/* This version does not allow 'host' to be selected unless its ActiveCall
+ * is newer than 'params->lastlih' which is the host with the oldest
+ * ActiveCall from the last pass (if it is provided). We filter out any hosts
+ * that are are held by other threads.
+ *
+ * There is a small problem here, but it may not be easily fixable. Say we
+ * select some host A, and give it back to GetSomeSpace_r. GSS_r for some
+ * reason cannot clear the callbacks on A, and so calls us again with
+ * lastlih = A. Suppose there is another host B that has the same ActiveCall
+ * time as A. We will now skip over host B, since
+ * 'hostB->ActiveCall > hostA->ActiveCall' is not true. This could result in
+ * us prematurely going to the GSS_r 2nd or 3rd pass, and making us a little
+ * inefficient. This should be pretty rare, though, except perhaps in cases
+ * with very small numbers of hosts.
+ *
+ * Also filter out any hosts with HOSTDELETED set. h_Enumerate_r should in
+ * theory not give these to us anyway, but be paranoid.
+ */
static int
-lih_r(register struct host *host, register int held,
- register struct host *hostp)
+lih0_r(struct host *host, void *rock)
{
- lih_host_held = 0;
+ struct lih_params *params = (struct lih_params *)rock;
+
+ /* OTHER_MUSTHOLD_LIH is because the h_Enum loop holds us once */
if (host->cblist
- && ((hostp && host != hostp) || (!held && !h_OtherHolds_r(host)))
- && (!lih_host || host->ActiveCall < lih_host->ActiveCall)) {
- lih_host = host;
+ && (!(host->hostFlags & HOSTDELETED))
+ && (host->refCount < OTHER_MUSTHOLD_LIH)
+ && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
+ && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
+
+ if (params->lih) {
+ h_Release_r(params->lih); /* release prev host */
+ }
+
+ h_Hold_r(host);
+ params->lih = host;
}
- if (!held) {
- held = 1;
- lih_host_held = 1;
+ return 0;
+}
+
+/* same as lih0_r, except we do not prevent held hosts from being selected. */
+static int
+lih1_r(struct host *host, void *rock)
+{
+ struct lih_params *params = (struct lih_params *)rock;
+
+ if (host->cblist
+ && (!(host->hostFlags & HOSTDELETED))
+ && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
+ && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
+
+ if (params->lih) {
+ h_Release_r(params->lih); /* release prev host */
+ }
+
+ h_Hold_r(host);
+ params->lih = host;
}
- return held;
+ return 0;
}
/* This could be upgraded to get more space each time */
-/* first pass: find the oldest host which isn't held by anyone */
-/* second pass: find the oldest host who isn't "me" */
+/* first pass: sequentially find the oldest host which isn't held by
+ anyone for which we can clear callbacks;
+ skipping 'hostp' */
+/* second pass: sequentially find the oldest host regardless of
+ whether or not the host is held; skipping 'hostp' */
+/* third pass: attempt to clear callbacks from 'hostp' */
/* always called with hostp unlocked */
+
+/* Note: hostlist is ordered most recently created host first and
+ * its order has no relationship to the most recently used. */
extern struct host *hostList;
static int
GetSomeSpace_r(struct host *hostp, int locked)
{
- register struct host *hp, *hp1 = (struct host *)0, *hp2 = hostList;
+ struct host *hp;
+ struct lih_params params;
int i = 0;
+ if (cbstuff.GotSomeSpaces == 0) {
+ /* only log this once; if GSS is getting called constantly, that's not
+ * good but don't make things worse by spamming the log. */
+ ViceLog(0, ("We have run out of callback space; forcing callback revocation. "
+ "This suggests the fileserver is configured with insufficient "
+ "callbacks; you probably want to increase the -cb fileserver "
+ "parameter (current setting: %u). The fileserver will continue "
+ "to operate, but this may indicate a severe performance problem\n",
+ cbstuff.nblks));
+ ViceLog(0, ("This message is logged at most once; for more information "
+ "see the OpenAFS documentation and fileserver xstat collection 3\n"));
+ }
+
cbstuff.GotSomeSpaces++;
ViceLog(5,
("GSS: First looking for timed out call backs via CleanupCallBacks\n"));
cbstuff.GSS3++;
return 0;
}
+
+ i = 0;
+ params.lastlih = NULL;
+
do {
- lih_host = 0;
- h_Enumerate_r(lih_r, hp2, (char *)hp1);
- hp = lih_host;
+ params.lih = NULL;
+
+ h_Enumerate_r(i == 0 ? lih0_r : lih1_r, hostList, ¶ms);
+
+ hp = params.lih;
+ if (params.lastlih) {
+ h_Release_r(params.lastlih);
+ params.lastlih = NULL;
+ }
+
if (hp) {
+ /* note that 'hp' was held by lih*_r; we will need to release it */
cbstuff.GSS4++;
- if (!ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
- if (lih_host_held)
- h_Release_r(hp);
+ if ((hp != hostp) && !ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
+ h_Release_r(hp);
return 0;
}
- if (lih_host_held)
- h_Release_r(hp);
- hp2 = hp->next;
+
+ params.lastlih = hp;
+ /* params.lastlih will be released on the next iteration, after
+ * h_Enumerate_r */
+
} else {
- hp2 = hostList;
- hp1 = hostp;
- cbstuff.GSS1++;
- ViceLog(5,
- ("GSS: Try harder for longest inactive host cnt= %d\n",
- i));
/*
* Next time try getting callbacks from any host even if
- * it's deleted (that's actually great since we can freely
- * remove its callbacks) or it's held since the only other
- * option is starvation for the file server (i.e. until the
- * callback timeout arrives).
+ * it's held, since the only other option is starvation for
+ * the file server (i.e. until the callback timeout arrives).
*/
i++;
+ params.lastlih = NULL;
+ cbstuff.GSS1++;
+ ViceLog(5,
+ ("GSS: Try harder for longest inactive host cnt= %d\n",
+ i));
}
} while (i < 2);
- /*
- * No choice to clear this host's callback state
- */
- /* third pass: we still haven't gotten any space, so we free what we had
- * previously passed over. */
+
+ /* Could not obtain space from other hosts, clear hostp's callback state */
cbstuff.GSS2++;
if (!locked) {
h_Lock_r(hostp);
ClearHostCallbacks_r(struct host *hp, int locked)
{
int code;
- int held = 0;
char hoststr[16];
+ struct rx_connection *cb_conn = NULL;
ViceLog(5,
- ("GSS: Delete longest inactive host %s\n",
- afs_inet_ntoa_r(hp->host, hoststr)));
- if (!(held = h_Held_r(hp)))
- h_Hold_r(hp);
+ ("GSS: Delete longest inactive host %p (%s:%d)\n",
+ hp, afs_inet_ntoa_r(hp->host, hoststr), ntohs(hp->port)));
+
+ if ((hp->hostFlags & HOSTDELETED)) {
+ /* hp could go away after reacquiring H_LOCK in h_NBLock_r, so we can't
+ * really use it; its callbacks will get cleared anyway when
+ * h_TossStuff_r gets its hands on it */
+ return 1;
+ }
+
+ h_Hold_r(hp);
/** Try a non-blocking lock. If the lock is already held return
* after releasing hold on hp
*/
if (!locked) {
- if (h_NBLock_r(hp)) {
- if (!held)
- h_Release_r(hp);
- return 1;
- }
+ if (h_NBLock_r(hp)) {
+ h_Release_r(hp);
+ return 1;
+ }
}
if (hp->Console & 2) {
/*
DeleteAllCallBacks_r(hp, 1);
if (hp->hostFlags & VENUSDOWN) {
hp->hostFlags &= ~RESETDONE; /* remember that we must do a reset */
- } else {
+ } else if (!(hp->hostFlags & HOSTDELETED)) {
/* host is up, try a call */
hp->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
+ cb_conn = hp->callback_rxcon;
+ rx_GetConnection(hp->callback_rxcon);
if (hp->interface) {
H_UNLOCK;
code =
- RXAFSCB_InitCallBackState3(hp->callback_rxcon, &FS_HostUUID);
- H_LOCK;
+ RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
} else {
H_UNLOCK;
- code = RXAFSCB_InitCallBackState(hp->callback_rxcon);
- H_LOCK;
+ code = RXAFSCB_InitCallBackState(cb_conn);
}
+ rx_PutConnection(cb_conn);
+ cb_conn = NULL;
+ H_LOCK;
hp->hostFlags |= ALTADDR; /* alternate addresses are valid */
if (code) {
/* failed, mark host down and need reset */
hp->hostFlags |= RESETDONE;
}
}
- if (!locked) {
- h_Unlock_r(hp);
- }
- if (!held)
- h_Release_r(hp);
+ if (!locked)
+ h_Unlock_r(hp);
+ h_Release_r(hp);
return 0;
}
fprintf(stderr, "%d CBs, %d FEs, (%d of total of %d 16-byte blocks)\n",
cbstuff.nCBs, cbstuff.nFEs, cbstuff.nCBs + cbstuff.nFEs,
cbstuff.nblks);
+ fprintf(stderr, "%d GSS1, %d GSS2, %d GSS3, %d GSS4, %d GSS5 (internal counters)\n",
+ cbstuff.GSS1, cbstuff.GSS2, cbstuff.GSS3, cbstuff.GSS4, cbstuff.GSS5);
return 0;
}
#define MAGIC 0x12345678 /* To check byte ordering of dump when it is read in */
+#define MAGICV2 0x12345679 /* To check byte ordering & version of dump when it is read in */
+
#ifndef INTERPRET_DUMP
+#ifdef AFS_DEMAND_ATTACH_FS
+/*
+ * demand attach fs
+ * callback state serialization
+ */
+static int cb_stateSaveTimeouts(struct fs_dump_state * state);
+static int cb_stateSaveFEHash(struct fs_dump_state * state);
+static int cb_stateSaveFEs(struct fs_dump_state * state);
+static int cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe);
+static int cb_stateRestoreTimeouts(struct fs_dump_state * state);
+static int cb_stateRestoreFEHash(struct fs_dump_state * state);
+static int cb_stateRestoreFEs(struct fs_dump_state * state);
+static int cb_stateRestoreFE(struct fs_dump_state * state);
+static int cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
+ struct iovec * iov, int niovecs);
+
+static int cb_stateVerifyFEHash(struct fs_dump_state * state);
+static int cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe);
+static int cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe);
+static int cb_stateVerifyTimeoutQueues(struct fs_dump_state * state);
+
+static int cb_stateFEToDiskEntry(struct FileEntry *, struct FEDiskEntry *);
+static int cb_stateDiskEntryToFE(struct fs_dump_state * state,
+ struct FEDiskEntry *, struct FileEntry *);
+
+static int cb_stateCBToDiskEntry(struct CallBack *, struct CBDiskEntry *);
+static int cb_stateDiskEntryToCB(struct fs_dump_state * state,
+ struct CBDiskEntry *, struct CallBack *);
+
+static int cb_stateFillHeader(struct callback_state_header * hdr);
+static int cb_stateCheckHeader(struct callback_state_header * hdr);
+
+static int cb_stateAllocMap(struct fs_dump_state * state);
+
int
-DumpCallBackState(void)
+cb_stateSave(struct fs_dump_state * state)
{
- int fd;
- afs_uint32 magic = MAGIC, now = FT_ApproxTime(), freelisthead;
+ int ret = 0;
- fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, O_WRONLY | O_CREAT | O_TRUNC,
- 0666);
- if (fd < 0) {
- ViceLog(0,
- ("Couldn't create callback dump file %s\n",
- AFSDIR_SERVER_CBKDUMP_FILEPATH));
- return 0;
- }
- (void)write(fd, &magic, sizeof(magic));
- (void)write(fd, &now, sizeof(now));
- (void)write(fd, &cbstuff, sizeof(cbstuff));
- (void)write(fd, TimeOuts, sizeof(TimeOuts));
- (void)write(fd, timeout, sizeof(timeout));
- (void)write(fd, &tfirst, sizeof(tfirst));
- freelisthead = cbtoi((struct CallBack *)CBfree);
- (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
- freelisthead = fetoi((struct FileEntry *)FEfree);
- (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
- (void)write(fd, HashTable, sizeof(HashTable));
- (void)write(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
- (void)write(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
- close(fd);
+ AssignInt64(state->eof_offset, &state->hdr->cb_offset);
- return 0;
-}
+ /* invalidate callback state header */
+ memset(state->cb_hdr, 0, sizeof(struct callback_state_header));
+ if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
+ sizeof(struct callback_state_header))) {
+ ret = 1;
+ goto done;
+ }
-#endif
+ fs_stateIncEOF(state, sizeof(struct callback_state_header));
-#ifdef INTERPRET_DUMP
+ /* dump timeout state */
+ if (cb_stateSaveTimeouts(state)) {
+ ret = 1;
+ goto done;
+ }
-/* This is only compiled in for the callback analyzer program */
-/* Returns the time of the dump */
-time_t
-ReadDump(char *file)
-{
- int fd;
- afs_uint32 magic, freelisthead;
- time_t now;
+ /* dump fe hashtable state */
+ if (cb_stateSaveFEHash(state)) {
+ ret = 1;
+ goto done;
+ }
- fd = open(file, O_RDONLY);
+ /* dump callback state */
+ if (cb_stateSaveFEs(state)) {
+ ret = 1;
+ goto done;
+ }
+
+ /* write the callback state header to disk */
+ cb_stateFillHeader(state->cb_hdr);
+ if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
+ sizeof(struct callback_state_header))) {
+ ret = 1;
+ goto done;
+ }
+
+ done:
+ return ret;
+}
+
+int
+cb_stateRestore(struct fs_dump_state * state)
+{
+ int ret = 0;
+
+ if (fs_stateReadHeader(state, &state->hdr->cb_offset, state->cb_hdr,
+ sizeof(struct callback_state_header))) {
+ ret = 1;
+ goto done;
+ }
+
+ if (cb_stateCheckHeader(state->cb_hdr)) {
+ ret = 1;
+ goto done;
+ }
+
+ if (cb_stateAllocMap(state)) {
+ ret = 1;
+ goto done;
+ }
+
+ if (cb_stateRestoreTimeouts(state)) {
+ ret = 1;
+ goto done;
+ }
+
+ if (cb_stateRestoreFEHash(state)) {
+ ret = 1;
+ goto done;
+ }
+
+ /* restore FEs and CBs from disk */
+ if (cb_stateRestoreFEs(state)) {
+ ret = 1;
+ goto done;
+ }
+
+ /* restore the timeout queue heads */
+ tfirst = state->cb_hdr->tfirst;
+
+ done:
+ return ret;
+}
+
+int
+cb_stateRestoreIndices(struct fs_dump_state * state)
+{
+ int i, ret = 0;
+ struct FileEntry * fe;
+ struct CallBack * cb;
+
+ /* restore indices in the FileEntry structures */
+ for (i = 1; i < state->fe_map.len; i++) {
+ if (state->fe_map.entries[i].new_idx) {
+ fe = itofe(state->fe_map.entries[i].new_idx);
+
+ /* restore the fe->fnext entry */
+ if (fe_OldToNew(state, fe->fnext, &fe->fnext)) {
+ ret = 1;
+ goto done;
+ }
+
+ /* restore the fe->firstcb entry */
+ if (cb_OldToNew(state, fe->firstcb, &fe->firstcb)) {
+ ret = 1;
+ goto done;
+ }
+ }
+ }
+
+ /* restore indices in the CallBack structures */
+ for (i = 1; i < state->cb_map.len; i++) {
+ if (state->cb_map.entries[i].new_idx) {
+ cb = itocb(state->cb_map.entries[i].new_idx);
+
+ /* restore the cb->cnext entry */
+ if (cb_OldToNew(state, cb->cnext, &cb->cnext)) {
+ ret = 1;
+ goto done;
+ }
+
+ /* restore the cb->fhead entry */
+ if (fe_OldToNew(state, cb->fhead, &cb->fhead)) {
+ ret = 1;
+ goto done;
+ }
+
+ /* restore the cb->hhead entry */
+ if (h_OldToNew(state, cb->hhead, &cb->hhead)) {
+ ret = 1;
+ goto done;
+ }
+
+ /* restore the cb->tprev entry */
+ if (cb_OldToNew(state, cb->tprev, &cb->tprev)) {
+ ret = 1;
+ goto done;
+ }
+
+ /* restore the cb->tnext entry */
+ if (cb_OldToNew(state, cb->tnext, &cb->tnext)) {
+ ret = 1;
+ goto done;
+ }
+
+ /* restore the cb->hprev entry */
+ if (cb_OldToNew(state, cb->hprev, &cb->hprev)) {
+ ret = 1;
+ goto done;
+ }
+
+ /* restore the cb->hnext entry */
+ if (cb_OldToNew(state, cb->hnext, &cb->hnext)) {
+ ret = 1;
+ goto done;
+ }
+ }
+ }
+
+ /* restore the timeout queue head indices */
+ for (i = 0; i < state->cb_timeout_hdr->records; i++) {
+ if (cb_OldToNew(state, timeout[i], &timeout[i])) {
+ ret = 1;
+ goto done;
+ }
+ }
+
+ /* restore the FE hash table queue heads */
+ for (i = 0; i < state->cb_fehash_hdr->records; i++) {
+ if (fe_OldToNew(state, HashTable[i], &HashTable[i])) {
+ ret = 1;
+ goto done;
+ }
+ }
+
+ done:
+ return ret;
+}
+
+int
+cb_stateVerify(struct fs_dump_state * state)
+{
+ int ret = 0;
+
+ if (cb_stateVerifyFEHash(state)) {
+ ret = 1;
+ }
+
+ if (cb_stateVerifyTimeoutQueues(state)) {
+ ret = 1;
+ }
+
+ return ret;
+}
+
+static int
+cb_stateVerifyFEHash(struct fs_dump_state * state)
+{
+ int ret = 0, i;
+ struct FileEntry * fe;
+ afs_uint32 fei, chain_len;
+
+ for (i = 0; i < FEHASH_SIZE; i++) {
+ chain_len = 0;
+ for (fei = HashTable[i], fe = itofe(fei);
+ fe;
+ fei = fe->fnext, fe = itofe(fei)) {
+ if (fei > cbstuff.nblks) {
+ ViceLog(0, ("cb_stateVerifyFEHash: error: index out of range (fei=%d)\n", fei));
+ ret = 1;
+ break;
+ }
+ if (cb_stateVerifyFE(state, fe)) {
+ ret = 1;
+ }
+ if (chain_len > FS_STATE_FE_MAX_HASH_CHAIN_LEN) {
+ ViceLog(0, ("cb_stateVerifyFEHash: error: hash chain %d length exceeds %d; assuming there's a loop\n",
+ i, FS_STATE_FE_MAX_HASH_CHAIN_LEN));
+ ret = 1;
+ break;
+ }
+ chain_len++;
+ }
+ }
+
+ return ret;
+}
+
+static int
+cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe)
+{
+ int ret = 0;
+
+ if ((fe->firstcb && !fe->ncbs) ||
+ (!fe->firstcb && fe->ncbs)) {
+ ViceLog(0, ("cb_stateVerifyFE: error: fe->firstcb does not agree with fe->ncbs (fei=%lu, fe->firstcb=%lu, fe->ncbs=%lu)\n",
+ afs_printable_uint32_lu(fetoi(fe)),
+ afs_printable_uint32_lu(fe->firstcb),
+ afs_printable_uint32_lu(fe->ncbs)));
+ ret = 1;
+ }
+ if (cb_stateVerifyFCBList(state, fe)) {
+ ViceLog(0, ("cb_stateVerifyFE: error: FCBList failed verification (fei=%lu)\n",
+ afs_printable_uint32_lu(fetoi(fe))));
+ ret = 1;
+ }
+
+ return ret;
+}
+
+static int
+cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe)
+{
+ int ret = 0;
+ afs_uint32 cbi, fei, chain_len = 0;
+ struct CallBack * cb;
+
+ fei = fetoi(fe);
+
+ for (cbi = fe->firstcb, cb = itocb(cbi);
+ cb;
+ cbi = cb->cnext, cb = itocb(cbi)) {
+ if (cbi > cbstuff.nblks) {
+ ViceLog(0, ("cb_stateVerifyFCBList: error: list index out of range (cbi=%d, ncbs=%d)\n",
+ cbi, cbstuff.nblks));
+ ret = 1;
+ goto done;
+ }
+ if (cb->fhead != fei) {
+ ViceLog(0, ("cb_stateVerifyFCBList: error: cb->fhead != fei (fei=%d, cb->fhead=%d)\n",
+ fei, cb->fhead));
+ ret = 1;
+ }
+ if (chain_len > FS_STATE_FCB_MAX_LIST_LEN) {
+ ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (fei=%d); assuming there's a loop\n",
+ FS_STATE_FCB_MAX_LIST_LEN, fei));
+ ret = 1;
+ goto done;
+ }
+ chain_len++;
+ }
+
+ if (fe->ncbs != chain_len) {
+ ViceLog(0, ("cb_stateVerifyFCBList: error: list length mismatch (len=%d, fe->ncbs=%d)\n",
+ chain_len, fe->ncbs));
+ ret = 1;
+ }
+
+ done:
+ return ret;
+}
+
+int
+cb_stateVerifyHCBList(struct fs_dump_state * state, struct host * host)
+{
+ int ret = 0;
+ afs_uint32 hi, chain_len, cbi;
+ struct CallBack *cb, *ncb;
+
+ hi = h_htoi(host);
+ chain_len = 0;
+
+ for (cbi = host->cblist, cb = itocb(cbi);
+ cb;
+ cbi = cb->hnext, cb = ncb) {
+ if (chain_len && (host->cblist == cbi)) {
+ /* we've wrapped around the circular list, and everything looks ok */
+ break;
+ }
+ if (cb->hhead != hi) {
+ ViceLog(0, ("cb_stateVerifyHCBList: error: incorrect cb->hhead (cbi=%d, h->index=%d, cb->hhead=%d)\n",
+ cbi, hi, cb->hhead));
+ ret = 1;
+ }
+ if (!cb->hprev || !cb->hnext) {
+ ViceLog(0, ("cb_stateVerifyHCBList: error: null index in circular list (cbi=%d, h->index=%d)\n",
+ cbi, hi));
+ ret = 1;
+ goto done;
+ }
+ if ((cb->hprev > cbstuff.nblks) ||
+ (cb->hnext > cbstuff.nblks)) {
+ ViceLog(0, ("cb_stateVerifyHCBList: error: list index out of range (cbi=%d, h->index=%d, cb->hprev=%d, cb->hnext=%d, nCBs=%d)\n",
+ cbi, hi, cb->hprev, cb->hnext, cbstuff.nblks));
+ ret = 1;
+ goto done;
+ }
+ ncb = itocb(cb->hnext);
+ if (cbi != ncb->hprev) {
+ ViceLog(0, ("cb_stateVerifyHCBList: error: corrupt linked list (cbi=%d, h->index=%d)\n",
+ cbi, hi));
+ ret = 1;
+ goto done;
+ }
+ if (chain_len > FS_STATE_HCB_MAX_LIST_LEN) {
+ ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (h->index=%d); assuming there's a loop\n",
+ FS_STATE_HCB_MAX_LIST_LEN, hi));
+ ret = 1;
+ goto done;
+ }
+ chain_len++;
+ }
+
+ done:
+ return ret;
+}
+
+static int
+cb_stateVerifyTimeoutQueues(struct fs_dump_state * state)
+{
+ int ret = 0, i;
+ afs_uint32 cbi, chain_len;
+ struct CallBack *cb, *ncb;
+
+ for (i = 0; i < CB_NUM_TIMEOUT_QUEUES; i++) {
+ chain_len = 0;
+ for (cbi = timeout[i], cb = itocb(cbi);
+ cb;
+ cbi = cb->tnext, cb = ncb) {
+ if (chain_len && (cbi == timeout[i])) {
+ /* we've wrapped around the circular list, and everything looks ok */
+ break;
+ }
+ if (cbi > cbstuff.nblks) {
+ ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: list index out of range (cbi=%d, tindex=%d)\n",
+ cbi, i));
+ ret = 1;
+ break;
+ }
+ if (itot(cb->thead) != &timeout[i]) {
+ ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: cb->thead points to wrong timeout queue (tindex=%d, cbi=%d, cb->thead=%d)\n",
+ i, cbi, cb->thead));
+ ret = 1;
+ }
+ if (!cb->tprev || !cb->tnext) {
+ ViceLog(0, ("cb_stateVerifyTimeoutQueues: null index in circular list (cbi=%d, tindex=%d)\n",
+ cbi, i));
+ ret = 1;
+ break;
+ }
+ if ((cb->tprev > cbstuff.nblks) ||
+ (cb->tnext > cbstuff.nblks)) {
+ ViceLog(0, ("cb_stateVerifyTimeoutQueues: list index out of range (cbi=%d, tindex=%d, cb->tprev=%d, cb->tnext=%d, nCBs=%d)\n",
+ cbi, i, cb->tprev, cb->tnext, cbstuff.nblks));
+ ret = 1;
+ break;
+ }
+ ncb = itocb(cb->tnext);
+ if (cbi != ncb->tprev) {
+ ViceLog(0, ("cb_stateVerifyTimeoutQueues: corrupt linked list (cbi=%d, tindex=%d)\n",
+ cbi, i));
+ ret = 1;
+ break;
+ }
+ if (chain_len > FS_STATE_TCB_MAX_LIST_LEN) {
+ ViceLog(0, ("cb_stateVerifyTimeoutQueues: list length exceeds %d (tindex=%d); assuming there's a loop\n",
+ FS_STATE_TCB_MAX_LIST_LEN, i));
+ ret = 1;
+ break;
+ }
+ chain_len++;
+ }
+ }
+
+ return ret;
+}
+
+static int
+cb_stateSaveTimeouts(struct fs_dump_state * state)
+{
+ int ret = 0;
+ struct iovec iov[2];
+
+ AssignInt64(state->eof_offset, &state->cb_hdr->timeout_offset);
+
+ memset(state->cb_timeout_hdr, 0, sizeof(struct callback_state_fehash_header));
+ state->cb_timeout_hdr->magic = CALLBACK_STATE_TIMEOUT_MAGIC;
+ state->cb_timeout_hdr->records = CB_NUM_TIMEOUT_QUEUES;
+ state->cb_timeout_hdr->len = sizeof(struct callback_state_timeout_header) +
+ (state->cb_timeout_hdr->records * sizeof(afs_uint32));
+
+ iov[0].iov_base = (char *)state->cb_timeout_hdr;
+ iov[0].iov_len = sizeof(struct callback_state_timeout_header);
+ iov[1].iov_base = (char *)timeout;
+ iov[1].iov_len = sizeof(timeout);
+
+ if (fs_stateSeek(state, &state->cb_hdr->timeout_offset)) {
+ ret = 1;
+ goto done;
+ }
+
+ if (fs_stateWriteV(state, iov, 2)) {
+ ret = 1;
+ goto done;
+ }
+
+ fs_stateIncEOF(state, state->cb_timeout_hdr->len);
+
+ done:
+ return ret;
+}
+
+static int
+cb_stateRestoreTimeouts(struct fs_dump_state * state)
+{
+ int ret = 0, len;
+
+ if (fs_stateReadHeader(state, &state->cb_hdr->timeout_offset,
+ state->cb_timeout_hdr,
+ sizeof(struct callback_state_timeout_header))) {
+ ret = 1;
+ goto done;
+ }
+
+ if (state->cb_timeout_hdr->magic != CALLBACK_STATE_TIMEOUT_MAGIC) {
+ ret = 1;
+ goto done;
+ }
+ if (state->cb_timeout_hdr->records != CB_NUM_TIMEOUT_QUEUES) {
+ ret = 1;
+ goto done;
+ }
+
+ len = state->cb_timeout_hdr->records * sizeof(afs_uint32);
+
+ if (state->cb_timeout_hdr->len !=
+ (sizeof(struct callback_state_timeout_header) + len)) {
+ ret = 1;
+ goto done;
+ }
+
+ if (fs_stateRead(state, timeout, len)) {
+ ret = 1;
+ goto done;
+ }
+
+ done:
+ return ret;
+}
+
+static int
+cb_stateSaveFEHash(struct fs_dump_state * state)
+{
+ int ret = 0;
+ struct iovec iov[2];
+
+ AssignInt64(state->eof_offset, &state->cb_hdr->fehash_offset);
+
+ memset(state->cb_fehash_hdr, 0, sizeof(struct callback_state_fehash_header));
+ state->cb_fehash_hdr->magic = CALLBACK_STATE_FEHASH_MAGIC;
+ state->cb_fehash_hdr->records = FEHASH_SIZE;
+ state->cb_fehash_hdr->len = sizeof(struct callback_state_fehash_header) +
+ (state->cb_fehash_hdr->records * sizeof(afs_uint32));
+
+ iov[0].iov_base = (char *)state->cb_fehash_hdr;
+ iov[0].iov_len = sizeof(struct callback_state_fehash_header);
+ iov[1].iov_base = (char *)HashTable;
+ iov[1].iov_len = sizeof(HashTable);
+
+ if (fs_stateSeek(state, &state->cb_hdr->fehash_offset)) {
+ ret = 1;
+ goto done;
+ }
+
+ if (fs_stateWriteV(state, iov, 2)) {
+ ret = 1;
+ goto done;
+ }
+
+ fs_stateIncEOF(state, state->cb_fehash_hdr->len);
+
+ done:
+ return ret;
+}
+
+static int
+cb_stateRestoreFEHash(struct fs_dump_state * state)
+{
+ int ret = 0, len;
+
+ if (fs_stateReadHeader(state, &state->cb_hdr->fehash_offset,
+ state->cb_fehash_hdr,
+ sizeof(struct callback_state_fehash_header))) {
+ ret = 1;
+ goto done;
+ }
+
+ if (state->cb_fehash_hdr->magic != CALLBACK_STATE_FEHASH_MAGIC) {
+ ret = 1;
+ goto done;
+ }
+ if (state->cb_fehash_hdr->records != FEHASH_SIZE) {
+ ret = 1;
+ goto done;
+ }
+
+ len = state->cb_fehash_hdr->records * sizeof(afs_uint32);
+
+ if (state->cb_fehash_hdr->len !=
+ (sizeof(struct callback_state_fehash_header) + len)) {
+ ret = 1;
+ goto done;
+ }
+
+ if (fs_stateRead(state, HashTable, len)) {
+ ret = 1;
+ goto done;
+ }
+
+ done:
+ return ret;
+}
+
+static int
+cb_stateSaveFEs(struct fs_dump_state * state)
+{
+ int ret = 0;
+ int fei, hash;
+ struct FileEntry *fe;
+
+ AssignInt64(state->eof_offset, &state->cb_hdr->fe_offset);
+
+ for (hash = 0; hash < FEHASH_SIZE ; hash++) {
+ for (fei = HashTable[hash]; fei; fei = fe->fnext) {
+ fe = itofe(fei);
+ if (cb_stateSaveFE(state, fe)) {
+ ret = 1;
+ goto done;
+ }
+ }
+ }
+
+ done:
+ return ret;
+}
+
+static int
+cb_stateRestoreFEs(struct fs_dump_state * state)
+{
+ int count, nFEs, ret = 0;
+
+ nFEs = state->cb_hdr->nFEs;
+
+ for (count = 0; count < nFEs; count++) {
+ if (cb_stateRestoreFE(state)) {
+ ret = 1;
+ goto done;
+ }
+ }
+
+ done:
+ return ret;
+}
+
+static int
+cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe)
+{
+ int ret = 0, iovcnt, cbi, written = 0;
+ afs_uint32 fei;
+ struct callback_state_entry_header hdr;
+ struct FEDiskEntry fedsk;
+ struct CBDiskEntry cbdsk[16];
+ struct iovec iov[16];
+ struct CallBack *cb;
+
+ fei = fetoi(fe);
+ if (fei > state->cb_hdr->fe_max) {
+ state->cb_hdr->fe_max = fei;
+ }
+
+ memset(&hdr, 0, sizeof(struct callback_state_entry_header));
+
+ if (cb_stateFEToDiskEntry(fe, &fedsk)) {
+ ret = 1;
+ goto done;
+ }
+
+ iov[0].iov_base = (char *)&hdr;
+ iov[0].iov_len = sizeof(hdr);
+ iov[1].iov_base = (char *)&fedsk;
+ iov[1].iov_len = sizeof(struct FEDiskEntry);
+ iovcnt = 2;
+
+ for (cbi = fe->firstcb, cb = itocb(cbi);
+ cb != NULL;
+ cbi = cb->cnext, cb = itocb(cbi), hdr.nCBs++) {
+ if (cbi > state->cb_hdr->cb_max) {
+ state->cb_hdr->cb_max = cbi;
+ }
+ if (cb_stateCBToDiskEntry(cb, &cbdsk[iovcnt])) {
+ ret = 1;
+ goto done;
+ }
+ cbdsk[iovcnt].index = cbi;
+ iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
+ iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
+ iovcnt++;
+ if ((iovcnt == 16) || (!cb->cnext)) {
+ if (fs_stateWriteV(state, iov, iovcnt)) {
+ ret = 1;
+ goto done;
+ }
+ written = 1;
+ iovcnt = 0;
+ }
+ }
+
+ hdr.magic = CALLBACK_STATE_ENTRY_MAGIC;
+ hdr.len = sizeof(hdr) + sizeof(struct FEDiskEntry) +
+ (hdr.nCBs * sizeof(struct CBDiskEntry));
+
+ if (!written) {
+ if (fs_stateWriteV(state, iov, iovcnt)) {
+ ret = 1;
+ goto done;
+ }
+ } else {
+ if (fs_stateWriteHeader(state, &state->eof_offset, &hdr, sizeof(hdr))) {
+ ret = 1;
+ goto done;
+ }
+ }
+
+ fs_stateIncEOF(state, hdr.len);
+
+ if (written) {
+ if (fs_stateSeek(state, &state->eof_offset)) {
+ ret = 1;
+ goto done;
+ }
+ }
+
+ state->cb_hdr->nFEs++;
+ state->cb_hdr->nCBs += hdr.nCBs;
+
+ done:
+ return ret;
+}
+
+static int
+cb_stateRestoreFE(struct fs_dump_state * state)
+{
+ int ret = 0, iovcnt, nCBs;
+ struct callback_state_entry_header hdr;
+ struct FEDiskEntry fedsk;
+ struct CBDiskEntry cbdsk[16];
+ struct iovec iov[16];
+ struct FileEntry * fe;
+
+ iov[0].iov_base = (char *)&hdr;
+ iov[0].iov_len = sizeof(hdr);
+ iov[1].iov_base = (char *)&fedsk;
+ iov[1].iov_len = sizeof(fedsk);
+ iovcnt = 2;
+
+ if (fs_stateReadV(state, iov, iovcnt)) {
+ ret = 1;
+ goto done;
+ }
+
+ if (hdr.magic != CALLBACK_STATE_ENTRY_MAGIC) {
+ ret = 1;
+ goto done;
+ }
+
+ fe = GetFE();
+ if (fe == NULL) {
+ ViceLog(0, ("cb_stateRestoreFE: ran out of free FileEntry structures\n"));
+ ret = 1;
+ goto done;
+ }
+
+ if (cb_stateDiskEntryToFE(state, &fedsk, fe)) {
+ ret = 1;
+ goto done;
+ }
+
+ if (hdr.nCBs) {
+ for (iovcnt = 0, nCBs = 0;
+ nCBs < hdr.nCBs;
+ nCBs++) {
+ iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
+ iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
+ iovcnt++;
+ if ((iovcnt == 16) || (nCBs == hdr.nCBs - 1)) {
+ if (fs_stateReadV(state, iov, iovcnt)) {
+ ret = 1;
+ goto done;
+ }
+ if (cb_stateRestoreCBs(state, fe, iov, iovcnt)) {
+ ret = 1;
+ goto done;
+ }
+ iovcnt = 0;
+ }
+ }
+ }
+
+ done:
+ return ret;
+}
+
+static int
+cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
+ struct iovec * iov, int niovecs)
+{
+ int ret = 0, idx;
+ struct CallBack * cb;
+ struct CBDiskEntry * cbdsk;
+
+ for (idx = 0; idx < niovecs; idx++) {
+ cbdsk = (struct CBDiskEntry *) iov[idx].iov_base;
+
+ if (cbdsk->cb.hhead < state->h_map.len &&
+ state->h_map.entries[cbdsk->cb.hhead].valid == FS_STATE_IDX_SKIPPED) {
+ continue;
+ }
+
+ if ((cb = GetCB()) == NULL) {
+ ViceLog(0, ("cb_stateRestoreCBs: ran out of free CallBack structures\n"));
+ ret = 1;
+ goto done;
+ }
+ if (cb_stateDiskEntryToCB(state, cbdsk, cb)) {
+ ViceLog(0, ("cb_stateRestoreCBs: corrupt CallBack disk entry\n"));
+ ret = 1;
+ goto done;
+ }
+ }
+
+ done:
+ return ret;
+}
+
+
+static int
+cb_stateFillHeader(struct callback_state_header * hdr)
+{
+ hdr->stamp.magic = CALLBACK_STATE_MAGIC;
+ hdr->stamp.version = CALLBACK_STATE_VERSION;
+ hdr->tfirst = tfirst;
+ return 0;
+}
+
+static int
+cb_stateCheckHeader(struct callback_state_header * hdr)
+{
+ int ret = 0;
+
+ if (hdr->stamp.magic != CALLBACK_STATE_MAGIC) {
+ ret = 1;
+ } else if (hdr->stamp.version != CALLBACK_STATE_VERSION) {
+ ret = 1;
+ } else if ((hdr->nFEs > cbstuff.nblks) || (hdr->nCBs > cbstuff.nblks)) {
+ ViceLog(0, ("cb_stateCheckHeader: saved callback state larger than callback memory allocation\n"));
+ ret = 1;
+ }
+ return ret;
+}
+
+/* disk entry conversion routines */
+static int
+cb_stateFEToDiskEntry(struct FileEntry * in, struct FEDiskEntry * out)
+{
+ memcpy(&out->fe, in, sizeof(struct FileEntry));
+ out->index = fetoi(in);
+ return 0;
+}
+
+static int
+cb_stateDiskEntryToFE(struct fs_dump_state * state,
+ struct FEDiskEntry * in, struct FileEntry * out)
+{
+ int ret = 0;
+
+ memcpy(out, &in->fe, sizeof(struct FileEntry));
+
+ /* setup FE map entry */
+ if (!in->index || (in->index >= state->fe_map.len)) {
+ ViceLog(0, ("cb_stateDiskEntryToFE: index (%d) out of range",
+ in->index));
+ ret = 1;
+ goto done;
+ }
+ state->fe_map.entries[in->index].valid = FS_STATE_IDX_VALID;
+ state->fe_map.entries[in->index].old_idx = in->index;
+ state->fe_map.entries[in->index].new_idx = fetoi(out);
+
+ done:
+ return ret;
+}
+
+static int
+cb_stateCBToDiskEntry(struct CallBack * in, struct CBDiskEntry * out)
+{
+ memcpy(&out->cb, in, sizeof(struct CallBack));
+ out->index = cbtoi(in);
+ return 0;
+}
+
+static int
+cb_stateDiskEntryToCB(struct fs_dump_state * state,
+ struct CBDiskEntry * in, struct CallBack * out)
+{
+ int ret = 0;
+
+ memcpy(out, &in->cb, sizeof(struct CallBack));
+
+ /* setup CB map entry */
+ if (!in->index || (in->index >= state->cb_map.len)) {
+ ViceLog(0, ("cb_stateDiskEntryToCB: index (%d) out of range\n",
+ in->index));
+ ret = 1;
+ goto done;
+ }
+ state->cb_map.entries[in->index].valid = FS_STATE_IDX_VALID;
+ state->cb_map.entries[in->index].old_idx = in->index;
+ state->cb_map.entries[in->index].new_idx = cbtoi(out);
+
+ done:
+ return ret;
+}
+
+/* index map routines */
+static int
+cb_stateAllocMap(struct fs_dump_state * state)
+{
+ state->fe_map.len = state->cb_hdr->fe_max + 1;
+ state->cb_map.len = state->cb_hdr->cb_max + 1;
+ state->fe_map.entries = (struct idx_map_entry_t *)
+ calloc(state->fe_map.len, sizeof(struct idx_map_entry_t));
+ state->cb_map.entries = (struct idx_map_entry_t *)
+ calloc(state->cb_map.len, sizeof(struct idx_map_entry_t));
+ return ((state->fe_map.entries != NULL) && (state->cb_map.entries != NULL)) ? 0 : 1;
+}
+
+int
+fe_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
+{
+ int ret = 0;
+
+ /* FEs use a one-based indexing system, so old==0 implies no mapping */
+ if (!old) {
+ *new = 0;
+ goto done;
+ }
+
+ if (old >= state->fe_map.len) {
+ ViceLog(0, ("fe_OldToNew: index %d is out of range\n", old));
+ ret = 1;
+ } else if (state->fe_map.entries[old].valid != FS_STATE_IDX_VALID ||
+ state->fe_map.entries[old].old_idx != old) { /* sanity check */
+ ViceLog(0, ("fe_OldToNew: index %d points to an invalid FileEntry record\n", old));
+ ret = 1;
+ } else {
+ *new = state->fe_map.entries[old].new_idx;
+ }
+
+ done:
+ return ret;
+}
+
+int
+cb_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
+{
+ int ret = 0;
+
+ /* CBs use a one-based indexing system, so old==0 implies no mapping */
+ if (!old) {
+ *new = 0;
+ goto done;
+ }
+
+ if (old >= state->cb_map.len) {
+ ViceLog(0, ("cb_OldToNew: index %d is out of range\n", old));
+ ret = 1;
+ } else if (state->cb_map.entries[old].valid != FS_STATE_IDX_VALID ||
+ state->cb_map.entries[old].old_idx != old) { /* sanity check */
+ ViceLog(0, ("cb_OldToNew: index %d points to an invalid CallBack record\n", old));
+ ret = 1;
+ } else {
+ *new = state->cb_map.entries[old].new_idx;
+ }
+
+ done:
+ return ret;
+}
+#endif /* AFS_DEMAND_ATTACH_FS */
+
+#define DumpBytes(fd,buf,req) if (write(fd, buf, req) < 0) ; /* don't care */
+
+static int
+DumpCallBackState_r(void)
+{
+ int fd, oflag;
+ afs_uint32 magic = MAGICV2, now = (afs_int32) time(NULL), freelisthead;
+
+ oflag = O_WRONLY | O_CREAT | O_TRUNC;
+#ifdef AFS_NT40_ENV
+ oflag |= O_BINARY;
+#endif
+ fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, oflag, 0666);
+ if (fd < 0) {
+ ViceLog(0,
+ ("Couldn't create callback dump file %s\n",
+ AFSDIR_SERVER_CBKDUMP_FILEPATH));
+ return 0;
+ }
+ /*
+ * Collect but ignoring the return value of write(2) here,
+ * to avoid compiler warnings on some platforms.
+ */
+ DumpBytes(fd, &magic, sizeof(magic));
+ DumpBytes(fd, &now, sizeof(now));
+ DumpBytes(fd, &cbstuff, sizeof(cbstuff));
+ DumpBytes(fd, TimeOuts, sizeof(TimeOuts));
+ DumpBytes(fd, timeout, sizeof(timeout));
+ DumpBytes(fd, &tfirst, sizeof(tfirst));
+ freelisthead = cbtoi((struct CallBack *)CBfree);
+ DumpBytes(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
+ freelisthead = fetoi((struct FileEntry *)FEfree);
+ DumpBytes(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
+ DumpBytes(fd, HashTable, sizeof(HashTable));
+ DumpBytes(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
+ DumpBytes(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
+ close(fd);
+
+ return 0;
+}
+
+int
+DumpCallBackState(void) {
+ int rc;
+
+ H_LOCK;
+ rc = DumpCallBackState_r();
+ H_UNLOCK;
+
+ return(rc);
+}
+
+#endif /* !INTERPRET_DUMP */
+
+#ifdef INTERPRET_DUMP
+
+static void
+ReadBytes(int fd, void *buf, size_t req)
+{
+ ssize_t count;
+
+ count = read(fd, buf, req);
+ if (count < 0) {
+ perror("read");
+ exit(-1);
+ } else if (count != req) {
+ fprintf(stderr, "read: premature EOF (expected %lu, got %lu)\n",
+ (unsigned long)req, (unsigned long)count);
+ exit(-1);
+ }
+}
+
+/* This is only compiled in for the callback analyzer program */
+/* Returns the time of the dump */
+time_t
+ReadDump(char *file, int timebits)
+{
+ int fd, oflag;
+ afs_uint32 magic, freelisthead;
+ afs_uint32 now;
+ afs_int64 now64;
+
+ oflag = O_RDONLY;
+#ifdef AFS_NT40_ENV
+ oflag |= O_BINARY;
+#endif
+ fd = open(file, oflag);
if (fd < 0) {
fprintf(stderr, "Couldn't read dump file %s\n", file);
exit(1);
}
- read(fd, &magic, sizeof(magic));
- if (magic != MAGIC) {
- fprintf(stderr,
- "Magic number of %s is invalid. You might be trying to\n",
- file);
- fprintf(stderr,
- "run this program on a machine type with a different byte ordering.\n");
- exit(1);
+ ReadBytes(fd, &magic, sizeof(magic));
+ if (magic == MAGICV2) {
+ timebits = 32;
+ } else {
+ if (magic != MAGIC) {
+ fprintf(stderr,
+ "Magic number of %s is invalid. You might be trying to\n",
+ file);
+ fprintf(stderr,
+ "run this program on a machine type with a different byte ordering.\n");
+ exit(1);
+ }
}
- read(fd, &now, sizeof(now));
- read(fd, &cbstuff, sizeof(cbstuff));
- read(fd, TimeOuts, sizeof(TimeOuts));
- read(fd, timeout, sizeof(timeout));
- read(fd, &tfirst, sizeof(tfirst));
- read(fd, &freelisthead, sizeof(freelisthead));
+ if (timebits == 64) {
+ ReadBytes(fd, &now64, sizeof(afs_int64));
+ now = (afs_int32) now64;
+ } else
+ ReadBytes(fd, &now, sizeof(afs_int32));
+
+ ReadBytes(fd, &cbstuff, sizeof(cbstuff));
+ ReadBytes(fd, TimeOuts, sizeof(TimeOuts));
+ ReadBytes(fd, timeout, sizeof(timeout));
+ ReadBytes(fd, &tfirst, sizeof(tfirst));
+ ReadBytes(fd, &freelisthead, sizeof(freelisthead));
CB = ((struct CallBack
- *)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
+ *)(calloc(cbstuff.nblks, sizeof(struct CallBack)))) - 1;
FE = ((struct FileEntry
*)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
CBfree = (struct CallBack *)itocb(freelisthead);
- read(fd, &freelisthead, sizeof(freelisthead));
+ ReadBytes(fd, &freelisthead, sizeof(freelisthead));
FEfree = (struct FileEntry *)itofe(freelisthead);
- read(fd, HashTable, sizeof(HashTable));
- read(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
- read(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
+ ReadBytes(fd, HashTable, sizeof(HashTable));
+ ReadBytes(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
+ ReadBytes(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
if (close(fd)) {
perror("Error reading dumpfile");
exit(1);
return now;
}
+#ifdef AFS_NT40_ENV
+#include "AFS_component_version_number.h"
+#else
#include "AFS_component_version_number.c"
+#endif
+
+static afs_uint32 *cbTrack;
int
main(int argc, char **argv)
{
int err = 0, cbi = 0, stats = 0, noptions = 0, all = 0, vol = 0, raw = 0;
static AFSFid fid;
- register struct FileEntry *fe;
- register struct CallBack *cb;
+ struct FileEntry *fe;
+ struct CallBack *cb;
time_t now;
+ int timebits = 32;
memset(&fid, 0, sizeof(fid));
argc--;
all = 1;
} else if (!strcmp(*argv, "-raw")) {
raw = 1;
+ } else if (!strcmp(*argv, "-timebits")) {
+ if (argc < 1) {
+ err++;
+ break;
+ }
+ argc--;
+ timebits = atoi(*++argv);
+ if ((timebits != 32)
+ && (timebits != 64)
+ )
+ err++;
} else if (!strcmp(*argv, "-volume")) {
if (argc < 1) {
err++;
}
if (err || argc != 1) {
fprintf(stderr,
- "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] callbackdumpfile\n");
+ "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] [-timebits 32"
+ "|64"
+ "] callbackdumpfile\n");
fprintf(stderr,
"[cbid is shown for each host in the hosts.dump file]\n");
exit(1);
}
- now = ReadDump(*argv);
+ now = ReadDump(*argv, timebits);
if (stats || noptions == 0) {
- time_t uxtfirst = UXtime(tfirst);
- printf("The time of the dump was %u %s", now, ctime(&now));
- printf("The last time cleanup ran was %u %s", uxtfirst,
+ time_t uxtfirst = UXtime(tfirst), tnow = now;
+ printf("The time of the dump was %u %s", (unsigned int) now, ctime(&tnow));
+ printf("The last time cleanup ran was %u %s", (unsigned int) uxtfirst,
ctime(&uxtfirst));
PrintCallBackStats();
}
+
+ cbTrack = calloc(cbstuff.nblks, sizeof(cbTrack[0]));
+
if (all || vol) {
- register hash;
- register afs_uint32 *feip;
- register struct CallBack *cb;
- register struct FileEntry *fe;
+ int hash;
+ afs_uint32 *feip;
+ struct CallBack *cb;
+ struct FileEntry *fe;
- for (hash = 0; hash < VHASH; hash++) {
- for (feip = &HashTable[hash]; fe = itofe(*feip);) {
+ for (hash = 0; hash < FEHASH_SIZE; hash++) {
+ for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
if (!vol || (fe->volid == vol)) {
- register struct CallBack *cbnext;
- for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
+ afs_uint32 fe_i = fetoi(fe);
+
+ for (cb = itocb(fe->firstcb); cb; cb = itocb(cb->cnext)) {
+ afs_uint32 cb_i = cbtoi(cb);
+
+ if (cb_i > cbstuff.nblks) {
+ printf("CB index out of range (%u > %d), stopped for this FE\n",
+ cb_i, cbstuff.nblks);
+ break;
+ }
+
+ if (cbTrack[cb_i]) {
+ printf("CB entry already claimed for FE[%u] (this is FE[%u]), stopped\n",
+ cbTrack[cb_i], fe_i);
+ break;
+ }
+ cbTrack[cb_i] = fe_i;
+
PrintCB(cb, now);
- cbnext = itocb(cb->cnext);
}
*feip = fe->fnext;
} else {
}
}
if (raw) {
- struct FileEntry *fe;
afs_int32 *p, i;
for (i = 1; i < cbstuff.nblks; i++) {
p = (afs_int32 *) & FE[i];
printf("%d:%12x%12x%12x%12x\n", i, p[0], p[1], p[2], p[3]);
}
}
+
+ free(cbTrack);
+ exit(0);
}
-int
-PrintCB(register struct CallBack *cb, afs_uint32 now)
+void
+PrintCB(struct CallBack *cb, afs_uint32 now)
{
struct FileEntry *fe = itofe(cb->fhead);
time_t expires = TIndexToTime(cb->thead);
- printf("vol=%u vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %d secs at %s",
- fe->volid, fe->vnode, fe->ncbs, cb->hhead, cb->status, fe->status,
- expires - now, ctime(&expires));
+ if (fe == NULL)
+ return;
+
+ printf("vol=%" AFS_VOLID_FMT " vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %lu secs at %s",
+ afs_printable_VolumeId_lu(fe->volid), fe->vnode, fe->ncbs,
+ cb->hhead, cb->status, fe->status, expires - now, ctime(&expires));
}
#endif
int i, j;
struct rx_connection **conns;
struct rx_connection *connSuccess = 0;
- afs_int32 *addr;
+ struct AddrPort *interfaces;
static struct rx_securityClass *sc = 0;
static struct AFSCBs tc = { 0, 0 };
char hoststr[16];
if (!host->interface)
return 1; /* failure */
- assert(host->interface->numberOfInterfaces > 0);
-
/* the only address is the primary interface */
- if (host->interface->numberOfInterfaces == 1)
+ if (host->interface->numberOfInterfaces <= 1)
return 1; /* failure */
/* initialise a security object only once */
sc = rxnull_NewClientSecurityObject();
i = host->interface->numberOfInterfaces;
- addr = calloc(i, sizeof(afs_int32));
+ interfaces = calloc(i, sizeof(struct AddrPort));
conns = calloc(i, sizeof(struct rx_connection *));
- if (!addr || !conns) {
- ViceLog(0,
- ("Failed malloc in MultiBreakCallBackAlternateAddress_r\n"));
- assert(0);
+ if (!interfaces || !conns) {
+ ViceLogThenPanic(0, ("Failed malloc in "
+ "MultiBreakCallBackAlternateAddress_r\n"));
}
/* initialize alternate rx connections */
for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
/* this is the current primary address */
- if (host->host == host->interface->addr[i])
+ if (host->host == host->interface->interface[i].addr &&
+ host->port == host->interface->interface[i].port)
continue;
- addr[j] = host->interface->addr[i];
+ interfaces[j] = host->interface->interface[i];
conns[j] =
- rx_NewConnection(host->interface->addr[i], host->port, 1, sc, 0);
+ rx_NewConnection(interfaces[j].addr,
+ interfaces[j].port, 1, sc, 0);
rx_SetConnDeadTime(conns[j], 2);
rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
j++;
}
- assert(j); /* at least one alternate address */
+ opr_Assert(j); /* at least one alternate address */
ViceLog(125,
- ("Starting multibreakcall back on all addr for host %s\n",
- afs_inet_ntoa_r(host->host, hoststr)));
+ ("Starting multibreakcall back on all addr for host %p (%s:%d)\n",
+ host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
H_UNLOCK;
multi_Rx(conns, j) {
multi_RXAFSCB_CallBack(afidp, &tc);
if (host->callback_rxcon)
rx_DestroyConnection(host->callback_rxcon);
host->callback_rxcon = conns[multi_i];
- host->host = addr[multi_i];
+ /* add then remove */
+ addInterfaceAddr_r(host, interfaces[multi_i].addr,
+ interfaces[multi_i].port);
+ removeInterfaceAddr_r(host, host->host, host->port);
+ host->host = interfaces[multi_i].addr;
+ host->port = interfaces[multi_i].port;
connSuccess = conns[multi_i];
rx_SetConnDeadTime(host->callback_rxcon, 50);
rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
ViceLog(125,
- ("multibreakcall success with addr %s\n",
- afs_inet_ntoa_r(addr[multi_i], hoststr)));
+ ("multibreakcall success with addr %s:%d\n",
+ afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
+ ntohs(interfaces[multi_i].port)));
H_UNLOCK;
multi_Abort;
}
if (conns[i] != connSuccess)
rx_DestroyConnection(conns[i]);
- free(addr);
+ free(interfaces);
free(conns);
if (connSuccess)
/*
-** try multiRX probes to host.
-** return 0 on success, non-zero on failure
+** try multi_RX probes to host.
+** return 0 on success, non-0 on failure
*/
int
MultiProbeAlternateAddress_r(struct host *host)
int i, j;
struct rx_connection **conns;
struct rx_connection *connSuccess = 0;
- afs_int32 *addr;
+ struct AddrPort *interfaces;
static struct rx_securityClass *sc = 0;
char hoststr[16];
if (!host->interface)
return 1; /* failure */
- assert(host->interface->numberOfInterfaces > 0);
-
/* the only address is the primary interface */
- if (host->interface->numberOfInterfaces == 1)
- return 1; /* failure */
+ if (host->interface->numberOfInterfaces <= 1)
+ return 1; /* failure */
/* initialise a security object only once */
if (!sc)
sc = rxnull_NewClientSecurityObject();
i = host->interface->numberOfInterfaces;
- addr = calloc(i, sizeof(afs_int32));
+ interfaces = calloc(i, sizeof(struct AddrPort));
conns = calloc(i, sizeof(struct rx_connection *));
- if (!addr || !conns) {
- ViceLog(0, ("Failed malloc in MultiProbeAlternateAddress_r\n"));
- assert(0);
+ if (!interfaces || !conns) {
+ ViceLogThenPanic(0, ("Failed malloc in "
+ "MultiProbeAlternateAddress_r\n"));
}
/* initialize alternate rx connections */
for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
/* this is the current primary address */
- if (host->host == host->interface->addr[i])
+ if (host->host == host->interface->interface[i].addr &&
+ host->port == host->interface->interface[i].port)
continue;
- addr[j] = host->interface->addr[i];
+ interfaces[j] = host->interface->interface[i];
conns[j] =
- rx_NewConnection(host->interface->addr[i], host->port, 1, sc, 0);
+ rx_NewConnection(interfaces[j].addr,
+ interfaces[j].port, 1, sc, 0);
rx_SetConnDeadTime(conns[j], 2);
rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
j++;
}
- assert(j); /* at least one alternate address */
+ opr_Assert(j); /* at least one alternate address */
ViceLog(125,
- ("Starting multiprobe on all addr for host %s\n",
- afs_inet_ntoa_r(host->host, hoststr)));
+ ("Starting multiprobe on all addr for host %p (%s:%d)\n",
+ host, afs_inet_ntoa_r(host->host, hoststr),
+ ntohs(host->port)));
H_UNLOCK;
multi_Rx(conns, j) {
multi_RXAFSCB_ProbeUuid(&host->interface->uuid);
if (host->callback_rxcon)
rx_DestroyConnection(host->callback_rxcon);
host->callback_rxcon = conns[multi_i];
- host->host = addr[multi_i];
+ /* add then remove */
+ addInterfaceAddr_r(host, interfaces[multi_i].addr,
+ interfaces[multi_i].port);
+ removeInterfaceAddr_r(host, host->host, host->port);
+ host->host = interfaces[multi_i].addr;
+ host->port = interfaces[multi_i].port;
connSuccess = conns[multi_i];
rx_SetConnDeadTime(host->callback_rxcon, 50);
rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
ViceLog(125,
- ("multiprobe success with addr %s\n",
- afs_inet_ntoa_r(addr[multi_i], hoststr)));
+ ("multiprobe success with addr %s:%d\n",
+ afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
+ ntohs(interfaces[multi_i].port)));
H_UNLOCK;
multi_Abort;
+ } else {
+ ViceLog(125,
+ ("multiprobe failure with addr %s:%d\n",
+ afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
+ ntohs(interfaces[multi_i].port)));
+
+ /* This is less than desirable but its the best we can do.
+ * The AFS Cache Manager will return either 0 for a Uuid
+ * match and a 1 for a non-match. If the error is 1 we
+ * therefore know that our mapping of IP address to Uuid
+ * is wrong. We should attempt to find the correct
+ * Uuid and fix the host tables.
+ */
+ if (multi_error == 1) {
+ /* remove the current alternate address from this host */
+ H_LOCK;
+ removeInterfaceAddr_r(host, interfaces[multi_i].addr, interfaces[multi_i].port);
+ H_UNLOCK;
+ }
+ }
+#ifdef AFS_DEMAND_ATTACH_FS
+ /* try to bail ASAP if the fileserver is shutting down */
+ FS_STATE_RDLOCK;
+ if (fs_state.mode == FS_MODE_SHUTDOWN) {
+ FS_STATE_UNLOCK;
+ multi_Abort;
}
+ FS_STATE_UNLOCK;
+#endif
}
multi_End_Ignore;
H_LOCK;
if (conns[i] != connSuccess)
rx_DestroyConnection(conns[i]);
- free(addr);
+ free(interfaces);
free(conns);
if (connSuccess)
git://git.openafs.org / openafs.git / blobdiff