/*!
 * \brief Query the AFSDB handler and wait for response.
 * \param  acellName
 * \return 0 for success. < 0 is error.
 */
static int
afs_GetCellHostsAFSDB(char *acellName)
{
    AFS_ASSERT_GLOCK();
    if (!afsdb_handler_running)
	return ENOENT;

    ObtainWriteLock(&afsdb_client_lock, 685);
    ObtainWriteLock(&afsdb_req_lock, 686);

    afsdb_req.cellname = acellName;

    afsdb_req.complete = 0;
    afsdb_req.pending = 1;
    afs_osi_Wakeup(&afsdb_req);
    ConvertWToRLock(&afsdb_req_lock);

    while (afsdb_handler_running && !afsdb_req.complete) {
	ReleaseReadLock(&afsdb_req_lock);
	afs_osi_Sleep(&afsdb_req);
	ObtainReadLock(&afsdb_req_lock);
    };

    ReleaseReadLock(&afsdb_req_lock);
    ReleaseWriteLock(&afsdb_client_lock);

    if (afsdb_req.cellname) {
	return 0;
    } else
	return ENOENT;
}

afs_int32
canWrite(int fid)
{
#ifndef AFS_PTHREAD_ENV
    afs_int32 code = 0;
#endif
    extern dumpSyncP dumpSyncPtr;

    ObtainWriteLock(&dumpSyncPtr->ds_lock);

    /* let the pipe drain */
    while (dumpSyncPtr->ds_bytes > 0) {
	if (dumpSyncPtr->ds_readerStatus == DS_WAITING) {
	    dumpSyncPtr->ds_readerStatus = 0;
#ifdef AFS_PTHREAD_ENV
	    CV_BROADCAST(&dumpSyncPtr->ds_readerStatus_cond);
#else
	    code = LWP_SignalProcess(&dumpSyncPtr->ds_readerStatus);
	    if (code)
		LogError(code, "canWrite: Signal delivery failed\n");
#endif
	}
	dumpSyncPtr->ds_writerStatus = DS_WAITING;
	ReleaseWriteLock(&dumpSyncPtr->ds_lock);
#ifdef AFS_PTHREAD_ENV
	MUTEX_ENTER(&dumpSyncPtr->ds_writerStatus_mutex);
	CV_WAIT(&dumpSyncPtr->ds_writerStatus_cond, &dumpSyncPtr->ds_writerStatus_mutex);
	MUTEX_EXIT(&dumpSyncPtr->ds_writerStatus_mutex);
#else
	LWP_WaitProcess(&dumpSyncPtr->ds_writerStatus);
#endif
	ObtainWriteLock(&dumpSyncPtr->ds_lock);
    }
    return (1);
}

int
DNew(struct dcache *adc, int page, struct DirBuffer *entry)
{
    /* Same as read, only do *not* even try to read the page, since it
     * probably doesn't exist. */
    struct buffer *tb;
    AFS_STATCNT(DNew);

    ObtainWriteLock(&afs_bufferLock, 264);
    if ((tb = afs_newslot(adc, page, NULL)) == 0) {
	ReleaseWriteLock(&afs_bufferLock);
	return EIO;
    }
    /* extend the chunk, if needed */
    /* Do it now, not in DFlush or afs_newslot when the data is written out,
     * since now our caller has adc->lock writelocked, and we can't acquire
     * that lock (or even map from a fid to a dcache) in afs_newslot or
     * DFlush due to lock hierarchy issues */
    if ((page + 1) * AFS_BUFFER_PAGESIZE > adc->f.chunkBytes) {
	afs_AdjustSize(adc, (page + 1) * AFS_BUFFER_PAGESIZE);
	osi_Assert(afs_WriteDCache(adc, 1) == 0);
    }
    ObtainWriteLock(&tb->lock, 265);
    tb->lockers++;
    ReleaseWriteLock(&afs_bufferLock);
    ReleaseWriteLock(&tb->lock);
    entry->buffer = tb;
    entry->data = tb->data;

    return 0;
}

/*
 * This is almost exactly like the PFlush() routine in afs_pioctl.c,
 * but that routine is static.  We are about to change a file from
 * bypassing caching to normal caching.  Therefore, we want to
 * throw out any existing VM pages for the file.  We keep track of
 * the number of times we go back and forth from caching to bypass.
 */
void
afs_TransitionToCaching(struct vcache *avc,
		        afs_ucred_t *acred,
			int aflags)
{
    int resetDesire = 0;
    int setManual = 0;

    if (!avc)
	return;

    if (aflags & TRANSChangeDesiredBit)
	resetDesire = 1;
    if (aflags & TRANSSetManualBit)
	setManual = 1;

#ifdef AFS_BOZONLOCK_ENV
    afs_BozonLock(&avc->pvnLock, avc);	/* Since afs_TryToSmush will do a pvn_vptrunc */
#else
    AFS_GLOCK();
#endif
    ObtainWriteLock(&avc->lock, 926);
    /*
     * Someone may have beat us to doing the transition - we had no lock
     * when we checked the flag earlier.  No cause to panic, just return.
     */
    if (!(avc->cachingStates & FCSBypass))
	goto done;

    /* Ok, we actually do need to flush */
    ObtainWriteLock(&afs_xcbhash, 957);
    afs_DequeueCallback(avc);
    avc->f.states &= ~(CStatd | CDirty);	/* next reference will re-stat cache entry */
    ReleaseWriteLock(&afs_xcbhash);
    /* now find the disk cache entries */
    afs_TryToSmush(avc, acred, 1);
    osi_dnlc_purgedp(avc);
    if (avc->linkData && !(avc->f.states & CCore)) {
	afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1);
	avc->linkData = NULL;
    }

    avc->cachingStates &= ~(FCSBypass);    /* Reset the bypass flag */
    if (resetDesire)
	avc->cachingStates &= ~(FCSDesireBypass);
    if (setManual)
	avc->cachingStates |= FCSManuallySet;
    avc->cachingTransitions++;

done:
    ReleaseWriteLock(&avc->lock);
#ifdef AFS_BOZONLOCK_ENV
    afs_BozonUnlock(&avc->pvnLock, avc);
#else
    AFS_GUNLOCK();
#endif
}

struct brequest *
afs_BQueue(short aopcode, struct vcache *avc,
	   afs_int32 dontwait, afs_int32 ause, afs_ucred_t *acred,
	   afs_size_t asparm0, afs_size_t asparm1, void *apparm0,
	   void *apparm1, void *apparm2)
{
    int i;
    struct brequest *tb;

    AFS_STATCNT(afs_BQueue);
    ObtainWriteLock(&afs_xbrs, 296);
    while (1) {
	tb = afs_brs;
	for (i = 0; i < NBRS; i++, tb++) {
	    if (tb->refCount == 0)
		break;
	}
	if (i < NBRS) {
	    /* found a buffer */
	    tb->opcode = aopcode;
	    tb->vc = avc;
	    tb->cred = acred;
	    if (tb->cred) {
		crhold(tb->cred);
	    }
	    if (avc) {
		AFS_FAST_HOLD(avc);
	    }
	    tb->refCount = ause + 1;
	    tb->size_parm[0] = asparm0;
	    tb->size_parm[1] = asparm1;
	    tb->ptr_parm[0] = apparm0;
	    tb->ptr_parm[1] = apparm1;
	    tb->ptr_parm[2] = apparm2;
	    tb->flags = 0;
	    tb->code_raw = tb->code_checkcode = 0;
	    tb->ts = afs_brs_count++;
	    /* if daemons are waiting for work, wake them up */
	    if (afs_brsDaemons > 0) {
		afs_osi_Wakeup(&afs_brsDaemons);
	    }
	    ReleaseWriteLock(&afs_xbrs);
	    return tb;
	}
	if (dontwait) {
	    ReleaseWriteLock(&afs_xbrs);
	    return NULL;
	}
	/* no free buffers, sleep a while */
	afs_brsWaiters++;
	ReleaseWriteLock(&afs_xbrs);
	afs_osi_Sleep(&afs_brsWaiters);
	ObtainWriteLock(&afs_xbrs, 301);
	afs_brsWaiters--;
    }
}

/*!
 * Remove a server from a cell's server list.
 * \param srvp Server to be removed.
 * \return
 */
void
afs_RemoveCellEntry(struct server *srvp)
{
    struct cell *tc;
    afs_int32 j, k;

    tc = srvp->cell;
    if (!tc)
	return;

    /* Remove the server structure from the cell list - if there */
    ObtainWriteLock(&tc->lock, 200);
    for (j = k = 0; j < AFS_MAXCELLHOSTS; j++) {
	if (!tc->cellHosts[j])
	    break;
	if (tc->cellHosts[j] != srvp) {
	    tc->cellHosts[k++] = tc->cellHosts[j];
	}
    }
    if (k == 0) {
	/* What do we do if we remove the last one? */
    }
    for (; k < AFS_MAXCELLHOSTS; k++) {
	tc->cellHosts[k] = 0;
    }
    ReleaseWriteLock(&tc->lock);
}

int afspag_PUnlog(char *ain, afs_int32 ainSize, struct AFS_UCRED **acred)
{
    register afs_int32 i;
    register struct unixuser *tu;
    afs_int32 pag, uid;

    AFS_STATCNT(PUnlog);
    if (!afs_resourceinit_flag)	/* afs daemons haven't started yet */
	return EIO;		/* Inappropriate ioctl for device */

    pag = PagInCred(*acred);
    uid = (pag == NOPAG) ? (*acred)->cr_uid : pag;
    i = UHash(uid);
    ObtainWriteLock(&afs_xuser, 823);
    for (tu = afs_users[i]; tu; tu = tu->next) {
	if (tu->uid == uid) {
	    tu->vid = UNDEFVID;
	    tu->states &= ~UHasTokens;
	    /* security is not having to say you're sorry */
	    memset((char *)&tu->ct, 0, sizeof(struct ClearToken));
#ifdef UKERNEL
	    /* set the expire times to 0, causes
	     * afs_GCUserData to remove this entry
	     */
	    tu->ct.EndTimestamp = 0;
	    tu->tokenTime = 0;
#endif /* UKERNEL */
	}
    }
    ReleaseWriteLock(&afs_xuser);
    return 0;
}

afs_int32
VnodeToSize(vnode_t * vp)
{
    int code;
    struct vattr vattr;

    /*
     * We lock xosi in osi_Stat, so we probably should
     * lock it here too - RWH.
     */
    ObtainWriteLock(&afs_xosi, 578);
    vattr.va_mask = AT_SIZE;
    AFS_GUNLOCK();
#ifdef AFS_SUN511_ENV
    code = VOP_GETATTR(vp, &vattr, 0, afs_osi_credp, NULL);
#else
    code = VOP_GETATTR(vp, &vattr, 0, afs_osi_credp);
#endif
    AFS_GLOCK();
    if (code) {
	osi_Panic("VnodeToSize");
    }
    ReleaseWriteLock(&afs_xosi);
    return (afs_int32) (vattr.va_size);
}

int
osi_UFSTruncate(struct osi_file *afile, afs_int32 asize)
{
    afs_ucred_t *oldCred;
    struct vattr tvattr;
    afs_int32 code;
    struct osi_stat tstat;
    AFS_STATCNT(osi_Truncate);

    /* This routine only shrinks files, and most systems
     * have very slow truncates, even when the file is already
     * small enough.  Check now and save some time.
     */
    code = afs_osi_Stat(afile, &tstat);
    if (code || tstat.size <= asize)
	return code;
    ObtainWriteLock(&afs_xosi, 321);
    AFS_GUNLOCK();
#ifdef AFS_DARWIN80_ENV
    VATTR_INIT(&tvattr);
    VATTR_SET(&tvattr, va_size, asize);
    code = vnode_setattr(afile->vnode, &tvattr, afs_osi_ctxtp);
#else
    VATTR_NULL(&tvattr);
    tvattr.va_size = asize;
    code = VOP_SETATTR(afile->vnode, &tvattr, &afs_osi_cred, current_proc());
#endif
    AFS_GLOCK();
    ReleaseWriteLock(&afs_xosi);
    return code;
}

int
osi_UFSTruncate(struct osi_file *afile, afs_int32 asize)
{
    afs_ucred_t *oldCred;
    struct vattr tvattr;
    afs_int32 code;
    struct osi_stat tstat;
    mon_state_t ms;
    AFS_STATCNT(osi_Truncate);

    /* This routine only shrinks files, and most systems
     * have very slow truncates, even when the file is already
     * small enough.  Check now and save some time.
     */
    code = afs_osi_Stat(afile, &tstat);
    if (code || tstat.size <= asize)
	return code;
    ObtainWriteLock(&afs_xosi, 321);
    AFS_GUNLOCK();
    tvattr.va_mask = AT_SIZE;
    tvattr.va_size = asize;
    AFS_VOP_SETATTR(afile->vnode, &tvattr, 0, &afs_osi_cred, code);
    AFS_GLOCK();
    ReleaseWriteLock(&afs_xosi);
    return code;
}

int
afs_MemCacheTruncate(struct osi_file *fP, int size)
{
    struct memCacheEntry *mceP = (struct memCacheEntry *)fP;
    AFS_STATCNT(afs_MemCacheTruncate);

    ObtainWriteLock(&mceP->afs_memLock, 313);
    /* old directory entry; g.c. */
    if (size == 0 && mceP->dataSize > memCacheBlkSize) {
	char *oldData = mceP->data;
	mceP->data = afs_osi_Alloc(memCacheBlkSize);
	if (mceP->data == NULL) {	/* no available memory */
	    mceP->data = oldData;
	    ReleaseWriteLock(&mceP->afs_memLock);
	    afs_warn("afs: afs_MemWriteBlk mem alloc failure (%d bytes)\n",
		     memCacheBlkSize);
	} else {
	    afs_osi_Free(oldData, mceP->dataSize);
	    mceP->dataSize = memCacheBlkSize;
	}
    }

    if (size < mceP->size)
	mceP->size = size;

    ReleaseWriteLock(&mceP->afs_memLock);
    return 0;
}

/*!
 * Find the first dcache of a file that has the specified fid.
 * Similar to afs_FindDCache, only that it takes a fid instead
 * of a vcache and it can get the first dcache.
 *
 * \param afid
 *
 * \return The found dcache or NULL.
 */
struct dcache *
afs_FindDCacheByFid(struct VenusFid *afid)
{
    afs_int32 i, index;
    struct dcache *tdc = NULL;

    i = DVHash(afid);
    ObtainWriteLock(&afs_xdcache, 758);
    for (index = afs_dvhashTbl[i]; index != NULLIDX;) {
	if (afs_indexUnique[index] == afid->Fid.Unique) {
	    tdc = afs_GetValidDSlot(index);
	    if (tdc) {
		ReleaseReadLock(&tdc->tlock);
		if (!FidCmp(&tdc->f.fid, afid)) {
		    break;		/* leaving refCount high for caller */
		}
		afs_PutDCache(tdc);
	    }
	}
	index = afs_dvnextTbl[index];
    }
    ReleaseWriteLock(&afs_xdcache);

    if (index == NULLIDX)
    	tdc = NULL;
    return tdc;
}

/*!
 * Generate a fake fid (vnode and uniquifier) for a vcache
 * (either dir or normal file). The vnode is generated via
 * afs_DisconVNode and the uniquifier by getting the highest
 * uniquifier on a hash chain and incrementing it by one.
 *
 * \param afid	 The fid structre that will be filled.
 * \param avtype Vnode type: VDIR/VREG.
 * \param lock	 True indicates that xvcache may be obtained,
 * 		 False that it is already held
 *
 * \note The cell number must be completed somewhere else.
 */
void
afs_GenFakeFid(struct VenusFid *afid, afs_uint32 avtype, int lock)
{
    struct vcache *tvc;
    afs_uint32 max_unique = 0, i;

    switch (avtype) {
    case VDIR:
    	afid->Fid.Vnode = afs_DisconVnode + 1;
	break;
    case VREG:
    case VLNK:
    	afid->Fid.Vnode = afs_DisconVnode;
	break;
    }

    if (lock)
	ObtainWriteLock(&afs_xvcache, 736);
    i = VCHash(afid);
    for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
        if (tvc->f.fid.Fid.Unique > max_unique)
	    max_unique = tvc->f.fid.Fid.Unique;
    }
    if (lock)
	ReleaseWriteLock(&afs_xvcache);

    afid->Fid.Unique = max_unique + 1;
    afs_DisconVnode += 2;
    if (!afs_DisconVnode)
    	afs_DisconVnode = 2;
}

/*!
 * Generate a fake fid for a disconnected shadow dir.
 * Similar to afs_GenFakeFid, only that it uses the dhash
 * to search for a uniquifier because a shadow dir lives only
 * in the dcache.
 *
 * \param afid
 *
 * \note Don't forget to fill in afid with Cell and Volume.
 */
void
afs_GenShadowFid(struct VenusFid *afid)
{
    afs_uint32 i, index, max_unique = 1;
    struct vcache *tvc = NULL;

    /* Try generating a fid that isn't used in the vhash. */
    do {
	/* Shadow Fids are always directories */
    	afid->Fid.Vnode = afs_DisconVnode + 1;

    	i = DVHash(afid);
    	ObtainWriteLock(&afs_xdcache, 737);
    	for (index = afs_dvhashTbl[i]; index != NULLIDX; index = i) {
            i = afs_dvnextTbl[index];
            if (afs_indexUnique[index] > max_unique)
	    	max_unique = afs_indexUnique[index];
    	}

	ReleaseWriteLock(&afs_xdcache);
	afid->Fid.Unique = max_unique + 1;
	afs_DisconVnode += 2;
	if (!afs_DisconVnode)
    	    afs_DisconVnode = 2;

	/* Is this a used vnode? */
    	ObtainSharedLock(&afs_xvcache, 762);
    	tvc = afs_FindVCache(afid, 0, 1);
    	ReleaseSharedLock(&afs_xvcache);
	if (tvc)
	    afs_PutVCache(tvc);
    } while (tvc);
}

int
DFlush(void)
{
    /* Flush all the modified buffers. */
    int i;
    struct buffer **tbp;
    afs_int32 code, rcode;

    rcode = 0;
    tbp = Buffers;
    ObtainReadLock(&afs_bufferLock);
    for (i = 0; i < nbuffers; i++, tbp++) {
	if ((*tbp)->dirty) {
	    ObtainWriteLock(&(*tbp)->lock);
	    (*tbp)->lockers++;
	    ReleaseReadLock(&afs_bufferLock);
	    if ((*tbp)->dirty) {
		code = ReallyWrite((*tbp)->fid, (*tbp)->page, (*tbp)->data);
		if (!code)
		    (*tbp)->dirty = 0;	/* Clear the dirty flag */
		if (code && !rcode) {
		    rcode = code;
		}
	    }
	    (*tbp)->lockers--;
	    ReleaseWriteLock(&(*tbp)->lock);
	    ObtainReadLock(&afs_bufferLock);
	}
    }
    ReleaseReadLock(&afs_bufferLock);
    return rcode;
}

int
DFlushEntry(afs_int32 *fid)
{
    /* Flush pages modified by one entry. */
    struct buffer *tb;
    int code;

    ObtainReadLock(&afs_bufferLock);
    for (tb = phTable[pHash(fid)]; tb; tb = tb->hashNext)
	if (FidEq(tb->fid, fid) && tb->dirty) {
	    ObtainWriteLock(&tb->lock);
	    if (tb->dirty) {
		code = ReallyWrite(tb->fid, tb->page, tb->data);
		if (code) {
		    ReleaseWriteLock(&tb->lock);
		    ReleaseReadLock(&afs_bufferLock);
		    return code;
		}
		tb->dirty = 0;
	    }
	    ReleaseWriteLock(&tb->lock);
	}
    ReleaseReadLock(&afs_bufferLock);
    return 0;
}

/* allocate space for sender */
void *
osi_AllocSmallSpace(size_t size)
{
    struct osi_packet *tp;

    AFS_STATCNT(osi_AllocSmallSpace);
    if (size > AFS_SMALLOCSIZ)
	osi_Panic("osi_AllocSmallS: size=%d\n", (int)size);

    if (!freeSmallList) {
	afs_stats_cmperf.SmallBlocksAlloced++;
	afs_stats_cmperf.SmallBlocksActive++;
	tp = afs_osi_Alloc(AFS_SMALLOCSIZ);
#ifdef KERNEL_HAVE_PIN
        pin((char *)tp, AFS_SMALLOCSIZ);
#endif
        return (char *)tp;
    }
    afs_stats_cmperf.SmallBlocksActive++;
    ObtainWriteLock(&osi_fsplock, 327);
    tp = freeSmallList;
    if (tp)
	freeSmallList = tp->next;
    ReleaseWriteLock(&osi_fsplock);
    return (char *)tp;
}

/* allocate space for sender */
void *
osi_AllocLargeSpace(size_t size)
{
    struct osi_packet *tp;

    AFS_ASSERT_GLOCK();

    AFS_STATCNT(osi_AllocLargeSpace);
    if (size > AFS_LRALLOCSIZ)
	osi_Panic("osi_AllocLargeSpace: size=%d\n", (int)size);
    afs_stats_cmperf.LargeBlocksActive++;
    if (!freePacketList) {
	char *p;

	afs_stats_cmperf.LargeBlocksAlloced++;
	p = afs_osi_Alloc(AFS_LRALLOCSIZ);
#ifdef  KERNEL_HAVE_PIN
	/*
	 * Need to pin this memory since under heavy conditions this memory
	 * could be swapped out; the problem is that we could inside rx where
	 * interrupts are disabled and thus we would panic if we don't pin it.
	 */
	pin(p, AFS_LRALLOCSIZ);
#endif
	return p;
    }
    ObtainWriteLock(&osi_flplock, 324);
    tp = freePacketList;
    if (tp)
	freePacketList = tp->next;
    ReleaseWriteLock(&osi_flplock);
    return (char *)tp;
}

int
afs_osi_Stat(struct osi_file *afile, struct osi_stat *astat)
{
    afs_int32 code;
    struct vattr tvattr;
    AFS_STATCNT(osi_Stat);
    ObtainWriteLock(&afs_xosi, 320);
    AFS_GUNLOCK();
#if defined(AFS_FBSD80_ENV)
    vn_lock(afile->vnode, LK_EXCLUSIVE | LK_RETRY);
    code = VOP_GETATTR(afile->vnode, &tvattr, afs_osi_credp);
    VOP_UNLOCK(afile->vnode, 0);
#else
    vn_lock(afile->vnode, LK_EXCLUSIVE | LK_RETRY, curthread);
    code = VOP_GETATTR(afile->vnode, &tvattr, afs_osi_credp, curthread);
    VOP_UNLOCK(afile->vnode, LK_EXCLUSIVE, curthread);
#endif
    AFS_GLOCK();
    if (code == 0) {
	astat->size = tvattr.va_size;
	astat->mtime = tvattr.va_mtime.tv_sec;
	astat->atime = tvattr.va_atime.tv_sec;
    }
    ReleaseWriteLock(&afs_xosi);
    return code;
}

/*
 * Invalidate the /afs vnode for dynroot; called when the underlying
 * directory has changed and needs to be re-read.
 */
void
afs_DynrootInvalidate(void)
{
    afs_int32 retry;
    struct vcache *tvc;
    struct VenusFid tfid;

    if (!afs_dynrootEnable)
	return;

    ObtainWriteLock(&afs_dynrootDirLock, 687);
    afs_dynrootVersion++;
    afs_dynrootVersionHigh = osi_Time();
    ReleaseWriteLock(&afs_dynrootDirLock);

    afs_GetDynrootFid(&tfid);
    do {
	retry = 0;
	ObtainReadLock(&afs_xvcache);
	tvc = afs_FindVCache(&tfid, &retry, 0);
	ReleaseReadLock(&afs_xvcache);
    } while (retry);
    if (tvc) {
	tvc->f.states &= ~(CStatd | CUnique);
	osi_dnlc_purgedp(tvc);
	afs_PutVCache(tvc);
    }
}