/*------------------------------------------------------------------------
* afs_BlackListOnce
*
* Description:
* Mark a server as invalid for further attempts of this request only.
*
* Arguments:
* areq : The request record associated with this operation.
* afid : The FID of the file involved in the action. This argument
* may be null if none was involved.
* tsp : pointer to a server struct for the server we wish to
* blacklist.
*
* Returns:
* Non-zero value if further servers are available to try,
* zero otherwise.
*
* Environment:
* This routine is typically called in situations where we believe
* one server out of a pool may have an error condition.
*
* Side Effects:
* As advertised.
*
* NOTE:
* The afs_Conn* routines use the list of invalidated servers to
* avoid reusing a server marked as invalid for this request.
*------------------------------------------------------------------------*/
static afs_int32
afs_BlackListOnce(struct vrequest *areq, struct VenusFid *afid,
struct server *tsp)
{
struct volume *tvp;
afs_int32 i;
afs_int32 serversleft = 0;
if (afid) {
tvp = afs_FindVolume(afid, READ_LOCK);
if (tvp) {
for (i = 0; i < MAXHOSTS; i++) {
if (tvp->serverHost[i] == tsp) {
areq->skipserver[i] = 1;
}
if (tvp->serverHost[i] &&
(tvp->serverHost[i]->addr->sa_flags &
SRVR_ISDOWN)) {
areq->skipserver[i] = 1;
}
}
afs_PutVolume(tvp, READ_LOCK);
for (i = 0; i < MAXHOSTS; i++) {
if (tvp->serverHost[i] && areq->skipserver[i] == 0) {
serversleft = 1;
break;
}
}
return serversleft;
}
}
return 1;
}
/**
* Note that areq may be null, in which case we don't bother to set any
* request status information.
* @param afid Volume FID.
* @param areq Request type.
* @param locktype Lock to be used.
* @return Volume or NULL if no result.
*/
struct volume *
afs_GetVolume(struct VenusFid *afid, struct vrequest *areq,
afs_int32 locktype)
{
struct volume *tv;
char *bp, tbuf[CVBS];
AFS_STATCNT(afs_GetVolume);
tv = afs_FindVolume(afid, locktype);
if (!tv) {
/* Do a dynroot check and add dynroot volume if found. */
if (afs_IsDynrootAnyFid(afid)) {
tv = afs_NewDynrootVolume(afid);
} else {
bp = afs_cv2string(&tbuf[CVBS], afid->Fid.Volume);
tv = afs_NewVolumeByName(bp, afid->Cell, 0, areq, locktype);
}
}
return tv;
} /*afs_GetVolume */
/* question: does afs_create need to set CDirty in the adp or the avc?
* I think we can get away without it, but I'm not sure. Note that
* afs_setattr is called in here for truncation.
*/
#ifdef AFS_SGI64_ENV
int
afs_create(OSI_VC_DECL(adp), char *aname, struct vattr *attrs, int flags,
int amode, struct vcache **avcp, afs_ucred_t *acred)
#else /* AFS_SGI64_ENV */
int
afs_create(OSI_VC_DECL(adp), char *aname, struct vattr *attrs,
enum vcexcl aexcl, int amode, struct vcache **avcp,
afs_ucred_t *acred)
#endif /* AFS_SGI64_ENV */
{
afs_int32 origCBs, origZaps, finalZaps;
struct vrequest *treq = NULL;
afs_int32 code;
struct afs_conn *tc;
struct VenusFid newFid;
struct AFSStoreStatus InStatus;
struct AFSFetchStatus *OutFidStatus, *OutDirStatus;
struct AFSVolSync tsync;
struct AFSCallBack CallBack;
afs_int32 now;
struct dcache *tdc;
afs_size_t offset, len;
struct server *hostp = 0;
struct vcache *tvc;
struct volume *volp = 0;
struct afs_fakestat_state fakestate;
struct rx_connection *rxconn;
XSTATS_DECLS;
OSI_VC_CONVERT(adp);
AFS_STATCNT(afs_create);
OutFidStatus = osi_AllocSmallSpace(sizeof(struct AFSFetchStatus));
OutDirStatus = osi_AllocSmallSpace(sizeof(struct AFSFetchStatus));
memset(&InStatus, 0, sizeof(InStatus));
if ((code = afs_CreateReq(&treq, acred)))
goto done2;
afs_Trace3(afs_iclSetp, CM_TRACE_CREATE, ICL_TYPE_POINTER, adp,
ICL_TYPE_STRING, aname, ICL_TYPE_INT32, amode);
afs_InitFakeStat(&fakestate);
#ifdef AFS_SGI65_ENV
/* If avcp is passed not null, it's the old reference to this file.
* We can use this to avoid create races. For now, just decrement
* the reference count on it.
*/
if (*avcp) {
AFS_RELE(AFSTOV(*avcp));
*avcp = NULL;
}
#endif
if (strlen(aname) > AFSNAMEMAX) {
code = ENAMETOOLONG;
goto done3;
}
if (!afs_ENameOK(aname)) {
code = EINVAL;
goto done3;
}
switch (attrs->va_type) {
case VBLK:
case VCHR:
#if !defined(AFS_SUN5_ENV)
case VSOCK:
#endif
case VFIFO:
/* We don't support special devices or FIFOs */
code = EINVAL;
goto done3;
default:
;
}
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&adp, &fakestate, treq);
if (code)
goto done;
tagain:
code = afs_VerifyVCache(adp, treq);
if (code)
goto done;
/** If the volume is read-only, return error without making an RPC to the
* fileserver
*/
if (adp->f.states & CRO) {
code = EROFS;
goto done;
}
if (AFS_IS_DISCONNECTED && !AFS_IS_DISCON_RW) {
code = ENETDOWN;
goto done;
}
tdc = afs_GetDCache(adp, (afs_size_t) 0, treq, &offset, &len, 1);
ObtainWriteLock(&adp->lock, 135);
if (tdc)
ObtainSharedLock(&tdc->lock, 630);
/*
* Make sure that the data in the cache is current. We may have
* received a callback while we were waiting for the write lock.
*/
if (!(adp->f.states & CStatd)
|| (tdc && !hsame(adp->f.m.DataVersion, tdc->f.versionNo))) {
ReleaseWriteLock(&adp->lock);
if (tdc) {
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
goto tagain;
}
if (tdc) {
/* see if file already exists. If it does, we only set
* the size attributes (to handle O_TRUNC) */
code = afs_dir_Lookup(tdc, aname, &newFid.Fid); /* use dnlc first xxx */
if (code == 0) {
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
ReleaseWriteLock(&adp->lock);
#ifdef AFS_SGI64_ENV
if (flags & VEXCL) {
#else
if (aexcl != NONEXCL) {
#endif
code = EEXIST; /* file exists in excl mode open */
goto done;
}
/* found the file, so use it */
newFid.Cell = adp->f.fid.Cell;
newFid.Fid.Volume = adp->f.fid.Fid.Volume;
tvc = NULL;
if (newFid.Fid.Unique == 0) {
tvc = afs_LookupVCache(&newFid, treq, NULL, adp, aname);
}
if (!tvc) /* lookup failed or wasn't called */
tvc = afs_GetVCache(&newFid, treq, NULL, NULL);
if (tvc) {
/* if the thing exists, we need the right access to open it.
* we must check that here, since no other checks are
* made by the open system call */
len = attrs->va_size; /* only do the truncate */
/*
* We used to check always for READ access before; the
* problem is that we will fail if the existing file
* has mode -w-w-w, which is wrong.
*/
if ((amode & VREAD)
&& !afs_AccessOK(tvc, PRSFS_READ, treq, CHECK_MODE_BITS)) {
afs_PutVCache(tvc);
code = EACCES;
goto done;
}
#if defined(AFS_DARWIN80_ENV)
if ((amode & VWRITE) || VATTR_IS_ACTIVE(attrs, va_data_size))
#elif defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
if ((amode & VWRITE) || (attrs->va_mask & AT_SIZE))
#else
if ((amode & VWRITE) || len != 0xffffffff)
#endif
{
/* needed for write access check */
tvc->f.parent.vnode = adp->f.fid.Fid.Vnode;
tvc->f.parent.unique = adp->f.fid.Fid.Unique;
/* need write mode for these guys */
if (!afs_AccessOK
(tvc, PRSFS_WRITE, treq, CHECK_MODE_BITS)) {
afs_PutVCache(tvc);
code = EACCES;
goto done;
}
}
#if defined(AFS_DARWIN80_ENV)
if (VATTR_IS_ACTIVE(attrs, va_data_size))
#elif defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
if (attrs->va_mask & AT_SIZE)
#else
if (len != 0xffffffff)
#endif
{
if (vType(tvc) != VREG) {
afs_PutVCache(tvc);
code = EISDIR;
goto done;
}
/* do a truncate */
#if defined(AFS_DARWIN80_ENV)
VATTR_INIT(attrs);
VATTR_SET_SUPPORTED(attrs, va_data_size);
VATTR_SET_ACTIVE(attrs, va_data_size);
#elif defined(UKERNEL)
attrs->va_mask = ATTR_SIZE;
#elif defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
attrs->va_mask = AT_SIZE;
#else
VATTR_NULL(attrs);
#endif
attrs->va_size = len;
ObtainWriteLock(&tvc->lock, 136);
tvc->f.states |= CCreating;
ReleaseWriteLock(&tvc->lock);
#if defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
#if defined(AFS_SGI64_ENV)
code =
afs_setattr(VNODE_TO_FIRST_BHV((vnode_t *) tvc),
attrs, 0, acred);
#else
code = afs_setattr(tvc, attrs, 0, acred);
#endif /* AFS_SGI64_ENV */
#else /* SUN5 || SGI */
code = afs_setattr(tvc, attrs, acred);
#endif /* SUN5 || SGI */
ObtainWriteLock(&tvc->lock, 137);
tvc->f.states &= ~CCreating;
ReleaseWriteLock(&tvc->lock);
if (code) {
afs_PutVCache(tvc);
goto done;
}
}
*avcp = tvc;
} else
code = ENOENT; /* shouldn't get here */
/* make sure vrefCount bumped only if code == 0 */
goto done;
}
}
/* if we create the file, we don't do any access checks, since
* that's how O_CREAT is supposed to work */
if (adp->f.states & CForeign) {
origCBs = afs_allCBs;
origZaps = afs_allZaps;
} else {
origCBs = afs_evenCBs; /* if changes, we don't really have a callback */
origZaps = afs_evenZaps; /* number of even numbered vnodes discarded */
}
InStatus.Mask = AFS_SETMODTIME | AFS_SETMODE | AFS_SETGROUP;
InStatus.ClientModTime = osi_Time();
InStatus.Group = (afs_int32) afs_cr_gid(acred);
if (AFS_NFSXLATORREQ(acred)) {
/*
* XXX The following is mainly used to fix a bug in the HP-UX
* nfs client where they create files with mode of 0 without
* doing any setattr later on to fix it. * XXX
*/
#if defined(AFS_AIX_ENV)
if (attrs->va_mode != -1) {
#else
#if defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
if (attrs->va_mask & AT_MODE) {
#else
if (attrs->va_mode != ((unsigned short)-1)) {
#endif
#endif
if (!attrs->va_mode)
attrs->va_mode = 0x1b6; /* XXX default mode: rw-rw-rw XXX */
}
}
if (!AFS_IS_DISCONNECTED) {
/* If not disconnected, connect to the server.*/
InStatus.UnixModeBits = attrs->va_mode & 0xffff; /* only care about protection bits */
do {
tc = afs_Conn(&adp->f.fid, treq, SHARED_LOCK, &rxconn);
if (tc) {
hostp = tc->srvr->server; /* remember for callback processing */
now = osi_Time();
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_CREATEFILE);
RX_AFS_GUNLOCK();
code =
RXAFS_CreateFile(rxconn, (struct AFSFid *)&adp->f.fid.Fid,
aname, &InStatus, (struct AFSFid *)
&newFid.Fid, OutFidStatus, OutDirStatus,
&CallBack, &tsync);
RX_AFS_GLOCK();
XSTATS_END_TIME;
CallBack.ExpirationTime += now;
} else
code = -1;
} while (afs_Analyze
(tc, rxconn, code, &adp->f.fid, treq, AFS_STATS_FS_RPCIDX_CREATEFILE,
SHARED_LOCK, NULL));
if ((code == EEXIST || code == UAEEXIST) &&
#ifdef AFS_SGI64_ENV
!(flags & VEXCL)
#else /* AFS_SGI64_ENV */
aexcl == NONEXCL
#endif
) {
/* if we get an EEXIST in nonexcl mode, just do a lookup */
if (tdc) {
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
ReleaseWriteLock(&adp->lock);
#if defined(AFS_SGI64_ENV)
code = afs_lookup(VNODE_TO_FIRST_BHV((vnode_t *) adp), aname, avcp,
NULL, 0, NULL, acred);
#elif defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
code = afs_lookup(adp, aname, avcp, NULL, 0, NULL, acred);
#elif defined(UKERNEL)
code = afs_lookup(adp, aname, avcp, acred, 0);
#elif !defined(AFS_DARWIN_ENV)
code = afs_lookup(adp, aname, avcp, acred);
#endif
goto done;
}
if (code) {
if (code < 0) {
ObtainWriteLock(&afs_xcbhash, 488);
afs_DequeueCallback(adp);
adp->f.states &= ~CStatd;
ReleaseWriteLock(&afs_xcbhash);
osi_dnlc_purgedp(adp);
}
ReleaseWriteLock(&adp->lock);
if (tdc) {
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
goto done;
}
} else {
/* Generate a fake FID for disconnected mode. */
newFid.Cell = adp->f.fid.Cell;
newFid.Fid.Volume = adp->f.fid.Fid.Volume;
afs_GenFakeFid(&newFid, VREG, 1);
} /* if (!AFS_IS_DISCON_RW) */
/* otherwise, we should see if we can make the change to the dir locally */
if (tdc)
UpgradeSToWLock(&tdc->lock, 631);
if (AFS_IS_DISCON_RW || afs_LocalHero(adp, tdc, OutDirStatus, 1)) {
/* we can do it locally */
ObtainWriteLock(&afs_xdcache, 291);
code = afs_dir_Create(tdc, aname, &newFid.Fid);
ReleaseWriteLock(&afs_xdcache);
if (code) {
ZapDCE(tdc);
DZap(tdc);
}
}
if (tdc) {
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc);
}
if (AFS_IS_DISCON_RW)
adp->f.m.LinkCount++;
newFid.Cell = adp->f.fid.Cell;
newFid.Fid.Volume = adp->f.fid.Fid.Volume;
ReleaseWriteLock(&adp->lock);
volp = afs_FindVolume(&newFid, READ_LOCK);
/* New tricky optimistic callback handling algorithm for file creation works
* as follows. We create the file essentially with no locks set at all. File
* server may thus handle operations from others cache managers as well as from
* this very own cache manager that reference the file in question before
* we managed to create the cache entry. However, if anyone else changes
* any of the status information for a file, we'll see afs_evenCBs increase
* (files always have even fids). If someone on this workstation manages
* to do something to the file, they'll end up having to create a cache
* entry for the new file. Either we'll find it once we've got the afs_xvcache
* lock set, or it was also *deleted* the vnode before we got there, in which case
* we will find evenZaps has changed, too. Thus, we only assume we have the right
* status information if no callbacks or vnode removals have occurred to even
* numbered files from the time the call started until the time that we got the xvcache
* lock set. Of course, this also assumes that any call that modifies a file first
* gets a write lock on the file's vnode, but if that weren't true, the whole cache manager
* would fail, since no call would be able to update the local vnode status after modifying
* a file on a file server. */
ObtainWriteLock(&afs_xvcache, 138);
if (adp->f.states & CForeign)
finalZaps = afs_allZaps; /* do this before calling newvcache */
else
finalZaps = afs_evenZaps; /* do this before calling newvcache */
/* don't need to call RemoveVCB, since only path leaving a callback is the
* one where we pass through afs_NewVCache. Can't have queued a VCB unless
* we created and freed an entry between file creation time and here, and the
* freeing of the vnode will change evenZaps. Don't need to update the VLRU
* queue, since the find will only succeed in the event of a create race, and
* then the vcache will be at the front of the VLRU queue anyway... */
if (!(tvc = afs_FindVCache(&newFid, 0, DO_STATS))) {
tvc = afs_NewVCache(&newFid, hostp);
if (tvc) {
int finalCBs;
ObtainWriteLock(&tvc->lock, 139);
ObtainWriteLock(&afs_xcbhash, 489);
finalCBs = afs_evenCBs;
/* add the callback in */
if (adp->f.states & CForeign) {
tvc->f.states |= CForeign;
finalCBs = afs_allCBs;
}
if (origCBs == finalCBs && origZaps == finalZaps) {
tvc->f.states |= CStatd; /* we've fake entire thing, so don't stat */
tvc->f.states &= ~CBulkFetching;
if (!AFS_IS_DISCON_RW) {
tvc->cbExpires = CallBack.ExpirationTime;
afs_QueueCallback(tvc, CBHash(CallBack.ExpirationTime), volp);
}
} else {
afs_DequeueCallback(tvc);
tvc->f.states &= ~(CStatd | CUnique);
tvc->callback = 0;
if (tvc->f.fid.Fid.Vnode & 1 || (vType(tvc) == VDIR))
osi_dnlc_purgedp(tvc);
}
ReleaseWriteLock(&afs_xcbhash);
if (AFS_IS_DISCON_RW) {
afs_DisconAddDirty(tvc, VDisconCreate, 0);
afs_GenDisconStatus(adp, tvc, &newFid, attrs, treq, VREG);
} else {
afs_ProcessFS(tvc, OutFidStatus, treq);
}
tvc->f.parent.vnode = adp->f.fid.Fid.Vnode;
tvc->f.parent.unique = adp->f.fid.Fid.Unique;
#if !defined(UKERNEL)
if (volp && (volp->states & VPartVisible))
tvc->f.states |= CPartVisible;
#endif
ReleaseWriteLock(&tvc->lock);
*avcp = tvc;
code = 0;
} else
code = ENOENT;
} else {
/* otherwise cache entry already exists, someone else must
* have created it. Comments used to say: "don't need write
* lock to *clear* these flags" but we should do it anyway.
* Code used to clear stat bit and callback, but I don't see
* the point -- we didn't have a create race, somebody else just
* snuck into NewVCache before we got here, probably a racing
* lookup.
*/
*avcp = tvc;
code = 0;
}
ReleaseWriteLock(&afs_xvcache);
done:
AFS_DISCON_UNLOCK();
done3:
if (volp)
afs_PutVolume(volp, READ_LOCK);
if (code == 0) {
if (afs_mariner)
afs_AddMarinerName(aname, *avcp);
/* return the new status in vattr */
afs_CopyOutAttrs(*avcp, attrs);
if (afs_mariner)
afs_MarinerLog("store$Creating", *avcp);
}
afs_PutFakeStat(&fakestate);
code = afs_CheckCode(code, treq, 20);
afs_DestroyReq(treq);
done2:
osi_FreeSmallSpace(OutFidStatus);
osi_FreeSmallSpace(OutDirStatus);
return code;
}
/*
* Check to see if we can track the change locally: requires that
* we have sufficiently recent info in data cache. If so, we
* know the new DataVersion number, and place it correctly in both the
* data and stat cache entries. This routine returns 1 if we should
* do the operation locally, and 0 otherwise.
*
* This routine must be called with the stat cache entry write-locked,
* and dcache entry write-locked.
*/
int
afs_LocalHero(struct vcache *avc, struct dcache *adc,
AFSFetchStatus * astat, int aincr)
{
afs_int32 ok;
afs_hyper_t avers;
AFS_STATCNT(afs_LocalHero);
hset64(avers, astat->dataVersionHigh, astat->DataVersion);
/* avers *is* the version number now, no matter what */
if (adc) {
/* does what's in the dcache *now* match what's in the vcache *now*,
* and do we have a valid callback? if not, our local copy is not "ok" */
ok = (hsame(avc->f.m.DataVersion, adc->f.versionNo) && avc->callback
&& (avc->f.states & CStatd) && avc->cbExpires >= osi_Time());
} else {
ok = 0;
}
if (ok) {
/* check that the DV on the server is what we expect it to be */
afs_hyper_t newDV;
hset(newDV, adc->f.versionNo);
hadd32(newDV, aincr);
if (!hsame(avers, newDV)) {
ok = 0;
}
}
#if defined(AFS_SGI_ENV)
osi_Assert(avc->v.v_type == VDIR);
#endif
/* The bulk status code used the length as a sequence number. */
/* Don't update the vcache entry unless the stats are current. */
if (avc->f.states & CStatd) {
hset(avc->f.m.DataVersion, avers);
#ifdef AFS_64BIT_CLIENT
FillInt64(avc->f.m.Length, astat->Length_hi, astat->Length);
#else /* AFS_64BIT_CLIENT */
avc->f.m.Length = astat->Length;
#endif /* AFS_64BIT_CLIENT */
avc->f.m.Date = astat->ClientModTime;
}
if (ok) {
/* we've been tracking things correctly */
adc->dflags |= DFEntryMod;
adc->f.versionNo = avers;
return 1;
} else {
if (adc) {
ZapDCE(adc);
DZap(adc);
}
if (avc->f.states & CStatd) {
osi_dnlc_purgedp(avc);
}
return 0;
}
}
static int
VLDB_Same(struct VenusFid *afid, struct vrequest *areq)
{
struct vrequest treq;
struct afs_conn *tconn;
int i, type = 0;
union {
struct vldbentry tve;
struct nvldbentry ntve;
struct uvldbentry utve;
} *v;
struct volume *tvp;
struct cell *tcell;
char *bp, tbuf[CVBS]; /* biggest volume id is 2^32, ~ 4*10^9 */
unsigned int changed;
struct server *(oldhosts[NMAXNSERVERS]);
AFS_STATCNT(CheckVLDB);
afs_FinalizeReq(areq);
if ((i = afs_InitReq(&treq, afs_osi_credp)))
return DUNNO;
v = afs_osi_Alloc(sizeof(*v));
tcell = afs_GetCell(afid->Cell, READ_LOCK);
bp = afs_cv2string(&tbuf[CVBS], afid->Fid.Volume);
do {
VSleep(2); /* Better safe than sorry. */
tconn =
afs_ConnByMHosts(tcell->cellHosts, tcell->vlport, tcell->cellNum,
&treq, SHARED_LOCK);
if (tconn) {
if (tconn->srvr->server->flags & SNO_LHOSTS) {
type = 0;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameO(tconn->id, bp, &v->tve);
RX_AFS_GLOCK();
} else if (tconn->srvr->server->flags & SYES_LHOSTS) {
type = 1;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameN(tconn->id, bp, &v->ntve);
RX_AFS_GLOCK();
} else {
type = 2;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameU(tconn->id, bp, &v->utve);
RX_AFS_GLOCK();
if (!(tconn->srvr->server->flags & SVLSRV_UUID)) {
if (i == RXGEN_OPCODE) {
type = 1;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameN(tconn->id, bp, &v->ntve);
RX_AFS_GLOCK();
if (i == RXGEN_OPCODE) {
type = 0;
tconn->srvr->server->flags |= SNO_LHOSTS;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameO(tconn->id, bp, &v->tve);
RX_AFS_GLOCK();
} else if (!i)
tconn->srvr->server->flags |= SYES_LHOSTS;
} else if (!i)
tconn->srvr->server->flags |= SVLSRV_UUID;
}
lastcode = i;
}
} else
i = -1;
} while (afs_Analyze(tconn, i, NULL, &treq, -1, /* no op code for this */
SHARED_LOCK, tcell));
afs_PutCell(tcell, READ_LOCK);
afs_Trace2(afs_iclSetp, CM_TRACE_CHECKVLDB, ICL_TYPE_FID, &afid,
ICL_TYPE_INT32, i);
if (i) {
afs_osi_Free(v, sizeof(*v));
return DUNNO;
}
/* have info, copy into serverHost array */
changed = 0;
tvp = afs_FindVolume(afid, WRITE_LOCK);
if (tvp) {
ObtainWriteLock(&tvp->lock, 107);
for (i = 0; i < NMAXNSERVERS && tvp->serverHost[i]; i++) {
oldhosts[i] = tvp->serverHost[i];
}
if (type == 2) {
InstallUVolumeEntry(tvp, &v->utve, afid->Cell, tcell, &treq);
} else if (type == 1) {
InstallNVolumeEntry(tvp, &v->ntve, afid->Cell);
} else {
InstallVolumeEntry(tvp, &v->tve, afid->Cell);
}
if (i < NMAXNSERVERS && tvp->serverHost[i]) {
changed = 1;
}
for (--i; !changed && i >= 0; i--) {
if (tvp->serverHost[i] != oldhosts[i]) {
changed = 1; /* also happens if prefs change. big deal. */
}
}
ReleaseWriteLock(&tvp->lock);
afs_PutVolume(tvp, WRITE_LOCK);
} else { /* can't find volume */
tvp = afs_GetVolume(afid, &treq, WRITE_LOCK);
if (tvp) {
afs_PutVolume(tvp, WRITE_LOCK);
afs_osi_Free(v, sizeof(*v));
return DIFFERENT;
} else {
afs_osi_Free(v, sizeof(*v));
return DUNNO;
}
}
afs_osi_Free(v, sizeof(*v));
return (changed ? DIFFERENT : SAME);
} /*VLDB_Same */
/*------------------------------------------------------------------------
* EXPORTED afs_Analyze
*
* Description:
* Analyze the outcome of an RPC operation, taking whatever support
* actions are necessary.
*
* Arguments:
* aconn : Ptr to the relevant connection on which the call was made.
* acode : The return code experienced by the RPC.
* afid : The FID of the file involved in the action. This argument
* may be null if none was involved.
* areq : The request record associated with this operation.
* op : which RPC we are analyzing.
* cellp : pointer to a cell struct. Must provide either fid or cell.
*
* Returns:
* Non-zero value if the related RPC operation should be retried,
* zero otherwise.
*
* Environment:
* This routine is typically called in a do-while loop, causing the
* embedded RPC operation to be called repeatedly if appropriate
* until whatever error condition (if any) is intolerable.
*
* Side Effects:
* As advertised.
*
* NOTE:
* The retry return value is used by afs_StoreAllSegments to determine
* if this is a temporary or permanent error.
*------------------------------------------------------------------------*/
int
afs_Analyze(register struct afs_conn *aconn, afs_int32 acode,
struct VenusFid *afid, register struct vrequest *areq, int op,
afs_int32 locktype, struct cell *cellp)
{
afs_int32 i;
struct srvAddr *sa;
struct server *tsp;
struct volume *tvp;
afs_int32 shouldRetry = 0;
afs_int32 serversleft = 1;
struct afs_stats_RPCErrors *aerrP;
afs_int32 markeddown;
if (AFS_IS_DISCONNECTED && !AFS_IN_SYNC) {
/* On reconnection, act as connected. XXX: for now.... */
/* SXW - This may get very tired after a while. We should try and
* intercept all RPCs before they get here ... */
/*printf("afs_Analyze: disconnected\n");*/
afs_FinalizeReq(areq);
if (aconn) {
/* SXW - I suspect that this will _never_ happen - we shouldn't
* get a connection because we're disconnected !!!*/
afs_PutConn(aconn, locktype);
}
return 0;
}
AFS_STATCNT(afs_Analyze);
afs_Trace4(afs_iclSetp, CM_TRACE_ANALYZE, ICL_TYPE_INT32, op,
ICL_TYPE_POINTER, aconn, ICL_TYPE_INT32, acode, ICL_TYPE_LONG,
areq->uid);
aerrP = (struct afs_stats_RPCErrors *)0;
if ((op >= 0) && (op < AFS_STATS_NUM_FS_RPC_OPS))
aerrP = &(afs_stats_cmfullperf.rpc.fsRPCErrors[op]);
afs_FinalizeReq(areq);
if (!aconn && areq->busyCount) { /* one RPC or more got VBUSY/VRESTARTING */
tvp = afs_FindVolume(afid, READ_LOCK);
if (tvp) {
afs_warnuser("afs: Waiting for busy volume %u (%s) in cell %s\n",
(afid ? afid->Fid.Volume : 0),
(tvp->name ? tvp->name : ""),
((tvp->serverHost[0]
&& tvp->serverHost[0]->cell) ? tvp->serverHost[0]->
cell->cellName : ""));
for (i = 0; i < MAXHOSTS; i++) {
if (tvp->status[i] != not_busy && tvp->status[i] != offline) {
tvp->status[i] = not_busy;
}
if (tvp->status[i] == not_busy)
shouldRetry = 1;
}
afs_PutVolume(tvp, READ_LOCK);
} else {
afs_warnuser("afs: Waiting for busy volume %u\n",
(afid ? afid->Fid.Volume : 0));
}
if (areq->busyCount > 100) {
if (aerrP)
(aerrP->err_Volume)++;
areq->volumeError = VOLBUSY;
shouldRetry = 0;
} else {
VSleep(afs_BusyWaitPeriod); /* poll periodically */
}
if (shouldRetry != 0)
areq->busyCount++;
return shouldRetry; /* should retry */
}
if (!aconn || !aconn->srvr) {
if (!areq->volumeError) {
if (aerrP)
(aerrP->err_Network)++;
if (hm_retry_int && !(areq->flags & O_NONBLOCK) && /* "hard" mount */
((afid && afs_IsPrimaryCellNum(afid->Cell))
|| (cellp && afs_IsPrimaryCell(cellp)))) {
if (!afid) {
afs_warnuser
("afs: hard-mount waiting for a vlserver to return to service\n");
VSleep(hm_retry_int);
afs_CheckServers(1, cellp);
shouldRetry = 1;
} else {
tvp = afs_FindVolume(afid, READ_LOCK);
if (!tvp || (tvp->states & VRO)) {
shouldRetry = hm_retry_RO;
} else {
shouldRetry = hm_retry_RW;
}
if (tvp)
afs_PutVolume(tvp, READ_LOCK);
if (shouldRetry) {
afs_warnuser
("afs: hard-mount waiting for volume %u\n",
afid->Fid.Volume);
VSleep(hm_retry_int);
afs_CheckServers(1, cellp);
}
}
} /* if (hm_retry_int ... */
else {
areq->networkError = 1;
}
}
return shouldRetry;
}
/* Find server associated with this connection. */
sa = aconn->srvr;
tsp = sa->server;
/* Before we do anything with acode, make sure we translate it back to
* a system error */
if ((acode & ~0xff) == ERROR_TABLE_BASE_uae)
acode = et_to_sys_error(acode);
if (acode == 0) {
/* If we previously took an error, mark this volume not busy */
if (areq->volumeError) {
tvp = afs_FindVolume(afid, READ_LOCK);
if (tvp) {
for (i = 0; i < MAXHOSTS; i++) {
if (tvp->serverHost[i] == tsp) {
tvp->status[i] = not_busy;
}
}
afs_PutVolume(tvp, READ_LOCK);
}
}
afs_PutConn(aconn, locktype);
return 0;
}
/* If network troubles, mark server as having bogued out again. */
/* VRESTARTING is < 0 because of backward compatibility issues
* with 3.4 file servers and older cache managers */
#ifdef AFS_64BIT_CLIENT
if (acode == -455)
acode = 455;
#endif /* AFS_64BIT_CLIENT */
if ((acode < 0) && (acode != VRESTARTING)) {
if (acode == RX_CALL_TIMEOUT) {
serversleft = afs_BlackListOnce(areq, afid, tsp);
areq->idleError++;
if (serversleft) {
shouldRetry = 1;
} else {
shouldRetry = 0;
}
/* By doing this, we avoid ever marking a server down
* in an idle timeout case. That's because the server is
* still responding and may only be letting a single vnode
* time out. We otherwise risk having the server continually
* be marked down, then up, then down again...
*/
goto out;
}
markeddown = afs_ServerDown(sa);
ForceNewConnections(sa); /**multi homed clients lock:afs_xsrvAddr? */
if (aerrP)
(aerrP->err_Server)++;
#if 0
/* retry *once* when the server is timed out in case of NAT */
if (markeddown && acode == RX_CALL_DEAD) {
aconn->forceConnectFS = 1;
shouldRetry = 1;
}
#endif
}
if (acode == VBUSY || acode == VRESTARTING) {
if (acode == VBUSY) {
areq->busyCount++;
if (aerrP)
(aerrP->err_VolumeBusies)++;
} else
areq->busyCount = 1;
tvp = afs_FindVolume(afid, READ_LOCK);
if (tvp) {
for (i = 0; i < MAXHOSTS; i++) {
if (tvp->serverHost[i] == tsp) {
tvp->status[i] = rdwr_busy; /* can't tell which yet */
/* to tell which, have to look at the op code. */
}
}
afs_PutVolume(tvp, READ_LOCK);
} else {
afs_warnuser("afs: Waiting for busy volume %u in cell %s\n",
(afid ? afid->Fid.Volume : 0), tsp->cell->cellName);
VSleep(afs_BusyWaitPeriod); /* poll periodically */
}
shouldRetry = 1;
acode = 0;
} else if (acode == VICETOKENDEAD
|| (acode & ~0xff) == ERROR_TABLE_BASE_RXK) {
/* any rxkad error is treated as token expiration */
struct unixuser *tu;
/*
* I'm calling these errors protection errors, since they involve
* faulty authentication.
*/
if (aerrP)
(aerrP->err_Protection)++;
tu = afs_FindUser(areq->uid, tsp->cell->cellNum, READ_LOCK);
if (tu) {
if (acode == VICETOKENDEAD) {
aconn->forceConnectFS = 1;
} else if (acode == RXKADEXPIRED) {
aconn->forceConnectFS = 0; /* don't check until new tokens set */
aconn->user->states |= UTokensBad;
afs_warnuser
("afs: Tokens for user of AFS id %d for cell %s have expired\n",
tu->vid, aconn->srvr->server->cell->cellName);
} else {
serversleft = afs_BlackListOnce(areq, afid, tsp);
areq->tokenError++;
if (serversleft) {
afs_warnuser
("afs: Tokens for user of AFS id %d for cell %s: rxkad error=%d\n",
tu->vid, aconn->srvr->server->cell->cellName, acode);
shouldRetry = 1;
} else {
areq->tokenError = 0;
aconn->forceConnectFS = 0; /* don't check until new tokens set */
aconn->user->states |= UTokensBad;
afs_warnuser
("afs: Tokens for user of AFS id %d for cell %s are discarded (rxkad error=%d)\n",
tu->vid, aconn->srvr->server->cell->cellName, acode);
}
}
afs_PutUser(tu, READ_LOCK);
} else {
/* The else case shouldn't be possible and should probably be replaced by a panic? */
if (acode == VICETOKENDEAD) {
aconn->forceConnectFS = 1;
} else if (acode == RXKADEXPIRED) {
aconn->forceConnectFS = 0; /* don't check until new tokens set */
aconn->user->states |= UTokensBad;
afs_warnuser
("afs: Tokens for user %d for cell %s have expired\n",
areq->uid, aconn->srvr->server->cell->cellName);
} else {
aconn->forceConnectFS = 0; /* don't check until new tokens set */
aconn->user->states |= UTokensBad;
afs_warnuser
("afs: Tokens for user %d for cell %s are discarded (rxkad error = %d)\n",
areq->uid, aconn->srvr->server->cell->cellName, acode);
}
}
shouldRetry = 1; /* Try again (as root). */
}
/* Check for access violation. */
else if (acode == EACCES) {
/* should mark access error in non-existent per-user global structure */
if (aerrP)
(aerrP->err_Protection)++;
areq->accessError = 1;
if (op == AFS_STATS_FS_RPCIDX_STOREDATA)
areq->permWriteError = 1;
shouldRetry = 0;
}
/* check for ubik errors; treat them like crashed servers */
else if (acode >= ERROR_TABLE_BASE_U && acode < ERROR_TABLE_BASE_U + 255) {
afs_ServerDown(sa);
if (aerrP)
(aerrP->err_Server)++;
shouldRetry = 1; /* retryable (maybe one is working) */
VSleep(1); /* just in case */
}
/* Check for bad volume data base / missing volume. */
else if (acode == VSALVAGE || acode == VOFFLINE || acode == VNOVOL
|| acode == VNOSERVICE || acode == VMOVED) {
struct cell *tcell;
int same;
shouldRetry = 1;
areq->volumeError = VOLMISSING;
if (aerrP)
(aerrP->err_Volume)++;
if (afid && (tcell = afs_GetCell(afid->Cell, 0))) {
same = VLDB_Same(afid, areq);
tvp = afs_FindVolume(afid, READ_LOCK);
if (tvp) {
for (i = 0; i < MAXHOSTS && tvp->serverHost[i]; i++) {
if (tvp->serverHost[i] == tsp) {
if (tvp->status[i] == end_not_busy)
tvp->status[i] = offline;
else
tvp->status[i]++;
} else if (!same) {
tvp->status[i] = not_busy; /* reset the others */
}
}
afs_PutVolume(tvp, READ_LOCK);
}
}
} else if (acode >= ERROR_TABLE_BASE_VL && acode <= ERROR_TABLE_BASE_VL + 255) { /* vlserver errors */
shouldRetry = 0;
areq->volumeError = VOLMISSING;
} else if (acode >= 0) {
if (aerrP)
(aerrP->err_Other)++;
if (op == AFS_STATS_FS_RPCIDX_STOREDATA)
areq->permWriteError = 1;
shouldRetry = 0; /* Other random Vice error. */
} else if (acode == RX_MSGSIZE) { /* same meaning as EMSGSIZE... */
VSleep(1); /* Just a hack for desperate times. */
if (aerrP)
(aerrP->err_Other)++;
shouldRetry = 1; /* packet was too big, please retry call */
}
if (acode < 0 && acode != RX_MSGSIZE && acode != VRESTARTING) {
/* If we get here, code < 0 and we have network/Server troubles.
* areq->networkError is not set here, since we always
* retry in case there is another server. However, if we find
* no connection (aconn == 0) we set the networkError flag.
*/
afs_MarkServerUpOrDown(sa, SRVR_ISDOWN);
if (aerrP)
(aerrP->err_Server)++;
VSleep(1); /* Just a hack for desperate times. */
shouldRetry = 1;
}
out:
/* now unlock the connection and return */
afs_PutConn(aconn, locktype);
return (shouldRetry);
} /*afs_Analyze */
/* don't set CDirty in here because RPC is called synchronously */
int
afs_symlink(OSI_VC_DECL(adp), char *aname, struct vattr *attrs,
char *atargetName, struct vcache **tvcp, afs_ucred_t *acred)
{
afs_uint32 now = 0;
struct vrequest *treq = NULL;
afs_int32 code = 0;
struct afs_conn *tc;
struct VenusFid newFid;
struct dcache *tdc;
afs_size_t offset, len;
afs_int32 alen;
struct server *hostp = 0;
struct vcache *tvc;
struct AFSStoreStatus InStatus;
struct AFSFetchStatus *OutFidStatus, *OutDirStatus;
struct AFSCallBack CallBack;
struct AFSVolSync tsync;
struct volume *volp = 0;
struct afs_fakestat_state fakestate;
struct rx_connection *rxconn;
XSTATS_DECLS;
OSI_VC_CONVERT(adp);
AFS_STATCNT(afs_symlink);
afs_Trace2(afs_iclSetp, CM_TRACE_SYMLINK, ICL_TYPE_POINTER, adp,
ICL_TYPE_STRING, aname);
OutFidStatus = osi_AllocSmallSpace(sizeof(struct AFSFetchStatus));
OutDirStatus = osi_AllocSmallSpace(sizeof(struct AFSFetchStatus));
memset(&InStatus, 0, sizeof(InStatus));
if ((code = afs_CreateReq(&treq, acred)))
goto done2;
afs_InitFakeStat(&fakestate);
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&adp, &fakestate, treq);
if (code)
goto done;
if (strlen(aname) > AFSNAMEMAX || strlen(atargetName) > AFSPATHMAX) {
code = ENAMETOOLONG;
goto done;
}
if (afs_IsDynroot(adp)) {
code = afs_DynrootVOPSymlink(adp, acred, aname, atargetName);
goto done;
}
if (afs_IsDynrootMount(adp)) {
code = EROFS;
goto done;
}
code = afs_VerifyVCache(adp, treq);
if (code) {
code = afs_CheckCode(code, treq, 30);
goto done;
}
/** If the volume is read-only, return error without making an RPC to the
* fileserver
*/
if (adp->f.states & CRO) {
code = EROFS;
goto done;
}
if (AFS_IS_DISCONNECTED && !AFS_IS_DISCON_RW) {
code = ENETDOWN;
goto done;
}
InStatus.Mask = AFS_SETMODTIME | AFS_SETMODE;
InStatus.ClientModTime = osi_Time();
alen = strlen(atargetName); /* we want it to include the null */
if ( (*atargetName == '#' || *atargetName == '%') && alen > 1 && atargetName[alen-1] == '.') {
InStatus.UnixModeBits = 0644; /* mt pt: null from "." at end */
if (alen == 1)
alen++; /* Empty string */
} else {
InStatus.UnixModeBits = 0755;
alen++; /* add in the null */
}
tdc = afs_GetDCache(adp, (afs_size_t) 0, treq, &offset, &len, 1);
volp = afs_FindVolume(&adp->f.fid, READ_LOCK); /*parent is also in same vol */
ObtainWriteLock(&adp->lock, 156);
if (tdc)
ObtainWriteLock(&tdc->lock, 636);
/* No further locks: if the SymLink succeeds, it does not matter what happens
* to our local copy of the directory. If somebody tampers with it in the meantime,
* the copy will be invalidated */
if (!AFS_IS_DISCON_RW) {
do {
tc = afs_Conn(&adp->f.fid, treq, SHARED_LOCK, &rxconn);
if (tc) {
hostp = tc->parent->srvr->server;
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_SYMLINK);
if (adp->f.states & CForeign) {
now = osi_Time();
RX_AFS_GUNLOCK();
code =
RXAFS_DFSSymlink(rxconn,
(struct AFSFid *)&adp->f.fid.Fid,
aname, atargetName, &InStatus,
(struct AFSFid *)&newFid.Fid,
OutFidStatus, OutDirStatus,
&CallBack, &tsync);
RX_AFS_GLOCK();
} else {
RX_AFS_GUNLOCK();
code =
RXAFS_Symlink(rxconn, (struct AFSFid *)&adp->f.fid.Fid,
aname, atargetName, &InStatus,
(struct AFSFid *)&newFid.Fid,
OutFidStatus, OutDirStatus, &tsync);
RX_AFS_GLOCK();
}
XSTATS_END_TIME;
} else
code = -1;
} while (afs_Analyze
(tc, rxconn, code, &adp->f.fid, treq, AFS_STATS_FS_RPCIDX_SYMLINK,
SHARED_LOCK, NULL));
} else {
newFid.Cell = adp->f.fid.Cell;
newFid.Fid.Volume = adp->f.fid.Fid.Volume;
afs_GenFakeFid(&newFid, VREG, 0);
}
ObtainWriteLock(&afs_xvcache, 40);
if (code) {
if (code < 0) {
afs_StaleVCache(adp);
}
ReleaseWriteLock(&adp->lock);
ReleaseWriteLock(&afs_xvcache);
if (tdc) {
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc);
}
goto done;
}
/* otherwise, we should see if we can make the change to the dir locally */
if (AFS_IS_DISCON_RW || afs_LocalHero(adp, tdc, OutDirStatus, 1)) {
/* we can do it locally */
ObtainWriteLock(&afs_xdcache, 293);
/* If the following fails because the name has been created in the meantime, the
* directory is out-of-date - the file server knows best! */
code = afs_dir_Create(tdc, aname, &newFid.Fid);
ReleaseWriteLock(&afs_xdcache);
if (code && !AFS_IS_DISCON_RW) {
ZapDCE(tdc); /* surprise error -- use invalid value */
DZap(tdc);
}
}
if (tdc) {
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc);
}
newFid.Cell = adp->f.fid.Cell;
newFid.Fid.Volume = adp->f.fid.Fid.Volume;
ReleaseWriteLock(&adp->lock);
/* now we're done with parent dir, create the link's entry. Note that
* no one can get a pointer to the new cache entry until we release
* the xvcache lock. */
tvc = afs_NewVCache(&newFid, hostp);
if (!tvc)
{
code = -2;
ReleaseWriteLock(&afs_xvcache);
goto done;
}
ObtainWriteLock(&tvc->lock, 157);
ObtainWriteLock(&afs_xcbhash, 500);
tvc->f.states |= CStatd; /* have valid info */
tvc->f.states &= ~CBulkFetching;
if (adp->f.states & CForeign) {
tvc->f.states |= CForeign;
/* We don't have to worry about losing the callback since we're doing it
* under the afs_xvcache lock actually, afs_NewVCache may drop the
* afs_xvcache lock, if it calls afs_FlushVCache */
tvc->cbExpires = CallBack.ExpirationTime + now;
afs_QueueCallback(tvc, CBHash(CallBack.ExpirationTime), volp);
} else {
tvc->cbExpires = 0x7fffffff; /* never expires, they can't change */
/* since it never expires, we don't have to queue the callback */
}
ReleaseWriteLock(&afs_xcbhash);
if (AFS_IS_DISCON_RW) {
attrs->va_mode = InStatus.UnixModeBits;
afs_GenDisconStatus(adp, tvc, &newFid, attrs, treq, VLNK);
code = afs_DisconCreateSymlink(tvc, atargetName, treq);
if (code) {
/* XXX - When this goes wrong, we need to tidy up the changes we made to
* the parent, and get rid of the vcache we just created */
ReleaseWriteLock(&tvc->lock);
ReleaseWriteLock(&afs_xvcache);
afs_PutVCache(tvc);
goto done;
}
afs_DisconAddDirty(tvc, VDisconCreate, 0);
} else {
afs_ProcessFS(tvc, OutFidStatus, treq);
}
if (!tvc->linkData) {
tvc->linkData = afs_osi_Alloc(alen);
osi_Assert(tvc->linkData != NULL);
strncpy(tvc->linkData, atargetName, alen - 1);
tvc->linkData[alen - 1] = 0;
}
ReleaseWriteLock(&tvc->lock);
ReleaseWriteLock(&afs_xvcache);
if (tvcp)
*tvcp = tvc;
else
afs_PutVCache(tvc);
code = 0;
done:
afs_PutFakeStat(&fakestate);
if (volp)
afs_PutVolume(volp, READ_LOCK);
AFS_DISCON_UNLOCK();
code = afs_CheckCode(code, treq, 31);
afs_DestroyReq(treq);
done2:
osi_FreeSmallSpace(OutFidStatus);
osi_FreeSmallSpace(OutDirStatus);
return code;
}
void
afs_CheckCallbacks(unsigned int secs)
{
struct vcache *tvc;
register struct afs_q *tq;
struct afs_q *uq;
afs_uint32 now;
struct volume *tvp;
register int safety;
ObtainWriteLock(&afs_xcbhash, 85); /* pretty likely I'm going to remove something */
now = osi_Time();
for (safety = 0, tq = cbHashT[base].head.prev;
(safety <= CBQ_LIMIT) && (tq != &(cbHashT[base].head));
tq = uq, safety++) {
uq = QPrev(tq);
tvc = CBQTOV(tq);
if (tvc->cbExpires < now + secs) { /* race #1 here */
/* Get the volume, and if its callback expiration time is more than secs
* seconds into the future, update this vcache entry and requeue it below
*/
if ((tvc->f.states & CRO)
&& (tvp = afs_FindVolume(&(tvc->f.fid), READ_LOCK))) {
if (tvp->expireTime > now + secs) {
tvc->cbExpires = tvp->expireTime; /* XXX race here */
} else {
int i;
for (i = 0; i < MAXHOSTS && tvp->serverHost[i]; i++) {
if (!(tvp->serverHost[i]->flags & SRVR_ISDOWN)) {
/* What about locking xvcache or vrefcount++ or
* write locking tvc? */
QRemove(tq);
tvc->f.states &= ~(CStatd | CMValid | CUnique);
if (!(tvc->f.states & (CVInit|CVFlushed)) &&
(tvc->f.fid.Fid.Vnode & 1 ||
(vType(tvc) == VDIR)))
osi_dnlc_purgedp(tvc);
tvc->dchint = NULL; /*invalidate em */
afs_ResetVolumeInfo(tvp);
break;
}
}
}
afs_PutVolume(tvp, READ_LOCK);
} else {
/* Do I need to worry about things like execsorwriters?
* What about locking xvcache or vrefcount++ or write locking tvc?
*/
QRemove(tq);
tvc->f.states &= ~(CStatd | CMValid | CUnique);
if (!(tvc->f.states & (CVInit|CVFlushed)) &&
(tvc->f.fid.Fid.Vnode & 1 || (vType(tvc) == VDIR)))
osi_dnlc_purgedp(tvc);
}
}
if ((tvc->cbExpires > basetime) && CBHash(tvc->cbExpires - basetime)) {
/* it's been renewed on us. Have to be careful not to put it back
* into this slot, or we may never get out of here.
*/
int slot;
slot = (CBHash(tvc->cbExpires - basetime) + base) % CBHTSIZE;
if (slot != base) {
if (QPrev(tq))
QRemove(&(tvc->callsort));
QAdd(&(cbHashT[slot].head), &(tvc->callsort));
/* XXX remember to update volume expiration time */
/* -- not needed for correctness, though */
}
}
}
if (safety > CBQ_LIMIT) {
afs_stats_cmperf.cbloops++;
if (afs_paniconwarn)
osi_Panic("CheckCallbacks");
afs_warn
("AFS Internal Error (minor): please contact AFS Product Support.\n");
ReleaseWriteLock(&afs_xcbhash);
afs_FlushCBs();
return;
} else
ReleaseWriteLock(&afs_xcbhash);
/* XXX future optimization:
if this item has been recently accessed, queue up a stat for it.
{
struct dcache * adc;
ObtainReadLock(&afs_xdcache);
if ((adc = tvc->quick.dc) && (adc->stamp == tvc->quick.stamp)
&& (afs_indexTimes[adc->index] > afs_indexCounter - 20)) {
queue up the stat request
}
ReleaseReadLock(&afs_xdcache);
}
*/
return;
} /* afs_CheckCallback */
static int
ClearCallBack(struct rx_connection *a_conn,
struct AFSFid *a_fid)
{
struct vcache *tvc;
int i;
struct VenusFid localFid;
struct volume *tv;
#ifdef AFS_DARWIN80_ENV
vnode_t vp;
#endif
AFS_STATCNT(ClearCallBack);
AFS_ASSERT_GLOCK();
/*
* XXXX Don't hold any server locks here because of callback protocol XXX
*/
localFid.Cell = 0;
localFid.Fid.Volume = a_fid->Volume;
localFid.Fid.Vnode = a_fid->Vnode;
localFid.Fid.Unique = a_fid->Unique;
/*
* Volume ID of zero means don't do anything.
*/
if (a_fid->Volume != 0) {
if (a_fid->Vnode == 0) {
struct afs_q *tq, *uq;
/*
* Clear callback for the whole volume. Zip through the
* hash chain, nullifying entries whose volume ID matches.
*/
loop1:
ObtainReadLock(&afs_xvcache);
i = VCHashV(&localFid);
for (tq = afs_vhashTV[i].prev; tq != &afs_vhashTV[i]; tq = uq) {
uq = QPrev(tq);
tvc = QTOVH(tq);
if (tvc->f.fid.Fid.Volume == a_fid->Volume) {
tvc->callback = NULL;
if (!localFid.Cell)
localFid.Cell = tvc->f.fid.Cell;
tvc->dchint = NULL; /* invalidate hints */
if (tvc->f.states & CVInit) {
ReleaseReadLock(&afs_xvcache);
afs_osi_Sleep(&tvc->f.states);
goto loop1;
}
#if defined(AFS_SGI_ENV) || defined(AFS_SUN5_ENV) || defined(AFS_HPUX_ENV) || defined(AFS_LINUX20_ENV)
AFS_FAST_HOLD(tvc);
#else
#ifdef AFS_DARWIN80_ENV
if (tvc->f.states & CDeadVnode) {
if (!(tvc->f.states & CBulkFetching)) {
ReleaseReadLock(&afs_xvcache);
afs_osi_Sleep(&tvc->f.states);
goto loop1;
}
}
vp = AFSTOV(tvc);
if (vnode_get(vp))
continue;
if (vnode_ref(vp)) {
AFS_GUNLOCK();
vnode_put(vp);
AFS_GLOCK();
continue;
}
if (tvc->f.states & (CBulkFetching|CDeadVnode)) {
AFS_GUNLOCK();
vnode_recycle(AFSTOV(tvc));
AFS_GLOCK();
}
#else
AFS_FAST_HOLD(tvc);
#endif
#endif
ReleaseReadLock(&afs_xvcache);
ObtainWriteLock(&afs_xcbhash, 449);
afs_DequeueCallback(tvc);
tvc->f.states &= ~(CStatd | CUnique | CBulkFetching);
afs_allCBs++;
if (tvc->f.fid.Fid.Vnode & 1)
afs_oddCBs++;
else
afs_evenCBs++;
ReleaseWriteLock(&afs_xcbhash);
if ((tvc->f.fid.Fid.Vnode & 1 || (vType(tvc) == VDIR)))
osi_dnlc_purgedp(tvc);
afs_Trace3(afs_iclSetp, CM_TRACE_CALLBACK,
ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32,
tvc->f.states, ICL_TYPE_INT32,
a_fid->Volume);
#ifdef AFS_DARWIN80_ENV
vnode_put(AFSTOV(tvc));
#endif
ObtainReadLock(&afs_xvcache);
uq = QPrev(tq);
AFS_FAST_RELE(tvc);
} else if ((tvc->f.states & CMValid)
&& (tvc->mvid->Fid.Volume == a_fid->Volume)) {
tvc->f.states &= ~CMValid;
if (!localFid.Cell)
localFid.Cell = tvc->mvid->Cell;
}
}
ReleaseReadLock(&afs_xvcache);
/*
* XXXX Don't hold any locks here XXXX
*/
tv = afs_FindVolume(&localFid, 0);
if (tv) {
afs_ResetVolumeInfo(tv);
afs_PutVolume(tv, 0);
/* invalidate mtpoint? */
}
} /*Clear callbacks for whole volume */
else {
/*
* Clear callbacks just for the one file.
*/
struct vcache *uvc;
afs_allCBs++;
if (a_fid->Vnode & 1)
afs_oddCBs++; /*Could do this on volume basis, too */
else
afs_evenCBs++; /*A particular fid was specified */
loop2:
ObtainReadLock(&afs_xvcache);
i = VCHash(&localFid);
for (tvc = afs_vhashT[i]; tvc; tvc = uvc) {
uvc = tvc->hnext;
if (tvc->f.fid.Fid.Vnode == a_fid->Vnode
&& tvc->f.fid.Fid.Volume == a_fid->Volume
&& tvc->f.fid.Fid.Unique == a_fid->Unique) {
tvc->callback = NULL;
tvc->dchint = NULL; /* invalidate hints */
if (tvc->f.states & CVInit) {
ReleaseReadLock(&afs_xvcache);
afs_osi_Sleep(&tvc->f.states);
goto loop2;
}
#if defined(AFS_SGI_ENV) || defined(AFS_SUN5_ENV) || defined(AFS_HPUX_ENV) || defined(AFS_LINUX20_ENV)
AFS_FAST_HOLD(tvc);
#else
#ifdef AFS_DARWIN80_ENV
if (tvc->f.states & CDeadVnode) {
if (!(tvc->f.states & CBulkFetching)) {
ReleaseReadLock(&afs_xvcache);
afs_osi_Sleep(&tvc->f.states);
goto loop2;
}
}
vp = AFSTOV(tvc);
if (vnode_get(vp))
continue;
if (vnode_ref(vp)) {
AFS_GUNLOCK();
vnode_put(vp);
AFS_GLOCK();
continue;
}
if (tvc->f.states & (CBulkFetching|CDeadVnode)) {
AFS_GUNLOCK();
vnode_recycle(AFSTOV(tvc));
AFS_GLOCK();
}
#else
AFS_FAST_HOLD(tvc);
#endif
#endif
ReleaseReadLock(&afs_xvcache);
ObtainWriteLock(&afs_xcbhash, 450);
afs_DequeueCallback(tvc);
tvc->f.states &= ~(CStatd | CUnique | CBulkFetching);
ReleaseWriteLock(&afs_xcbhash);
if ((tvc->f.fid.Fid.Vnode & 1 || (vType(tvc) == VDIR)))
osi_dnlc_purgedp(tvc);
afs_Trace3(afs_iclSetp, CM_TRACE_CALLBACK,
ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32,
tvc->f.states, ICL_TYPE_LONG, 0);
#ifdef CBDEBUG
lastCallBack_vnode = afid->Vnode;
lastCallBack_dv = tvc->mstat.DataVersion.low;
osi_GetuTime(&lastCallBack_time);
#endif /* CBDEBUG */
#ifdef AFS_DARWIN80_ENV
vnode_put(AFSTOV(tvc));
#endif
ObtainReadLock(&afs_xvcache);
uvc = tvc->hnext;
AFS_FAST_RELE(tvc);
}
} /*Walk through hash table */
ReleaseReadLock(&afs_xvcache);
} /*Clear callbacks for one file */
}
/*Fid has non-zero volume ID */
/*
* Always return a predictable value.
*/
return (0);
} /*ClearCallBack */