Beispiel #1
0
/*
 * afs_TruncateAllSegments
 *
 * Description:
 *	Truncate a cache file.
 *
 * Parameters:
 *	avc  : Ptr to vcache entry to truncate.
 *	alen : Number of bytes to make the file.
 *	areq : Ptr to request structure.
 *
 * Environment:
 *	Called with avc write-locked; in VFS40 systems, pvnLock is also
 *	held.
 */
int
afs_TruncateAllSegments(struct vcache *avc, afs_size_t alen,
			struct vrequest *areq, afs_ucred_t *acred)
{
    struct dcache *tdc;
    afs_int32 code;
    afs_int32 index;
    afs_size_t newSize;

    int dcCount, dcPos;
    struct dcache **tdcArray = NULL;

    AFS_STATCNT(afs_TruncateAllSegments);
    avc->f.m.Date = osi_Time();
    afs_Trace3(afs_iclSetp, CM_TRACE_TRUNCALL, ICL_TYPE_POINTER, avc,
	       ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length),
	       ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(alen));
    if (alen >= avc->f.m.Length) {
	/*
	 * Special speedup since Sun's vm extends the file this way;
	 * we've never written to the file thus we can just set the new
	 * length and avoid the needless calls below.
	 * Also used for ftruncate calls which can extend the file.
	 * To completely minimize the possible extra StoreMini RPC, we really
	 * should keep the ExtendedPos as well and clear this flag if we
	 * truncate below that value before we store the file back.
	 */
	avc->f.states |= CExtendedFile;
	avc->f.m.Length = alen;
	return 0;
    }
#if	(defined(AFS_SUN5_ENV))

    /* Zero unused portion of last page */
    osi_VM_PreTruncate(avc, alen, acred);

#endif

#if	(defined(AFS_SUN5_ENV))
    ObtainWriteLock(&avc->vlock, 546);
    avc->activeV++;		/* Block new getpages */
    ReleaseWriteLock(&avc->vlock);
#endif

    ReleaseWriteLock(&avc->lock);
    AFS_GUNLOCK();

    /* Flush pages beyond end-of-file. */
    osi_VM_Truncate(avc, alen, acred);

    AFS_GLOCK();
    ObtainWriteLock(&avc->lock, 79);

    avc->f.m.Length = alen;

    if (alen < avc->f.truncPos)
	avc->f.truncPos = alen;
    code = DVHash(&avc->f.fid);

    /* block out others from screwing with this table */
    ObtainWriteLock(&afs_xdcache, 287);

    dcCount = 0;
    for (index = afs_dvhashTbl[code]; index != NULLIDX;) {
	if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
	    tdc = afs_GetValidDSlot(index);
	    if (!tdc) {
		ReleaseWriteLock(&afs_xdcache);
		code = EIO;
		goto done;
	    }
	    ReleaseReadLock(&tdc->tlock);
	    if (!FidCmp(&tdc->f.fid, &avc->f.fid))
		dcCount++;
	    afs_PutDCache(tdc);
	}
	index = afs_dvnextTbl[index];
    }

    /* Now allocate space where we can save those dcache entries, and
     * do a second pass over them..  Since we're holding xdcache, it
     * shouldn't be changing.
     */
    tdcArray = osi_Alloc(dcCount * sizeof(struct dcache *));
    dcPos = 0;

    for (index = afs_dvhashTbl[code]; index != NULLIDX; index = afs_dvnextTbl[index]) {
	if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
	    tdc = afs_GetValidDSlot(index);
	    if (!tdc) {
		/* make sure we put back all of the tdcArray members before
		 * bailing out */
		/* remember, the last valid tdc is at dcPos-1, so start at
		 * dcPos-1, not at dcPos itself. */
		for (dcPos = dcPos - 1; dcPos >= 0; dcPos--) {
		    tdc = tdcArray[dcPos];
		    afs_PutDCache(tdc);
		}
		code = EIO;
		goto done;
	    }
	    ReleaseReadLock(&tdc->tlock);
	    if (!FidCmp(&tdc->f.fid, &avc->f.fid)) {
		/* same file, and modified, we'll store it back */
		if (dcPos < dcCount) {
		    tdcArray[dcPos++] = tdc;
		} else {
		    afs_PutDCache(tdc);
		}
	    } else {
		afs_PutDCache(tdc);
	    }
	}
    }

    ReleaseWriteLock(&afs_xdcache);

    /* Now we loop over the array of dcache entries and truncate them */
    for (index = 0; index < dcPos; index++) {
	struct osi_file *tfile;

	tdc = tdcArray[index];

	newSize = alen - AFS_CHUNKTOBASE(tdc->f.chunk);
	if (newSize < 0)
	    newSize = 0;
	ObtainSharedLock(&tdc->lock, 672);
	if (newSize < tdc->f.chunkBytes && newSize < MAX_AFS_UINT32) {
	    UpgradeSToWLock(&tdc->lock, 673);
	    tdc->f.states |= DWriting;
	    tfile = afs_CFileOpen(&tdc->f.inode);
	    afs_CFileTruncate(tfile, (afs_int32)newSize);
	    afs_CFileClose(tfile);
	    afs_AdjustSize(tdc, (afs_int32)newSize);
	    if (alen < tdc->validPos) {
                if (alen < AFS_CHUNKTOBASE(tdc->f.chunk))
                    tdc->validPos = 0;
                else
                    tdc->validPos = alen;
            }
	    ConvertWToSLock(&tdc->lock);
	}
	ReleaseSharedLock(&tdc->lock);
	afs_PutDCache(tdc);
    }

    code = 0;

 done:
    if (tdcArray) {
	osi_Free(tdcArray, dcCount * sizeof(struct dcache *));
    }
#if	(defined(AFS_SUN5_ENV))
    ObtainWriteLock(&avc->vlock, 547);
    if (--avc->activeV == 0 && (avc->vstates & VRevokeWait)) {
	avc->vstates &= ~VRevokeWait;
	afs_osi_Wakeup((char *)&avc->vstates);
    }
    ReleaseWriteLock(&avc->vlock);
#endif

    return code;
}
Beispiel #2
0
int
afs_MemRead(struct vcache *avc, struct uio *auio,
	    afs_ucred_t *acred, daddr_t albn, struct buf **abpp,
	    int noLock)
{
    afs_size_t totalLength;
    afs_size_t transferLength;
    afs_size_t filePos;
    afs_size_t offset, tlen;
    afs_size_t len = 0;
    afs_int32 trimlen;
    struct dcache *tdc = 0;
    afs_int32 error, trybusy = 1;
    afs_int32 code;
    struct vrequest *treq = NULL;
#ifdef AFS_DARWIN80_ENV
    uio_t tuiop = NULL;
#else
    struct uio tuio;
    struct uio *tuiop = &tuio;
    struct iovec *tvec;
    memset(&tuio, 0, sizeof(tuio));
#endif

    AFS_STATCNT(afs_MemRead);
    if (avc->vc_error)
	return EIO;

    /* check that we have the latest status info in the vnode cache */
    if ((code = afs_CreateReq(&treq, acred)))
	return code;
    if (!noLock) {
	code = afs_VerifyVCache(avc, treq);
	if (code) {
	    code = afs_CheckCode(code, treq, 8);	/* failed to get it */
	    afs_DestroyReq(treq);
	    return code;
	}
    }
#ifndef	AFS_VM_RDWR_ENV
    if (AFS_NFSXLATORREQ(acred)) {
	if (!afs_AccessOK
	    (avc, PRSFS_READ, treq,
	     CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
	    code = afs_CheckCode(EACCES, treq, 9);
	    afs_DestroyReq(treq);
	    return code;
	}
    }
#endif

#ifndef AFS_DARWIN80_ENV
    tvec = (struct iovec *)osi_AllocSmallSpace(sizeof(struct iovec));
    memset(tvec, 0, sizeof(struct iovec));
#endif
    totalLength = AFS_UIO_RESID(auio);
    filePos = AFS_UIO_OFFSET(auio);
    afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
	       ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(filePos), ICL_TYPE_INT32,
	       totalLength, ICL_TYPE_OFFSET,
	       ICL_HANDLE_OFFSET(avc->f.m.Length));
    error = 0;
    transferLength = 0;
    if (!noLock)
	ObtainReadLock(&avc->lock);
#if	defined(AFS_TEXT_ENV) && !defined(AFS_VM_RDWR_ENV)
    if (avc->flushDV.high == AFS_MAXDV && avc->flushDV.low == AFS_MAXDV) {
	hset(avc->flushDV, avc->f.m.DataVersion);
    }
#endif

    /*
     * Locks held:
     * avc->lock(R)
     */

    /* This bit is bogus. We're checking to see if the read goes past the
     * end of the file. If so, we should be zeroing out all of the buffers
     * that the client has passed into us (there is a danger that we may leak
     * kernel memory if we do not). However, this behaviour is disabled by
     * not setting len before this segment runs, and by setting len to 0
     * immediately we enter it. In addition, we also need to check for a read
     * which partially goes off the end of the file in the while loop below.
     */

    if (filePos >= avc->f.m.Length) {
	if (len > AFS_ZEROS)
	    len = sizeof(afs_zeros);	/* and in 0 buffer */
	len = 0;
#ifdef AFS_DARWIN80_ENV
	trimlen = len;
	tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
	afsio_copy(auio, &tuio, tvec);
	trimlen = len;
	afsio_trim(&tuio, trimlen);
#endif
	AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
    }

    while (avc->f.m.Length > 0 && totalLength > 0) {
	/* read all of the cached info */
	if (filePos >= avc->f.m.Length)
	    break;		/* all done */
	if (noLock) {
	    if (tdc) {
		ReleaseReadLock(&tdc->lock);
		afs_PutDCache(tdc);
	    }
	    tdc = afs_FindDCache(avc, filePos);
	    if (tdc) {
		ObtainReadLock(&tdc->lock);
		offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
		len = tdc->f.chunkBytes - offset;
	    }
	} else {
	    int versionOk;
	    /* a tricky question: does the presence of the DFFetching flag
	     * mean that we're fetching the latest version of the file?  No.
	     * The server could update the file as soon as the fetch responsible
	     * for the setting of the DFFetching flag completes.
	     *
	     * However, the presence of the DFFetching flag (visible under
	     * a dcache read lock since it is set and cleared only under a
	     * dcache write lock) means that we're fetching as good a version
	     * as was known to this client at the time of the last call to
	     * afs_VerifyVCache, since the latter updates the stat cache's
	     * m.DataVersion field under a vcache write lock, and from the
	     * time that the DFFetching flag goes on in afs_GetDCache (before
	     * the fetch starts), to the time it goes off (after the fetch
	     * completes), afs_GetDCache keeps at least a read lock on the
	     * vcache entry.
	     *
	     * This means that if the DFFetching flag is set, we can use that
	     * data for any reads that must come from the current version of
	     * the file (current == m.DataVersion).
	     *
	     * Another way of looking at this same point is this: if we're
	     * fetching some data and then try do an afs_VerifyVCache, the
	     * VerifyVCache operation will not complete until after the
	     * DFFetching flag is turned off and the dcache entry's f.versionNo
	     * field is updated.
	     *
	     * Note, by the way, that if DFFetching is set,
	     * m.DataVersion > f.versionNo (the latter is not updated until
	     * after the fetch completes).
	     */
	    if (tdc) {
		ReleaseReadLock(&tdc->lock);
		afs_PutDCache(tdc);	/* before reusing tdc */
	    }
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
    try_background:
#endif
	    tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 2);
	    ObtainReadLock(&tdc->lock);
	    /* now, first try to start transfer, if we'll need the data.  If
	     * data already coming, we don't need to do this, obviously.  Type
	     * 2 requests never return a null dcache entry, btw.
	     */
	    if (!(tdc->dflags & DFFetching)
		&& !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
		/* have cache entry, it is not coming in now,
		 * and we'll need new data */
	      tagain:
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
		if (trybusy && (!afs_BBusy() || (afs_protocols & VICEP_ACCESS))) {
#else
		if (trybusy && !afs_BBusy()) {
#endif
		    struct brequest *bp;
		    /* daemon is not busy */
		    ObtainSharedLock(&tdc->mflock, 665);
		    if (!(tdc->mflags & DFFetchReq)) {
			int dontwait = B_DONTWAIT;
			/* start the daemon (may already be running, however) */
			UpgradeSToWLock(&tdc->mflock, 666);
			tdc->mflags |= DFFetchReq;
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
			if (afs_protocols & VICEP_ACCESS)
			    dontwait = 0;
#endif
			bp = afs_BQueue(BOP_FETCH, avc, dontwait, 0, acred,
					(afs_size_t) filePos, (afs_size_t) 0,
					tdc, (void *)0, (void *)0);
			if (!bp) {
			    tdc->mflags &= ~DFFetchReq;
			    trybusy = 0;	/* Avoid bkg daemon since they're too busy */
			    ReleaseWriteLock(&tdc->mflock);
			    goto tagain;
			}
			ConvertWToSLock(&tdc->mflock);
			/* don't use bp pointer! */
		    }
		    code = 0;
		    ConvertSToRLock(&tdc->mflock);
		    while (!code && tdc->mflags & DFFetchReq) {
			afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT,
				   ICL_TYPE_STRING, __FILE__, ICL_TYPE_INT32,
				   __LINE__, ICL_TYPE_POINTER, tdc,
				   ICL_TYPE_INT32, tdc->dflags);
			/* don't need waiting flag on this one */
			ReleaseReadLock(&tdc->mflock);
			ReleaseReadLock(&tdc->lock);
			ReleaseReadLock(&avc->lock);
			code = afs_osi_SleepSig(&tdc->validPos);
			ObtainReadLock(&avc->lock);
			ObtainReadLock(&tdc->lock);
			ObtainReadLock(&tdc->mflock);
		    }
		    ReleaseReadLock(&tdc->mflock);
		    if (code) {
			error = code;
			break;
		    }
		}
	    }
	    /* now data may have started flowing in (if DFFetching is on).  If
	     * data is now streaming in, then wait for some interesting stuff.
	     */
	    code = 0;
	    while (!code && (tdc->dflags & DFFetching)
		   && tdc->validPos <= filePos) {
		/* too early: wait for DFFetching flag to vanish,
		 * or data to appear */
		afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
			   __FILE__, ICL_TYPE_INT32, __LINE__,
			   ICL_TYPE_POINTER, tdc, ICL_TYPE_INT32,
			   tdc->dflags);
		ReleaseReadLock(&tdc->lock);
		ReleaseReadLock(&avc->lock);
		code = afs_osi_SleepSig(&tdc->validPos);
		ObtainReadLock(&avc->lock);
		ObtainReadLock(&tdc->lock);
	    }
	    if (code) {
		error = code;
		break;
	    }
	    /* fetching flag gone, data is here, or we never tried
	     * (BBusy for instance) */
	    len = tdc->validPos - filePos;
	    versionOk = hsame(avc->f.m.DataVersion, tdc->f.versionNo) ? 1 : 0;
	    if (tdc->dflags & DFFetching) {
		/* still fetching, some new data is here:
		 * compute length and offset */
		offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
	    } else {
		/* no longer fetching, verify data version
		 * (avoid new GetDCache call) */
		if (versionOk && len > 0) {
		    offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
		} else {
		    /* don't have current data, so get it below */
		    afs_Trace3(afs_iclSetp, CM_TRACE_VERSIONNO,
			       ICL_TYPE_INT64, ICL_HANDLE_OFFSET(filePos),
			       ICL_TYPE_HYPER, &avc->f.m.DataVersion,
			       ICL_TYPE_HYPER, &tdc->f.versionNo);
#if 0
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS) {
printf("afs_read: DV mismatch? %d instead of %d for %u.%u.%u\n", tdc->f.versionNo.low, avc->f.m.DataVersion.low, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
printf("afs_read: validPos %llu filePos %llu totalLength %lld m.Length %llu noLock %d\n", tdc->validPos, filePos, totalLength, avc->f.m.Length, noLock);
printf("afs_read: or len too low? %lld for %u.%u.%u\n", len, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
}
#endif
#endif
		    ReleaseReadLock(&tdc->lock);
		    afs_PutDCache(tdc);
		    tdc = NULL;
		}
	    }

	    if (!tdc) {
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
		if (afs_protocols & VICEP_ACCESS) { /* avoid foreground fetch */
		    if (!versionOk) {
printf("afs_read: avoid forground %u.%u.%u\n", avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
		        goto try_background;
		    }
#if 0
printf("afs_read: forground %u.%u.%u\n", avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
#endif
		}
#endif
		/* If we get here, it was not possible to start the
		 * background daemon. With flag == 1 afs_GetDCache
		 * does the FetchData rpc synchronously.
		 */
		ReleaseReadLock(&avc->lock);
		tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 1);
		ObtainReadLock(&avc->lock);
		if (tdc)
		    ObtainReadLock(&tdc->lock);
	    }
	}

	afs_Trace3(afs_iclSetp, CM_TRACE_VNODEREAD, ICL_TYPE_POINTER, tdc,
		   ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(offset),
		   ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(len));
	if (!tdc) {
	    error = EIO;
	    break;
	}

	/*
	 * Locks held:
	 * avc->lock(R)
	 * tdc->lock(R)
	 */

	if (len > totalLength)
	    len = totalLength;	/* will read len bytes */
	if (len <= 0) {		/* shouldn't get here if DFFetching is on */
	    /* read past the end of a chunk, may not be at next chunk yet, and yet
	     * also not at eof, so may have to supply fake zeros */
	    len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset;	/* bytes left in chunk addr space */
	    if (len > totalLength)
		len = totalLength;	/* and still within xfr request */
	    tlen = avc->f.m.Length - offset;	/* and still within file */
	    if (len > tlen)
		len = tlen;
	    if (len > AFS_ZEROS)
		len = sizeof(afs_zeros);	/* and in 0 buffer */
#ifdef AFS_DARWIN80_ENV
	    trimlen = len;
            tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
	    afsio_copy(auio, &tuio, tvec);
	    trimlen = len;
	    afsio_trim(&tuio, trimlen);
#endif
	        AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
	    if (code) {
		error = code;
		break;
	    }
	} else {
	    /* get the data from the mem cache */

	    /* mung uio structure to be right for this transfer */
#ifdef AFS_DARWIN80_ENV
	    trimlen = len;
            tuiop = afsio_darwin_partialcopy(auio, trimlen);
	    uio_setoffset(tuiop, offset);
#else
	    afsio_copy(auio, &tuio, tvec);
	    trimlen = len;
	    afsio_trim(&tuio, trimlen);
	    tuio.afsio_offset = offset;
#endif

	    code = afs_MemReadUIO(&tdc->f.inode, tuiop);

	    if (code) {
		error = code;
		break;
	    }
	}
	/* otherwise we've read some, fixup length, etc and continue with next seg */
	len = len - AFS_UIO_RESID(tuiop);	/* compute amount really transferred */
	trimlen = len;
	afsio_skip(auio, trimlen);	/* update input uio structure */
	totalLength -= len;
	transferLength += len;
	filePos += len;

	if (len <= 0)
	    break;		/* surprise eof */
#ifdef AFS_DARWIN80_ENV
	if (tuiop) {
	    uio_free(tuiop);
	    tuiop = 0;
	}
#endif
    }				/* the whole while loop */

    /*
     * Locks held:
     * avc->lock(R)
     * tdc->lock(R) if tdc
     */

    /* if we make it here with tdc non-zero, then it is the last chunk we
     * dealt with, and we have to release it when we're done.  We hold on
     * to it in case we need to do a prefetch.
     */
    if (tdc) {
	ReleaseReadLock(&tdc->lock);
	/*
         * try to queue prefetch, if needed. If DataVersion is zero there
	 * should not be any more: files with DV 0 never have been stored
	 * on the fileserver, symbolic links and directories never require
	 * more than a single chunk.
	 */
	if (!noLock && !(hiszero(avc->f.m.DataVersion)) &&
#ifndef AFS_VM_RDWR_ENV
	    afs_preCache
#else
	    1
#endif
	    ) {
	    afs_PrefetchChunk(avc, tdc, acred, treq);
	}
	afs_PutDCache(tdc);
    }
    if (!noLock)
	ReleaseReadLock(&avc->lock);
#ifdef AFS_DARWIN80_ENV
    if (tuiop)
       uio_free(tuiop);
#else
    osi_FreeSmallSpace(tvec);
#endif
    error = afs_CheckCode(error, treq, 10);
    afs_DestroyReq(treq);
    return error;
}

/* called with the dcache entry triggering the fetch, the vcache entry involved,
 * and a vrequest for the read call.  Marks the dcache entry as having already
 * triggered a prefetch, starts the prefetch going and sets the DFFetchReq
 * flag in the prefetched block, so that the next call to read knows to wait
 * for the daemon to start doing things.
 *
 * This function must be called with the vnode at least read-locked, and
 * no locks on the dcache, because it plays around with dcache entries.
 */
void
afs_PrefetchChunk(struct vcache *avc, struct dcache *adc,
		  afs_ucred_t *acred, struct vrequest *areq)
{
    struct dcache *tdc;
    afs_size_t offset;
    afs_size_t j1, j2;		/* junk vbls for GetDCache to trash */

    offset = adc->f.chunk + 1;	/* next chunk we'll need */
    offset = AFS_CHUNKTOBASE(offset);	/* base of next chunk */
    ObtainReadLock(&adc->lock);
    ObtainSharedLock(&adc->mflock, 662);
    if (offset < avc->f.m.Length && !(adc->mflags & DFNextStarted)
	&& !afs_BBusy()) {
	struct brequest *bp;

	UpgradeSToWLock(&adc->mflock, 663);
	adc->mflags |= DFNextStarted;	/* we've tried to prefetch for this guy */
	ReleaseWriteLock(&adc->mflock);
	ReleaseReadLock(&adc->lock);

	tdc = afs_GetDCache(avc, offset, areq, &j1, &j2, 2);	/* type 2 never returns 0 */
        /*
         * In disconnected mode, type 2 can return 0 because it doesn't
         * make any sense to allocate a dcache we can never fill
         */
         if (tdc == NULL)
             return;

	ObtainSharedLock(&tdc->mflock, 651);
	if (!(tdc->mflags & DFFetchReq)) {
	    /* ask the daemon to do the work */
	    UpgradeSToWLock(&tdc->mflock, 652);
	    tdc->mflags |= DFFetchReq;	/* guaranteed to be cleared by BKG or GetDCache */
	    /* last parm (1) tells bkg daemon to do an afs_PutDCache when it is done,
	     * since we don't want to wait for it to finish before doing so ourselves.
	     */
	    bp = afs_BQueue(BOP_FETCH, avc, B_DONTWAIT, 0, acred,
			    (afs_size_t) offset, (afs_size_t) 1, tdc,
			    (void *)0, (void *)0);
	    if (!bp) {
		/* Bkg table full; just abort non-important prefetching to avoid deadlocks */
		tdc->mflags &= ~DFFetchReq;
		ReleaseWriteLock(&tdc->mflock);
		afs_PutDCache(tdc);

		/*
		 * DCLOCKXXX: This is a little sketchy, since someone else
		 * could have already started a prefetch..  In practice,
		 * this probably doesn't matter; at most it would cause an
		 * extra slot in the BKG table to be used up when someone
		 * prefetches this for the second time.
		 */
		ObtainReadLock(&adc->lock);
		ObtainWriteLock(&adc->mflock, 664);
		adc->mflags &= ~DFNextStarted;
		ReleaseWriteLock(&adc->mflock);
		ReleaseReadLock(&adc->lock);
	    } else {
		ReleaseWriteLock(&tdc->mflock);
	    }
	} else {
	    ReleaseSharedLock(&tdc->mflock);
	    afs_PutDCache(tdc);
	}
    } else {
	ReleaseSharedLock(&adc->mflock);
	ReleaseReadLock(&adc->lock);
    }
}

int
afs_UFSRead(struct vcache *avc, struct uio *auio,
	    afs_ucred_t *acred, daddr_t albn, struct buf **abpp,
	    int noLock)
{
    afs_size_t totalLength;
    afs_size_t transferLength;
    afs_size_t filePos;
    afs_size_t offset, tlen;
    afs_size_t len = 0;
    afs_int32 trimlen;
    struct dcache *tdc = 0;
    afs_int32 error;
    struct osi_file *tfile;
    afs_int32 code;
    int trybusy = 1;
    struct vrequest *treq = NULL;
#ifdef AFS_DARWIN80_ENV
    uio_t tuiop=NULL;
#else
    struct uio tuio;
    struct uio *tuiop = &tuio;
    struct iovec *tvec;
    memset(&tuio, 0, sizeof(tuio));
#endif

    AFS_STATCNT(afs_UFSRead);
    if (avc && avc->vc_error)
	return EIO;

    AFS_DISCON_LOCK();

    /* check that we have the latest status info in the vnode cache */
    if ((code = afs_CreateReq(&treq, acred)))
	return code;
    if (!noLock) {
	if (!avc)
	    osi_Panic("null avc in afs_UFSRead");
	else {
	    code = afs_VerifyVCache(avc, treq);
	    if (code) {
		code = afs_CheckCode(code, treq, 11);	/* failed to get it */
		afs_DestroyReq(treq);
		AFS_DISCON_UNLOCK();
		return code;
	    }
	}
    }
#ifndef	AFS_VM_RDWR_ENV
    if (AFS_NFSXLATORREQ(acred)) {
	if (!afs_AccessOK
	    (avc, PRSFS_READ, treq,
	     CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
	    AFS_DISCON_UNLOCK();
	    code = afs_CheckCode(EACCES, treq, 12);
	    afs_DestroyReq(treq);
	    return code;
	}
    }
#endif

#ifndef AFS_DARWIN80_ENV
    tvec = (struct iovec *)osi_AllocSmallSpace(sizeof(struct iovec));
    memset(tvec, 0, sizeof(struct iovec));
#endif
    totalLength = AFS_UIO_RESID(auio);
    filePos = AFS_UIO_OFFSET(auio);
    afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
	       ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(filePos), ICL_TYPE_INT32,
	       totalLength, ICL_TYPE_OFFSET,
	       ICL_HANDLE_OFFSET(avc->f.m.Length));
    error = 0;
    transferLength = 0;
    if (!noLock)
	ObtainReadLock(&avc->lock);
#if	defined(AFS_TEXT_ENV) && !defined(AFS_VM_RDWR_ENV)
    if (avc->flushDV.high == AFS_MAXDV && avc->flushDV.low == AFS_MAXDV) {
	hset(avc->flushDV, avc->f.m.DataVersion);
    }
#endif

    /* This bit is bogus. We're checking to see if the read goes past the
     * end of the file. If so, we should be zeroing out all of the buffers
     * that the client has passed into us (there is a danger that we may leak
     * kernel memory if we do not). However, this behaviour is disabled by
     * not setting len before this segment runs, and by setting len to 0
     * immediately we enter it. In addition, we also need to check for a read
     * which partially goes off the end of the file in the while loop below.
     */

    if (filePos >= avc->f.m.Length) {
	if (len > AFS_ZEROS)
	    len = sizeof(afs_zeros);	/* and in 0 buffer */
	len = 0;
#ifdef AFS_DARWIN80_ENV
	trimlen = len;
	tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
	afsio_copy(auio, &tuio, tvec);
	trimlen = len;
	afsio_trim(&tuio, trimlen);
#endif
	AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
    }

    while (avc->f.m.Length > 0 && totalLength > 0) {
	/* read all of the cached info */
	if (filePos >= avc->f.m.Length)
	    break;		/* all done */
	if (noLock) {
	    if (tdc) {
		ReleaseReadLock(&tdc->lock);
		afs_PutDCache(tdc);
	    }
	    tdc = afs_FindDCache(avc, filePos);
	    if (tdc) {
		ObtainReadLock(&tdc->lock);
		offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
		len = tdc->validPos - filePos;
	    }
	} else {
	    int versionOk;
	    /* a tricky question: does the presence of the DFFetching flag
	     * mean that we're fetching the latest version of the file?  No.
	     * The server could update the file as soon as the fetch responsible
	     * for the setting of the DFFetching flag completes.
	     *
	     * However, the presence of the DFFetching flag (visible under
	     * a dcache read lock since it is set and cleared only under a
	     * dcache write lock) means that we're fetching as good a version
	     * as was known to this client at the time of the last call to
	     * afs_VerifyVCache, since the latter updates the stat cache's
	     * m.DataVersion field under a vcache write lock, and from the
	     * time that the DFFetching flag goes on in afs_GetDCache (before
	     * the fetch starts), to the time it goes off (after the fetch
	     * completes), afs_GetDCache keeps at least a read lock on the
	     * vcache entry.
	     *
	     * This means that if the DFFetching flag is set, we can use that
	     * data for any reads that must come from the current version of
	     * the file (current == m.DataVersion).
	     *
	     * Another way of looking at this same point is this: if we're
	     * fetching some data and then try do an afs_VerifyVCache, the
	     * VerifyVCache operation will not complete until after the
	     * DFFetching flag is turned off and the dcache entry's f.versionNo
	     * field is updated.
	     *
	     * Note, by the way, that if DFFetching is set,
	     * m.DataVersion > f.versionNo (the latter is not updated until
	     * after the fetch completes).
	     */
	    if (tdc) {
		ReleaseReadLock(&tdc->lock);
		afs_PutDCache(tdc);	/* before reusing tdc */
	    }
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
    try_background:
#endif
	    tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 2);
	    if (!tdc) {
	        error = ENETDOWN;
	        break;
	    }

	    ObtainReadLock(&tdc->lock);
	    /* now, first try to start transfer, if we'll need the data.  If
	     * data already coming, we don't need to do this, obviously.  Type
	     * 2 requests never return a null dcache entry, btw. */
	    if (!(tdc->dflags & DFFetching)
		&& !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
		/* have cache entry, it is not coming in now, and we'll need new data */
	      tagain:
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
		if (trybusy && (!afs_BBusy() || (afs_protocols & VICEP_ACCESS))) {
#else
		if (trybusy && !afs_BBusy()) {
#endif
		    struct brequest *bp;
		    /* daemon is not busy */
		    ObtainSharedLock(&tdc->mflock, 667);
		    if (!(tdc->mflags & DFFetchReq)) {
			int dontwait = B_DONTWAIT;
			UpgradeSToWLock(&tdc->mflock, 668);
			tdc->mflags |= DFFetchReq;
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
			if (afs_protocols & VICEP_ACCESS)
			    dontwait = 0;
#endif
			bp = afs_BQueue(BOP_FETCH, avc, dontwait, 0, acred,
					(afs_size_t) filePos, (afs_size_t) 0,
					tdc, (void *)0, (void *)0);
			if (!bp) {
			    /* Bkg table full; retry deadlocks */
			    tdc->mflags &= ~DFFetchReq;
			    trybusy = 0;	/* Avoid bkg daemon since they're too busy */
			    ReleaseWriteLock(&tdc->mflock);
			    goto tagain;
			}
			ConvertWToSLock(&tdc->mflock);
		    }
		    code = 0;
		    ConvertSToRLock(&tdc->mflock);
		    while (!code && tdc->mflags & DFFetchReq) {
			afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT,
				   ICL_TYPE_STRING, __FILE__, ICL_TYPE_INT32,
				   __LINE__, ICL_TYPE_POINTER, tdc,
				   ICL_TYPE_INT32, tdc->dflags);
			/* don't need waiting flag on this one */
			ReleaseReadLock(&tdc->mflock);
			ReleaseReadLock(&tdc->lock);
			ReleaseReadLock(&avc->lock);
			code = afs_osi_SleepSig(&tdc->validPos);
			ObtainReadLock(&avc->lock);
			ObtainReadLock(&tdc->lock);
			ObtainReadLock(&tdc->mflock);
		    }
		    ReleaseReadLock(&tdc->mflock);
		    if (code) {
			error = code;
			break;
		    }
		}
	    }
	    /* now data may have started flowing in (if DFFetching is on).  If
	     * data is now streaming in, then wait for some interesting stuff.
	     */
	    code = 0;
	    while (!code && (tdc->dflags & DFFetching)
		   && tdc->validPos <= filePos) {
		/* too early: wait for DFFetching flag to vanish,
		 * or data to appear */
		afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
			   __FILE__, ICL_TYPE_INT32, __LINE__,
			   ICL_TYPE_POINTER, tdc, ICL_TYPE_INT32,
			   tdc->dflags);
		ReleaseReadLock(&tdc->lock);
		ReleaseReadLock(&avc->lock);
		code = afs_osi_SleepSig(&tdc->validPos);
		ObtainReadLock(&avc->lock);
		ObtainReadLock(&tdc->lock);
	    }
	    if (code) {
		error = code;
		break;
	    }
	    /* fetching flag gone, data is here, or we never tried
	     * (BBusy for instance) */
	    len = tdc->validPos - filePos;
	    versionOk = hsame(avc->f.m.DataVersion, tdc->f.versionNo) ? 1 : 0;
	    if (tdc->dflags & DFFetching) {
		/* still fetching, some new data is here:
		 * compute length and offset */
		offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
	    } else {
		/* no longer fetching, verify data version (avoid new
		 * GetDCache call) */
		if (versionOk && len > 0) {
		    offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
		} else {
		    /* don't have current data, so get it below */
		    afs_Trace3(afs_iclSetp, CM_TRACE_VERSIONNO,
			       ICL_TYPE_INT64, ICL_HANDLE_OFFSET(filePos),
			       ICL_TYPE_HYPER, &avc->f.m.DataVersion,
			       ICL_TYPE_HYPER, &tdc->f.versionNo);
#if 0
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS) {
printf("afs_read: DV mismatch? %d instead of %d for %u.%u.%u\n", tdc->f.versionNo.low, avc->f.m.DataVersion.low, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
printf("afs_read: validPos %llu filePos %llu totalLength %d m.Length %llu noLock %d\n", tdc->validPos, filePos, totalLength, avc->f.m.Length, noLock);
printf("afs_read: or len too low? %lld for %u.%u.%u\n", len, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
}
#endif
#endif
		    ReleaseReadLock(&tdc->lock);
		    afs_PutDCache(tdc);
		    tdc = NULL;
		}
	    }

	    if (!tdc) {
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
		if (afs_protocols & VICEP_ACCESS) { /* avoid foreground fetch */
		    if (!versionOk) {
printf("afs_read: avoid forground %u.%u.%u\n", avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
			goto try_background;
		    }
		}
#endif
		/* If we get here, it was not possible to start the
		 * background daemon. With flag == 1 afs_GetDCache
		 * does the FetchData rpc synchronously.
		 */
		ReleaseReadLock(&avc->lock);
		tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 1);
		ObtainReadLock(&avc->lock);
		if (tdc)
		    ObtainReadLock(&tdc->lock);
	    }
	}

	if (!tdc) {
	    error = EIO;
	    break;
	}
	len = tdc->validPos - filePos;
	afs_Trace3(afs_iclSetp, CM_TRACE_VNODEREAD, ICL_TYPE_POINTER, tdc,
		   ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(offset),
		   ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(len));
	if (len > totalLength)
	    len = totalLength;	/* will read len bytes */
	if (len <= 0) {		/* shouldn't get here if DFFetching is on */
	    afs_Trace4(afs_iclSetp, CM_TRACE_VNODEREAD2, ICL_TYPE_POINTER,
		       tdc, ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(tdc->validPos),
		       ICL_TYPE_INT32, tdc->f.chunkBytes, ICL_TYPE_INT32,
		       tdc->dflags);
	    /* read past the end of a chunk, may not be at next chunk yet, and yet
	     * also not at eof, so may have to supply fake zeros */
	    len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset;	/* bytes left in chunk addr space */
	    if (len > totalLength)
		len = totalLength;	/* and still within xfr request */
	    tlen = avc->f.m.Length - offset;	/* and still within file */
	    if (len > tlen)
		len = tlen;
	    if (len > AFS_ZEROS)
		len = sizeof(afs_zeros);	/* and in 0 buffer */
#ifdef AFS_DARWIN80_ENV
	    trimlen = len;
            tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
	    afsio_copy(auio, &tuio, tvec);
	    trimlen = len;
	    afsio_trim(&tuio, trimlen);
#endif
	    AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
	    if (code) {
		error = code;
		break;
	    }
	} else {
	    /* get the data from the file */
	    tfile = (struct osi_file *)osi_UFSOpen(&tdc->f.inode);
#ifdef AFS_DARWIN80_ENV
	    trimlen = len;
            tuiop = afsio_darwin_partialcopy(auio, trimlen);
	    uio_setoffset(tuiop, offset);
#else
	    /* mung uio structure to be right for this transfer */
	    afsio_copy(auio, &tuio, tvec);
	    trimlen = len;
	    afsio_trim(&tuio, trimlen);
	    tuio.afsio_offset = offset;
#endif

#if defined(AFS_AIX41_ENV)
	    AFS_GUNLOCK();
	    code =
		VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL,
			  NULL, afs_osi_credp);
	    AFS_GLOCK();
#elif defined(AFS_AIX32_ENV)
	    code =
		VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL);
	    /* Flush all JFS pages now for big performance gain in big file cases
	     * If we do something like this, must check to be sure that AFS file
	     * isn't mmapped... see afs_gn_map() for why.
	     */
/*
	  if (tfile->vnode->v_gnode && tfile->vnode->v_gnode->gn_seg) {
 many different ways to do similar things:
   so far, the best performing one is #2, but #1 might match it if we
   straighten out the confusion regarding which pages to flush.  It
   really does matter.
   1.	    vm_flushp(tfile->vnode->v_gnode->gn_seg, 0, len/PAGESIZE - 1);
   2.	    vm_releasep(tfile->vnode->v_gnode->gn_seg, offset/PAGESIZE,
			(len + PAGESIZE-1)/PAGESIZE);
   3.	    vms_inactive(tfile->vnode->v_gnode->gn_seg) Doesn't work correctly
   4.  	    vms_delete(tfile->vnode->v_gnode->gn_seg) probably also fails
	    tfile->vnode->v_gnode->gn_seg = NULL;
   5.       deletep
   6.       ipgrlse
   7.       ifreeseg
          Unfortunately, this seems to cause frequent "cache corruption" episodes.
   	    vm_releasep(tfile->vnode->v_gnode->gn_seg, offset/PAGESIZE,
			(len + PAGESIZE-1)/PAGESIZE);
	  }
*/
#elif defined(AFS_AIX_ENV)
	    code =
		VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, (off_t) & offset,
			  &tuio, NULL, NULL, -1);
#elif defined(AFS_SUN5_ENV)
	    AFS_GUNLOCK();
#ifdef AFS_SUN510_ENV
	    VOP_RWLOCK(tfile->vnode, 0, NULL);
	    code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp, NULL);
	    VOP_RWUNLOCK(tfile->vnode, 0, NULL);
#else
	    VOP_RWLOCK(tfile->vnode, 0);
	    code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
	    VOP_RWUNLOCK(tfile->vnode, 0);
#endif
	    AFS_GLOCK();
#elif defined(AFS_SGI_ENV)
	    AFS_GUNLOCK();
	    AFS_VOP_RWLOCK(tfile->vnode, VRWLOCK_READ);
	    AFS_VOP_READ(tfile->vnode, &tuio, IO_ISLOCKED, afs_osi_credp,
			 code);
	    AFS_VOP_RWUNLOCK(tfile->vnode, VRWLOCK_READ);
	    AFS_GLOCK();
#elif defined(AFS_HPUX100_ENV)
	    AFS_GUNLOCK();
	    code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, afs_osi_credp);
	    AFS_GLOCK();
#elif defined(AFS_LINUX20_ENV)
	    AFS_GUNLOCK();
	    code = osi_rdwr(tfile, &tuio, UIO_READ);
	    AFS_GLOCK();
#elif defined(AFS_DARWIN80_ENV)
	    AFS_GUNLOCK();
	    code = VNOP_READ(tfile->vnode, tuiop, 0, afs_osi_ctxtp);
	    AFS_GLOCK();
#elif defined(AFS_DARWIN_ENV)
	    AFS_GUNLOCK();
	    VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, current_proc());
	    code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
	    VOP_UNLOCK(tfile->vnode, 0, current_proc());
	    AFS_GLOCK();
#elif defined(AFS_FBSD80_ENV)
	    AFS_GUNLOCK();
	    VOP_LOCK(tfile->vnode, LK_EXCLUSIVE);
	    code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
	    VOP_UNLOCK(tfile->vnode, 0);
	    AFS_GLOCK();
#elif defined(AFS_FBSD_ENV)
	    AFS_GUNLOCK();
	    VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curthread);
	    code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
	    VOP_UNLOCK(tfile->vnode, 0, curthread);
	    AFS_GLOCK();
#elif defined(AFS_NBSD_ENV)
            AFS_GUNLOCK();
            VOP_LOCK(tfile->vnode, LK_EXCLUSIVE);
            code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
            VOP_UNLOCK(tfile->vnode, 0);
            AFS_GLOCK();

#elif defined(AFS_XBSD_ENV)
	    AFS_GUNLOCK();
	    VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curproc);
	    code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
	    VOP_UNLOCK(tfile->vnode, 0, curproc);
	    AFS_GLOCK();
#else
	    code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, afs_osi_credp);
#endif
	    osi_UFSClose(tfile);

	    if (code) {
		error = code;
		break;
	    }
	}
	/* otherwise we've read some, fixup length, etc and continue with next seg */
	len = len - AFS_UIO_RESID(tuiop);	/* compute amount really transferred */
	trimlen = len;
	afsio_skip(auio, trimlen);	/* update input uio structure */
	totalLength -= len;
	transferLength += len;
	filePos += len;
	if (len <= 0)
	    break;		/* surprise eof */
#ifdef AFS_DARWIN80_ENV
	if (tuiop) {
	    uio_free(tuiop);
	    tuiop = 0;
	}
#endif
    }

    /* if we make it here with tdc non-zero, then it is the last chunk we
     * dealt with, and we have to release it when we're done.  We hold on
     * to it in case we need to do a prefetch, obviously.
     */
    if (tdc) {
	ReleaseReadLock(&tdc->lock);
#if !defined(AFS_VM_RDWR_ENV)
	/*
         * try to queue prefetch, if needed. If DataVersion is zero there
	 * should not be any more: files with DV 0 never have been stored
	 * on the fileserver, symbolic links and directories never require
	 * more than a single chunk.
	 */
	if (!noLock && !(hiszero(avc->f.m.DataVersion))) {
	    if (!(tdc->mflags & DFNextStarted))
		afs_PrefetchChunk(avc, tdc, acred, treq);
	}
#endif
	afs_PutDCache(tdc);
    }
    if (!noLock)
	ReleaseReadLock(&avc->lock);

#ifdef AFS_DARWIN80_ENV
    if (tuiop)
       uio_free(tuiop);
#else
    osi_FreeSmallSpace(tvec);
#endif
    AFS_DISCON_UNLOCK();
    error = afs_CheckCode(error, treq, 13);
    afs_DestroyReq(treq);
    return error;
}
Beispiel #3
0
int
afs_StoreAllSegments(struct vcache *avc, struct vrequest *areq,
		     int sync)
{
    struct dcache *tdc;
    afs_int32 code = 0;
    afs_int32 index;
    afs_int32 origCBs, foreign = 0;
    int hash;
    afs_hyper_t newDV, oldDV;	/* DV when we start, and finish, respectively */
    struct dcache **dcList;
    unsigned int i, j, minj, moredata, high, off;
    afs_size_t maxStoredLength;	/* highest offset we've written to server. */
    int safety, marineronce = 0;

    AFS_STATCNT(afs_StoreAllSegments);

    hset(oldDV, avc->f.m.DataVersion);
    hset(newDV, avc->f.m.DataVersion);
    hash = DVHash(&avc->f.fid);
    foreign = (avc->f.states & CForeign);
    dcList = osi_AllocLargeSpace(AFS_LRALLOCSIZ);
    afs_Trace2(afs_iclSetp, CM_TRACE_STOREALL, ICL_TYPE_POINTER, avc,
	       ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length));
#if !defined(AFS_AIX32_ENV) && !defined(AFS_SGI65_ENV)
    /* In the aix vm implementation we need to do the vm_writep even
     * on the memcache case since that's we adjust the file's size
     * and finish flushing partial vm pages.
     */
    if ((cacheDiskType != AFS_FCACHE_TYPE_MEM) ||
	(sync & AFS_VMSYNC_INVAL) || (sync & AFS_VMSYNC) ||
	(sync & AFS_LASTSTORE))
#endif /* !AFS_AIX32_ENV && !AFS_SGI65_ENV */
    {
	/* If we're not diskless, reading a file may stress the VM
	 * system enough to cause a pageout, and this vnode would be
	 * locked when the pageout occurs.  We can prevent this problem
	 * by making sure all dirty pages are already flushed.  We don't
	 * do this when diskless because reading a diskless (i.e.
	 * memory-resident) chunk doesn't require using new VM, and we
	 * also don't want to dump more dirty data into a diskless cache,
	 * since they're smaller, and we might exceed its available
	 * space.
	 */
#if	defined(AFS_SUN5_ENV)
	if (sync & AFS_VMSYNC_INVAL)	/* invalidate VM pages */
	    osi_VM_TryToSmush(avc, CRED(), 1);
	else
#endif
	    osi_VM_StoreAllSegments(avc);
    }
    if (AFS_IS_DISCONNECTED && !AFS_IN_SYNC) {
	/* This will probably make someone sad ... */
	/*printf("Net down in afs_StoreSegments\n");*/
	return ENETDOWN;
    }
    ConvertWToSLock(&avc->lock);

    /*
     * Subsequent code expects a sorted list, and it expects all the
     * chunks in the list to be contiguous, so we need a sort and a
     * while loop in here, too - but this will work for a first pass...
     * 92.10.05 - OK, there's a sort in here now.  It's kind of a modified
     *            bin sort, I guess.  Chunk numbers start with 0
     *
     * - Have to get a write lock on xdcache because GetDSlot might need it (if
     *   the chunk doesn't have a dcache struct).
     *   This seems like overkill in most cases.
     * - I'm not sure that it's safe to do "index = .hvNextp", then unlock
     *   xdcache, then relock xdcache and try to use index.  It is done
     *   a lot elsewhere in the CM, but I'm not buying that argument.
     * - should be able to check IFDataMod without doing the GetDSlot (just
     *   hold afs_xdcache).  That way, it's easy to do this without the
     *   writelock on afs_xdcache, and we save unneccessary disk
     *   operations. I don't think that works, 'cuz the next pointers
     *   are still on disk.
     */
    origCBs = afs_allCBs;

    maxStoredLength = 0;
    minj = 0;

    do {
	memset(dcList, 0, NCHUNKSATONCE * sizeof(struct dcache *));
	high = 0;
	moredata = FALSE;

	/* lock and start over from beginning of hash chain
	 * in order to avoid a race condition. */
	ObtainWriteLock(&afs_xdcache, 284);
	index = afs_dvhashTbl[hash];

	for (j = 0; index != NULLIDX;) {
	    if ((afs_indexFlags[index] & IFDataMod)
		&& (afs_indexUnique[index] == avc->f.fid.Fid.Unique)) {
		tdc = afs_GetValidDSlot(index);	/* refcount+1. */
		if (!tdc) {
		    ReleaseWriteLock(&afs_xdcache);
		    code = EIO;
		    goto done;
		}
		ReleaseReadLock(&tdc->tlock);
		if (!FidCmp(&tdc->f.fid, &avc->f.fid) && tdc->f.chunk >= minj) {
		    off = tdc->f.chunk - minj;
		    if (off < NCHUNKSATONCE) {
			if (dcList[off])
			    osi_Panic("dclist slot already in use!");
			if (afs_mariner && !marineronce) {
			    /* first chunk only */
			    afs_MarinerLog("store$Storing", avc);
			    marineronce++;
			}
			dcList[off] = tdc;
			if (off > high)
			    high = off;
			j++;
			/* DCLOCKXXX: chunkBytes is protected by tdc->lock which we
			 * can't grab here, due to lock ordering with afs_xdcache.
			 * So, disable this shortcut for now.  -- kolya 2001-10-13
			 */
			/* shortcut: big win for little files */
			/* tlen -= tdc->f.chunkBytes;
			 * if (tlen <= 0)
			 *    break;
			 */
		    } else {
			moredata = TRUE;
			afs_PutDCache(tdc);
			if (j == NCHUNKSATONCE)
			    break;
		    }
		} else {
		    afs_PutDCache(tdc);
		}
	    }
	    index = afs_dvnextTbl[index];
	}
	ReleaseWriteLock(&afs_xdcache);

	/* this guy writes chunks, puts back dcache structs, and bumps newDV */
	/* "moredata" just says "there are more dirty chunks yet to come".
	 */
	if (j) {
	    code =
		afs_CacheStoreVCache(dcList, avc, areq, sync,
				   minj, high, moredata,
				   &newDV, &maxStoredLength);
	    /* Release any zero-length dcache entries in our interval
	     * that we locked but didn't store back above.
	     */
	    for (j = 0; j <= high; j++) {
		tdc = dcList[j];
		if (tdc) {
		    osi_Assert(tdc->f.chunkBytes == 0);
		    ReleaseSharedLock(&tdc->lock);
		    afs_PutDCache(tdc);
		}
	    }
	}
	/* if (j) */
	minj += NCHUNKSATONCE;
    } while (!code && moredata);

 done:
    UpgradeSToWLock(&avc->lock, 29);

    /* send a trivial truncation store if did nothing else */
    if (code == 0) {
	/*
	 * Call StoreMini if we haven't written enough data to extend the
	 * file at the fileserver to the client's notion of the file length.
	 */
	if ((avc->f.truncPos != AFS_NOTRUNC)
	    || ((avc->f.states & CExtendedFile)
		&& (maxStoredLength < avc->f.m.Length))) {
	    code = afs_StoreMini(avc, areq);
	    if (code == 0)
		hadd32(newDV, 1);	/* just bumped here, too */
	}
	avc->f.states &= ~CExtendedFile;
    }

    /*
     * Finally, turn off DWriting, turn on DFEntryMod,
     * update f.versionNo.
     * A lot of this could be integrated into the loop above
     */
    if (!code) {
	afs_hyper_t h_unset;
	hones(h_unset);

	minj = 0;

	do {
	    moredata = FALSE;
	    memset(dcList, 0,
		   NCHUNKSATONCE * sizeof(struct dcache *));

	    /* overkill, but it gets the lock in case GetDSlot needs it */
	    ObtainWriteLock(&afs_xdcache, 285);

	    for (j = 0, safety = 0, index = afs_dvhashTbl[hash];
		 index != NULLIDX && safety < afs_cacheFiles + 2;
	         index = afs_dvnextTbl[index]) {

		if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
		    tdc = afs_GetValidDSlot(index);
		    if (!tdc) {
			/* This is okay; since manipulating the dcaches at this
			 * point is best-effort. We only get a dcache here to
			 * increment the dv and turn off DWriting. If we were
			 * supposed to do that for a dcache, but could not
			 * due to an I/O error, it just means the dv won't
			 * be updated so we don't be able to use that cached
			 * chunk in the future. That's inefficient, but not
			 * an error. */
			continue;
		    }
		    ReleaseReadLock(&tdc->tlock);

		    if (!FidCmp(&tdc->f.fid, &avc->f.fid)
			&& tdc->f.chunk >= minj) {
			off = tdc->f.chunk - minj;
			if (off < NCHUNKSATONCE) {
			    /* this is the file, and the correct chunk range */
			    if (j >= NCHUNKSATONCE)
				osi_Panic
				    ("Too many dcache entries in range\n");
			    dcList[j++] = tdc;
			} else {
			    moredata = TRUE;
			    afs_PutDCache(tdc);
			    if (j == NCHUNKSATONCE)
				break;
			}
		    } else {
			afs_PutDCache(tdc);
		    }
		}
	    }
	    ReleaseWriteLock(&afs_xdcache);

	    for (i = 0; i < j; i++) {
		/* Iterate over the dcache entries we collected above */
		tdc = dcList[i];
		ObtainSharedLock(&tdc->lock, 677);

		/* was code here to clear IFDataMod, but it should only be done
		 * in storedcache and storealldcache.
		 */
		/* Only increase DV if we had up-to-date data to start with.
		 * Otherwise, we could be falsely upgrading an old chunk
		 * (that we never read) into one labelled with the current
		 * DV #.  Also note that we check that no intervening stores
		 * occurred, otherwise we might mislabel cache information
		 * for a chunk that we didn't store this time
		 */
		/* Don't update the version number if it's not yet set. */
		if (!hsame(tdc->f.versionNo, h_unset)
		    && hcmp(tdc->f.versionNo, oldDV) >= 0) {

		    if ((!(afs_dvhack || foreign)
			 && hsame(avc->f.m.DataVersion, newDV))
			|| ((afs_dvhack || foreign)
			    && (origCBs == afs_allCBs))) {
			/* no error, this is the DV */

			UpgradeSToWLock(&tdc->lock, 678);
			hset(tdc->f.versionNo, avc->f.m.DataVersion);
			tdc->dflags |= DFEntryMod;
			/* DWriting may not have gotten cleared above, if all
			 * we did was a StoreMini */
			tdc->f.states &= ~DWriting;
			ConvertWToSLock(&tdc->lock);
		    }
		}

		ReleaseSharedLock(&tdc->lock);
		afs_PutDCache(tdc);
	    }

	    minj += NCHUNKSATONCE;

	} while (moredata);
    }

    if (code) {
	/*
	 * Invalidate chunks after an error for ccores files since
	 * afs_inactive won't be called for these and they won't be
	 * invalidated. Also discard data if it's a permanent error from the
	 * fileserver.
	 */
	if (areq->permWriteError || (avc->f.states & CCore)) {
	    afs_InvalidateAllSegments(avc);
	}
    }
    afs_Trace3(afs_iclSetp, CM_TRACE_STOREALLDONE, ICL_TYPE_POINTER, avc,
	       ICL_TYPE_INT32, avc->f.m.Length, ICL_TYPE_INT32, code);
    /* would like a Trace5, but it doesn't exist... */
    afs_Trace3(afs_iclSetp, CM_TRACE_AVCLOCKER, ICL_TYPE_POINTER, avc,
	       ICL_TYPE_INT32, avc->lock.wait_states, ICL_TYPE_INT32,
	       avc->lock.excl_locked);
    afs_Trace4(afs_iclSetp, CM_TRACE_AVCLOCKEE, ICL_TYPE_POINTER, avc,
	       ICL_TYPE_INT32, avc->lock.wait_states, ICL_TYPE_INT32,
	       avc->lock.readers_reading, ICL_TYPE_INT32,
	       avc->lock.num_waiting);

    /*
     * Finally, if updated DataVersion matches newDV, we did all of the
     * stores.  If mapDV indicates that the page cache was flushed up
     * to when we started the store, then we can relabel them as flushed
     * as recently as newDV.
     * Turn off CDirty bit because the stored data is now in sync with server.
     */
    if (code == 0 && hcmp(avc->mapDV, oldDV) >= 0) {
	if ((!(afs_dvhack || foreign) && hsame(avc->f.m.DataVersion, newDV))
	    || ((afs_dvhack || foreign) && (origCBs == afs_allCBs))) {
	    hset(avc->mapDV, newDV);
	    avc->f.states &= ~CDirty;
	}
    }
    osi_FreeLargeSpace(dcList);

    /* If not the final write a temporary error is ok. */
    if (code && !areq->permWriteError && !(sync & AFS_LASTSTORE))
	code = 0;

    return code;

}				/*afs_StoreAllSegments (new 03/02/94) */
Beispiel #4
0
/*!
 * \brief Entry point for user-space AFSDB request handler.
 * Reads cell data from kerlenMsg and add new cell, or alias.
 * \param acellName Cell name. If a cell is found, it's name will be filled in here.
 * \param acellNameLen Cell name length.
 * \param kernelMsg Buffer containing data about host count, time out, and cell hosts ids.
 * \return 0 for success, < 0 for error.
 */
int
afs_AFSDBHandler(char *acellName, int acellNameLen, afs_int32 * kernelMsg)
{
    afs_int32 timeout, code;
    afs_int32 cellHosts[AFS_MAXCELLHOSTS];

    if (afsdb_handler_shutdown)
	return -2;
    afsdb_handler_running = 1;

    ObtainSharedLock(&afsdb_req_lock, 683);
    if (afsdb_req.pending) {
	int i, hostCount;

	UpgradeSToWLock(&afsdb_req_lock, 684);
	hostCount = kernelMsg[0];
	timeout = kernelMsg[1];
	if (timeout)
	    timeout += osi_Time();

	for (i = 0; i < AFS_MAXCELLHOSTS; i++) {
	    if (i >= hostCount)
		cellHosts[i] = 0;
	    else
		cellHosts[i] = kernelMsg[2 + i];
	}

	if (hostCount)
	    code = afs_NewCell(acellName, cellHosts, CNoSUID, NULL, 0, 0,
			       timeout);

	if (!hostCount || (code && code != EEXIST))
	    /* null out the cellname if the lookup failed */
	    afsdb_req.cellname = NULL;
	else
	    /* If we found an alias, create it */
	    if (afs_strcasecmp(afsdb_req.cellname, acellName))
		afs_NewCellAlias(afsdb_req.cellname, acellName);

	/* Request completed, wake up the relevant thread */
	afsdb_req.pending = 0;
	afsdb_req.complete = 1;
	afs_osi_Wakeup(&afsdb_req);
	ConvertWToSLock(&afsdb_req_lock);
    }
    ConvertSToRLock(&afsdb_req_lock);

    /* Wait for a request */
    while (afsdb_req.pending == 0 && afs_termState != AFSOP_STOP_AFSDB) {
	ReleaseReadLock(&afsdb_req_lock);
	afs_osi_Sleep(&afsdb_req);
	ObtainReadLock(&afsdb_req_lock);
    }

    /* Check if we're shutting down */
    if (afs_termState == AFSOP_STOP_AFSDB) {
	ReleaseReadLock(&afsdb_req_lock);

	/* Inform anyone waiting for us that we're going away */
	afsdb_handler_shutdown = 1;
	afsdb_handler_running = 0;
	afs_osi_Wakeup(&afsdb_req);

	afs_termState = AFSOP_STOP_RXEVENT;
	afs_osi_Wakeup(&afs_termState);
	return -2;
    }

    /* Return the lookup request to userspace */
    strncpy(acellName, afsdb_req.cellname, acellNameLen);
    ReleaseReadLock(&afsdb_req_lock);
    return 0;
}
Beispiel #5
0
/**
 * Connects to a server by it's server address.
 *
 * @param sap Server address.
 * @param aport Server port.
 * @param acell
 * @param tu Connect as this user.
 * @param force_if_down
 * @param create
 * @param locktype Specifies type of lock to be used for this function.
 *
 * @return The new connection.
 */
struct afs_conn *
afs_ConnBySA(struct srvAddr *sap, unsigned short aport, afs_int32 acell,
	     struct unixuser *tu, int force_if_down, afs_int32 create,
	     afs_int32 locktype, struct rx_connection **rxconn)
{
    int glocked, foundvec;
    struct afs_conn *tc = NULL;
    struct sa_conn_vector *tcv = NULL;
    struct rx_securityClass *csec; /*Security class object */
    int isec; /*Security index */
    int service;

    *rxconn = NULL;

    /* find cached connection */
    ObtainSharedLock(&afs_xconn, 15);
    foundvec = 0;
    for (tcv = sap->conns; tcv; tcv = tcv->next) {
        if (tcv->user == tu && tcv->port == aport) {
            /* return most eligible conn */
            if (!foundvec)
                foundvec = 1;
            UpgradeSToWLock(&afs_xconn, 37);
            tc = find_preferred_connection(tcv, create);
            ConvertWToSLock(&afs_xconn);
            break;
        }
    }

    if (!tc && !create) {
        /* Not found and can't create a new one. */
        ReleaseSharedLock(&afs_xconn);
        return NULL;
    }

    if (AFS_IS_DISCONNECTED && !AFS_IN_SYNC) {
        afs_warnuser("afs_ConnBySA: disconnected\n");
        ReleaseSharedLock(&afs_xconn);
        return NULL;
    }

    if (!foundvec && create) {
	/* No such connection vector exists.  Create one and splice it in.
	 * Make sure the server record has been marked as used (for the purposes
	 * of calculating up & down times, it's now considered to be an
	 * ``active'' server).  Also make sure the server's lastUpdateEvalTime
	 * gets set, marking the time of its ``birth''.
	 */
	UpgradeSToWLock(&afs_xconn, 37);
        new_conn_vector(tcv);

        tcv->user = tu;
        tcv->port = aport;
        tcv->srvr = sap;
        tcv->next = sap->conns;
        sap->conns = tcv;

        /* all struct afs_conn ptrs come from here */
        tc = find_preferred_connection(tcv, create);

	afs_ActivateServer(sap);

	ConvertWToSLock(&afs_xconn);
    } /* end of if (!tcv) */

    if (!tc) {
        /* Not found and no alternatives. */
        ReleaseSharedLock(&afs_xconn);
        return NULL;
    }

    if (tu->states & UTokensBad) {
	/* we may still have an authenticated RPC connection here,
	 * we'll have to create a new, unauthenticated, connection.
	 * Perhaps a better way to do this would be to set
	 * conn->forceConnectFS on all conns when the token first goes
	 * bad, but that's somewhat trickier, due to locking
	 * constraints (though not impossible).
	 */
	if (tc->id && (rx_SecurityClassOf(tc->id) != 0)) {
	    tc->forceConnectFS = 1;	/* force recreation of connection */
	}
	tu->states &= ~UHasTokens;      /* remove the authentication info */
    }

    glocked = ISAFS_GLOCK();
    if (tc->forceConnectFS) {
	UpgradeSToWLock(&afs_xconn, 38);
	csec = (struct rx_securityClass *)0;
	if (tc->id) {
	    if (glocked)
                AFS_GUNLOCK();
            rx_SetConnSecondsUntilNatPing(tc->id, 0);
            rx_DestroyConnection(tc->id);
	    if (glocked)
                AFS_GLOCK();
	}
	/*
	 * Stupid hack to determine if using vldb service or file system
	 * service.
	 */
	if (aport == sap->server->cell->vlport)
	    service = 52;
	else
	    service = 1;
	isec = 0;

	csec = afs_pickSecurityObject(tc, &isec);

	if (glocked)
            AFS_GUNLOCK();
	tc->id = rx_NewConnection(sap->sa_ip, aport, service, csec, isec);
	if (glocked)
            AFS_GLOCK();
	if (service == 52) {
	    rx_SetConnHardDeadTime(tc->id, afs_rx_harddead);
	}
	/* set to a RX_CALL_TIMEOUT error to allow MTU retry to trigger */
	rx_SetServerConnIdleDeadErr(tc->id, RX_CALL_DEAD);
	rx_SetConnIdleDeadTime(tc->id, afs_rx_idledead);

	/*
	 * Only do this for the base connection, not per-user.
	 * Will need to be revisited if/when CB gets security.
	 */
	if ((isec == 0) && (service != 52) && !(tu->states & UTokensBad) &&
	    (tu->viceId == UNDEFVID)
#ifndef UKERNEL /* ukernel runs as just one uid anyway */
	    && (tu->uid == 0)
#endif
	    )
	    rx_SetConnSecondsUntilNatPing(tc->id, 20);

	tc->forceConnectFS = 0;	/* apparently we're appropriately connected now */
	if (csec)
	    rxs_Release(csec);
	ConvertWToSLock(&afs_xconn);
    } /* end of if (tc->forceConnectFS)*/

    *rxconn = tc->id;
    rx_GetConnection(*rxconn);

    ReleaseSharedLock(&afs_xconn);
    return tc;
}
Beispiel #6
0
/**
 * Connects to a server by it's server address.
 *
 * @param sap Server address.
 * @param aport Server port.
 * @param acell
 * @param tu Connect as this user.
 * @param force_if_down
 * @param create
 * @param locktype Specifies type of lock to be used for this function.
 *
 * @return The new connection.
 */
struct afs_conn *
afs_ConnBySA(struct srvAddr *sap, unsigned short aport, afs_int32 acell,
	     struct unixuser *tu, int force_if_down, afs_int32 create,
	     afs_int32 locktype)
{
    struct afs_conn *tc = 0;
    struct rx_securityClass *csec;	/*Security class object */
    int isec;			/*Security index */
    int service;

    if (!sap || ((sap->sa_flags & SRVR_ISDOWN) && !force_if_down)) {
	/* sa is known down, and we don't want to force it.  */
	return NULL;
    }

    ObtainSharedLock(&afs_xconn, 15);
    /* Get conn by port and user. */
    for (tc = sap->conns; tc; tc = tc->next) {
	if (tc->user == tu && tc->port == aport) {
	    break;
	}
    }

    if (!tc && !create) {
	/* Not found and can't create a new one. */
	ReleaseSharedLock(&afs_xconn);
	return NULL;
    }
    
    if (AFS_IS_DISCONNECTED && !AFS_IN_SYNC) {
        afs_warnuser("afs_ConnBySA: disconnected\n");
        ReleaseSharedLock(&afs_xconn);
        return NULL;
    }

    if (!tc) {
	/* No such connection structure exists.  Create one and splice it in.
	 * Make sure the server record has been marked as used (for the purposes
	 * of calculating up & down times, it's now considered to be an
	 * ``active'' server).  Also make sure the server's lastUpdateEvalTime
	 * gets set, marking the time of its ``birth''.
	 */
	UpgradeSToWLock(&afs_xconn, 37);
	tc = (struct afs_conn *)afs_osi_Alloc(sizeof(struct afs_conn));
	memset(tc, 0, sizeof(struct afs_conn));

	tc->user = tu;
	tc->port = aport;
	tc->srvr = sap;
	tc->refCount = 0;	/* bumped below */
	tc->forceConnectFS = 1;
	tc->id = (struct rx_connection *)0;
	tc->next = sap->conns;
	sap->conns = tc;
	afs_ActivateServer(sap);

	ConvertWToSLock(&afs_xconn);
    } /* end of if (!tc) */
    tc->refCount++;

    if (tu->states & UTokensBad) {
	/* we may still have an authenticated RPC connection here,
	 * we'll have to create a new, unauthenticated, connection.
	 * Perhaps a better way to do this would be to set
	 * conn->forceConnectFS on all conns when the token first goes
	 * bad, but that's somewhat trickier, due to locking
	 * constraints (though not impossible).
	 */
	if (tc->id && (rx_SecurityClassOf(tc->id) != 0)) {
	    tc->forceConnectFS = 1;	/* force recreation of connection */
	}
	tu->vid = UNDEFVID;	/* forcibly disconnect the authentication info */
    }

    if (tc->forceConnectFS) {
	UpgradeSToWLock(&afs_xconn, 38);
	csec = (struct rx_securityClass *)0;
	if (tc->id) {
	    AFS_GUNLOCK();
	    rx_DestroyConnection(tc->id);
	    AFS_GLOCK();
	}
	/*
	 * Stupid hack to determine if using vldb service or file system
	 * service.
	 */
	if (aport == sap->server->cell->vlport)
	    service = 52;
	else
	    service = 1;
	isec = 0;

	csec = afs_pickSecurityObject(tc, &isec);

	AFS_GUNLOCK();
	tc->id = rx_NewConnection(sap->sa_ip, aport, service, csec, isec);
	AFS_GLOCK();
	if (service == 52) {
	    rx_SetConnHardDeadTime(tc->id, afs_rx_harddead);
	}
	/* set to a RX_CALL_TIMEOUT error to allow MTU retry to trigger */
	rx_SetServerConnIdleDeadErr(tc->id, RX_CALL_DEAD);
	rx_SetConnIdleDeadTime(tc->id, afs_rx_idledead);
	rx_SetMsgsizeRetryErr(tc->id, RX_MSGSIZE);

	/*
	 * Only do this for the base connection, not per-user.
	 * Will need to be revisited if/when CB gets security.
	 */
	if ((isec == 0) && (service != 52) && !(tu->states & UTokensBad) &&
	    (tu->vid == UNDEFVID))
	    rx_SetConnSecondsUntilNatPing(tc->id, 20);

	tc->forceConnectFS = 0;	/* apparently we're appropriately connected now */
	if (csec)
	    rxs_Release(csec);
	ConvertWToSLock(&afs_xconn);
    } /* end of if (tc->forceConnectFS)*/

    ReleaseSharedLock(&afs_xconn);
    return tc;
}
int
afs_MemRead(register struct vcache *avc, struct uio *auio,
            struct AFS_UCRED *acred, daddr_t albn, struct buf **abpp,
            int noLock)
{
    afs_size_t totalLength;
    afs_size_t transferLength;
    afs_size_t filePos;
    afs_size_t offset, len, tlen;
    afs_int32 trimlen;
    struct dcache *tdc = 0;
    afs_int32 error, trybusy = 1;
#ifdef AFS_DARWIN80_ENV
    uio_t tuiop = NULL;
#else
    struct uio tuio;
    struct uio *tuiop = &tuio;
    struct iovec *tvec;
#endif
    afs_int32 code;
    struct vrequest treq;

    AFS_STATCNT(afs_MemRead);
    if (avc->vc_error)
        return EIO;

    /* check that we have the latest status info in the vnode cache */
    if ((code = afs_InitReq(&treq, acred)))
        return code;
    if (!noLock) {
        code = afs_VerifyVCache(avc, &treq);
        if (code) {
            code = afs_CheckCode(code, &treq, 8);	/* failed to get it */
            return code;
        }
    }
#ifndef	AFS_VM_RDWR_ENV
    if (AFS_NFSXLATORREQ(acred)) {
        if (!afs_AccessOK
                (avc, PRSFS_READ, &treq,
                 CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
            return afs_CheckCode(EACCES, &treq, 9);
        }
    }
#endif

#ifndef AFS_DARWIN80_ENV
    tvec = (struct iovec *)osi_AllocSmallSpace(sizeof(struct iovec));
#endif
    totalLength = AFS_UIO_RESID(auio);
    filePos = AFS_UIO_OFFSET(auio);
    afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
               ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(filePos), ICL_TYPE_INT32,
               totalLength, ICL_TYPE_OFFSET,
               ICL_HANDLE_OFFSET(avc->f.m.Length));
    error = 0;
    transferLength = 0;
    if (!noLock)
        ObtainReadLock(&avc->lock);
#if	defined(AFS_TEXT_ENV) && !defined(AFS_VM_RDWR_ENV)
    if (avc->flushDV.high == AFS_MAXDV && avc->flushDV.low == AFS_MAXDV) {
        hset(avc->flushDV, avc->f.m.DataVersion);
    }
#endif

    /*
     * Locks held:
     * avc->lock(R)
     */
    if (filePos >= avc->f.m.Length) {
        if (len > AFS_ZEROS)
            len = sizeof(afs_zeros);	/* and in 0 buffer */
        len = 0;
#ifdef AFS_DARWIN80_ENV
        trimlen = len;
        tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
        afsio_copy(auio, &tuio, tvec);
        trimlen = len;
        afsio_trim(&tuio, trimlen);
#endif
        AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
    }

    while (avc->f.m.Length > 0 && totalLength > 0) {
        /* read all of the cached info */
        if (filePos >= avc->f.m.Length)
            break;		/* all done */
        if (noLock) {
            if (tdc) {
                ReleaseReadLock(&tdc->lock);
                afs_PutDCache(tdc);
            }
            tdc = afs_FindDCache(avc, filePos);
            if (tdc) {
                ObtainReadLock(&tdc->lock);
                offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
                len = tdc->f.chunkBytes - offset;
            }
        } else {
            /* a tricky question: does the presence of the DFFetching flag
             * mean that we're fetching the latest version of the file?  No.
             * The server could update the file as soon as the fetch responsible
             * for the setting of the DFFetching flag completes.
             *
             * However, the presence of the DFFetching flag (visible under
             * a dcache read lock since it is set and cleared only under a
             * dcache write lock) means that we're fetching as good a version
             * as was known to this client at the time of the last call to
             * afs_VerifyVCache, since the latter updates the stat cache's
             * m.DataVersion field under a vcache write lock, and from the
             * time that the DFFetching flag goes on in afs_GetDCache (before
             * the fetch starts), to the time it goes off (after the fetch
             * completes), afs_GetDCache keeps at least a read lock on the
             * vcache entry.
             *
             * This means that if the DFFetching flag is set, we can use that
             * data for any reads that must come from the current version of
             * the file (current == m.DataVersion).
             *
             * Another way of looking at this same point is this: if we're
             * fetching some data and then try do an afs_VerifyVCache, the
             * VerifyVCache operation will not complete until after the
             * DFFetching flag is turned off and the dcache entry's f.versionNo
             * field is updated.
             *
             * Note, by the way, that if DFFetching is set,
             * m.DataVersion > f.versionNo (the latter is not updated until
             * after the fetch completes).
             */
            if (tdc) {
                ReleaseReadLock(&tdc->lock);
                afs_PutDCache(tdc);	/* before reusing tdc */
            }
            tdc = afs_GetDCache(avc, filePos, &treq, &offset, &len, 2);
            ObtainReadLock(&tdc->lock);
            /* now, first try to start transfer, if we'll need the data.  If
             * data already coming, we don't need to do this, obviously.  Type
             * 2 requests never return a null dcache entry, btw.
             */
            if (!(tdc->dflags & DFFetching)
                    && !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
                /* have cache entry, it is not coming in now,
                 * and we'll need new data */
tagain:
                if (trybusy && !afs_BBusy()) {
                    struct brequest *bp;
                    /* daemon is not busy */
                    ObtainSharedLock(&tdc->mflock, 665);
                    if (!(tdc->mflags & DFFetchReq)) {
                        /* start the daemon (may already be running, however) */
                        UpgradeSToWLock(&tdc->mflock, 666);
                        tdc->mflags |= DFFetchReq;
                        bp = afs_BQueue(BOP_FETCH, avc, B_DONTWAIT, 0, acred,
                                        (afs_size_t) filePos, (afs_size_t) 0,
                                        tdc);
                        if (!bp) {
                            tdc->mflags &= ~DFFetchReq;
                            trybusy = 0;	/* Avoid bkg daemon since they're too busy */
                            ReleaseWriteLock(&tdc->mflock);
                            goto tagain;
                        }
                        ConvertWToSLock(&tdc->mflock);
                        /* don't use bp pointer! */
                    }
                    code = 0;
                    ConvertSToRLock(&tdc->mflock);
                    while (!code && tdc->mflags & DFFetchReq) {
                        afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT,
                                   ICL_TYPE_STRING, __FILE__, ICL_TYPE_INT32,
                                   __LINE__, ICL_TYPE_POINTER, tdc,
                                   ICL_TYPE_INT32, tdc->dflags);
                        /* don't need waiting flag on this one */
                        ReleaseReadLock(&tdc->mflock);
                        ReleaseReadLock(&tdc->lock);
                        ReleaseReadLock(&avc->lock);
                        code = afs_osi_SleepSig(&tdc->validPos);
                        ObtainReadLock(&avc->lock);
                        ObtainReadLock(&tdc->lock);
                        ObtainReadLock(&tdc->mflock);
                    }
                    ReleaseReadLock(&tdc->mflock);
                    if (code) {
                        error = code;
                        break;
                    }
                }
            }
            /* now data may have started flowing in (if DFFetching is on).  If
             * data is now streaming in, then wait for some interesting stuff.
             */
            code = 0;
            while (!code && (tdc->dflags & DFFetching)
                    && tdc->validPos <= filePos) {
                /* too early: wait for DFFetching flag to vanish,
                 * or data to appear */
                afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
                           __FILE__, ICL_TYPE_INT32, __LINE__,
                           ICL_TYPE_POINTER, tdc, ICL_TYPE_INT32,
                           tdc->dflags);
                ReleaseReadLock(&tdc->lock);
                ReleaseReadLock(&avc->lock);
                code = afs_osi_SleepSig(&tdc->validPos);
                ObtainReadLock(&avc->lock);
                ObtainReadLock(&tdc->lock);
            }
            if (code) {
                error = code;
                break;
            }
            /* fetching flag gone, data is here, or we never tried
             * (BBusy for instance) */
            if (tdc->dflags & DFFetching) {
                /* still fetching, some new data is here:
                 * compute length and offset */
                offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
                len = tdc->validPos - filePos;
            } else {
                /* no longer fetching, verify data version
                 * (avoid new GetDCache call) */
                if (hsame(avc->f.m.DataVersion, tdc->f.versionNo)
                        && ((len = tdc->validPos - filePos) > 0)) {
                    offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
                } else {
                    /* don't have current data, so get it below */
                    afs_Trace3(afs_iclSetp, CM_TRACE_VERSIONNO,
                               ICL_TYPE_INT64, ICL_HANDLE_OFFSET(filePos),
                               ICL_TYPE_HYPER, &avc->f.m.DataVersion,
                               ICL_TYPE_HYPER, &tdc->f.versionNo);
                    ReleaseReadLock(&tdc->lock);
                    afs_PutDCache(tdc);
                    tdc = NULL;
                }
            }

            if (!tdc) {
                /* If we get, it was not possible to start the
                 * background daemon. With flag == 1 afs_GetDCache
                 * does the FetchData rpc synchronously.
                 */
                ReleaseReadLock(&avc->lock);
                tdc = afs_GetDCache(avc, filePos, &treq, &offset, &len, 1);
                ObtainReadLock(&avc->lock);
                if (tdc)
                    ObtainReadLock(&tdc->lock);
            }
        }

        afs_Trace3(afs_iclSetp, CM_TRACE_VNODEREAD, ICL_TYPE_POINTER, tdc,
                   ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(offset),
                   ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(len));
        if (!tdc) {
            error = EIO;
            break;
        }

        /*
         * Locks held:
         * avc->lock(R)
         * tdc->lock(R)
         */

        if (len > totalLength)
            len = totalLength;	/* will read len bytes */
        if (len <= 0) {		/* shouldn't get here if DFFetching is on */
            /* read past the end of a chunk, may not be at next chunk yet, and yet
             * also not at eof, so may have to supply fake zeros */
            len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset;	/* bytes left in chunk addr space */
            if (len > totalLength)
                len = totalLength;	/* and still within xfr request */
            tlen = avc->f.m.Length - offset;	/* and still within file */
            if (len > tlen)
                len = tlen;
            if (len > AFS_ZEROS)
                len = sizeof(afs_zeros);	/* and in 0 buffer */
#ifdef AFS_DARWIN80_ENV
            trimlen = len;
            tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
            afsio_copy(auio, &tuio, tvec);
            trimlen = len;
            afsio_trim(&tuio, trimlen);
#endif
            AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
            if (code) {
                error = code;
                break;
            }
        } else {
            /* get the data from the mem cache */

            /* mung uio structure to be right for this transfer */
#ifdef AFS_DARWIN80_ENV
            trimlen = len;
            tuiop = afsio_darwin_partialcopy(auio, trimlen);
            uio_setoffset(tuiop, offset);
#else
            afsio_copy(auio, &tuio, tvec);
            trimlen = len;
            afsio_trim(&tuio, trimlen);
            tuio.afsio_offset = offset;
#endif

            code = afs_MemReadUIO(tdc->f.inode, tuiop);

            if (code) {
                error = code;
                break;
            }
        }
        /* otherwise we've read some, fixup length, etc and continue with next seg */
        len = len - AFS_UIO_RESID(tuiop);	/* compute amount really transferred */
        trimlen = len;
        afsio_skip(auio, trimlen);	/* update input uio structure */
        totalLength -= len;
        transferLength += len;
        filePos += len;

        if (len <= 0)
            break;		/* surprise eof */
#ifdef AFS_DARWIN80_ENV
        if (tuiop) {
            uio_free(tuiop);
            tuiop = 0;
        }
#endif
    }				/* the whole while loop */

    /*
     * Locks held:
     * avc->lock(R)
     * tdc->lock(R) if tdc
     */

    /* if we make it here with tdc non-zero, then it is the last chunk we
     * dealt with, and we have to release it when we're done.  We hold on
     * to it in case we need to do a prefetch.
     */
    if (tdc) {
        ReleaseReadLock(&tdc->lock);
        /* try to queue prefetch, if needed */
        if (!noLock &&
#ifndef AFS_VM_RDWR_ENV
                afs_preCache
#else
                1
#endif
           ) {
            afs_PrefetchChunk(avc, tdc, acred, &treq);
        }
        afs_PutDCache(tdc);
    }
    if (!noLock)
        ReleaseReadLock(&avc->lock);
#ifdef AFS_DARWIN80_ENV
    if (tuiop)
        uio_free(tuiop);
#else
    osi_FreeSmallSpace(tvec);
#endif
    error = afs_CheckCode(error, &treq, 10);
    return error;
}
int
afs_UFSRead(register struct vcache *avc, struct uio *auio,
            struct AFS_UCRED *acred, daddr_t albn, struct buf **abpp,
            int noLock)
{
    afs_size_t totalLength;
    afs_size_t transferLength;
    afs_size_t filePos;
    afs_size_t offset, len, tlen;
    afs_int32 trimlen;
    struct dcache *tdc = 0;
    afs_int32 error;
#ifdef AFS_DARWIN80_ENV
    uio_t tuiop=NULL;
#else
    struct uio tuio;
    struct uio *tuiop = &tuio;
    struct iovec *tvec;
#endif
    struct osi_file *tfile;
    afs_int32 code;
    int trybusy = 1;
    struct vrequest treq;

    AFS_STATCNT(afs_UFSRead);
    if (avc && avc->vc_error)
        return EIO;

    AFS_DISCON_LOCK();

    /* check that we have the latest status info in the vnode cache */
    if ((code = afs_InitReq(&treq, acred)))
        return code;
    if (!noLock) {
        if (!avc)
            osi_Panic("null avc in afs_UFSRead");
        else {
            code = afs_VerifyVCache(avc, &treq);
            if (code) {
                code = afs_CheckCode(code, &treq, 11);	/* failed to get it */
                AFS_DISCON_UNLOCK();
                return code;
            }
        }
    }
#ifndef	AFS_VM_RDWR_ENV
    if (AFS_NFSXLATORREQ(acred)) {
        if (!afs_AccessOK
                (avc, PRSFS_READ, &treq,
                 CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
            AFS_DISCON_UNLOCK();
            return afs_CheckCode(EACCES, &treq, 12);
        }
    }
#endif

#ifndef AFS_DARWIN80_ENV
    tvec = (struct iovec *)osi_AllocSmallSpace(sizeof(struct iovec));
#endif
    totalLength = AFS_UIO_RESID(auio);
    filePos = AFS_UIO_OFFSET(auio);
    afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
               ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(filePos), ICL_TYPE_INT32,
               totalLength, ICL_TYPE_OFFSET,
               ICL_HANDLE_OFFSET(avc->f.m.Length));
    error = 0;
    transferLength = 0;
    if (!noLock)
        ObtainReadLock(&avc->lock);
#if	defined(AFS_TEXT_ENV) && !defined(AFS_VM_RDWR_ENV)
    if (avc->flushDV.high == AFS_MAXDV && avc->flushDV.low == AFS_MAXDV) {
        hset(avc->flushDV, avc->f.m.DataVersion);
    }
#endif

    if (filePos >= avc->f.m.Length) {
        if (len > AFS_ZEROS)
            len = sizeof(afs_zeros);	/* and in 0 buffer */
        len = 0;
#ifdef AFS_DARWIN80_ENV
        trimlen = len;
        tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
        afsio_copy(auio, &tuio, tvec);
        trimlen = len;
        afsio_trim(&tuio, trimlen);
#endif
        AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
    }

    while (avc->f.m.Length > 0 && totalLength > 0) {
        /* read all of the cached info */
        if (filePos >= avc->f.m.Length)
            break;		/* all done */
        if (noLock) {
            if (tdc) {
                ReleaseReadLock(&tdc->lock);
                afs_PutDCache(tdc);
            }
            tdc = afs_FindDCache(avc, filePos);
            if (tdc) {
                ObtainReadLock(&tdc->lock);
                offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
                len = tdc->validPos - filePos;
            }
        } else {
            /* a tricky question: does the presence of the DFFetching flag
             * mean that we're fetching the latest version of the file?  No.
             * The server could update the file as soon as the fetch responsible
             * for the setting of the DFFetching flag completes.
             *
             * However, the presence of the DFFetching flag (visible under
             * a dcache read lock since it is set and cleared only under a
             * dcache write lock) means that we're fetching as good a version
             * as was known to this client at the time of the last call to
             * afs_VerifyVCache, since the latter updates the stat cache's
             * m.DataVersion field under a vcache write lock, and from the
             * time that the DFFetching flag goes on in afs_GetDCache (before
             * the fetch starts), to the time it goes off (after the fetch
             * completes), afs_GetDCache keeps at least a read lock on the
             * vcache entry.
             *
             * This means that if the DFFetching flag is set, we can use that
             * data for any reads that must come from the current version of
             * the file (current == m.DataVersion).
             *
             * Another way of looking at this same point is this: if we're
             * fetching some data and then try do an afs_VerifyVCache, the
             * VerifyVCache operation will not complete until after the
             * DFFetching flag is turned off and the dcache entry's f.versionNo
             * field is updated.
             *
             * Note, by the way, that if DFFetching is set,
             * m.DataVersion > f.versionNo (the latter is not updated until
             * after the fetch completes).
             */
            if (tdc) {
                ReleaseReadLock(&tdc->lock);
                afs_PutDCache(tdc);	/* before reusing tdc */
            }
            tdc = afs_GetDCache(avc, filePos, &treq, &offset, &len, 2);
#ifdef AFS_DISCON_ENV
            if (!tdc) {
                printf("Network down in afs_read");
                error = ENETDOWN;
                break;
            }
#endif /* AFS_DISCON_ENV */

            ObtainReadLock(&tdc->lock);
            /* now, first try to start transfer, if we'll need the data.  If
             * data already coming, we don't need to do this, obviously.  Type
             * 2 requests never return a null dcache entry, btw. */
            if (!(tdc->dflags & DFFetching)
                    && !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
                /* have cache entry, it is not coming in now, and we'll need new data */
tagain:
                if (trybusy && !afs_BBusy()) {
                    struct brequest *bp;
                    /* daemon is not busy */
                    ObtainSharedLock(&tdc->mflock, 667);
                    if (!(tdc->mflags & DFFetchReq)) {
                        UpgradeSToWLock(&tdc->mflock, 668);
                        tdc->mflags |= DFFetchReq;
                        bp = afs_BQueue(BOP_FETCH, avc, B_DONTWAIT, 0, acred,
                                        (afs_size_t) filePos, (afs_size_t) 0,
                                        tdc);
                        if (!bp) {
                            /* Bkg table full; retry deadlocks */
                            tdc->mflags &= ~DFFetchReq;
                            trybusy = 0;	/* Avoid bkg daemon since they're too busy */
                            ReleaseWriteLock(&tdc->mflock);
                            goto tagain;
                        }
                        ConvertWToSLock(&tdc->mflock);
                    }
                    code = 0;
                    ConvertSToRLock(&tdc->mflock);
                    while (!code && tdc->mflags & DFFetchReq) {
                        afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT,
                                   ICL_TYPE_STRING, __FILE__, ICL_TYPE_INT32,
                                   __LINE__, ICL_TYPE_POINTER, tdc,
                                   ICL_TYPE_INT32, tdc->dflags);
                        /* don't need waiting flag on this one */
                        ReleaseReadLock(&tdc->mflock);
                        ReleaseReadLock(&tdc->lock);
                        ReleaseReadLock(&avc->lock);
                        code = afs_osi_SleepSig(&tdc->validPos);
                        ObtainReadLock(&avc->lock);
                        ObtainReadLock(&tdc->lock);
                        ObtainReadLock(&tdc->mflock);
                    }
                    ReleaseReadLock(&tdc->mflock);
                    if (code) {
                        error = code;
                        break;
                    }
                }
            }
            /* now data may have started flowing in (if DFFetching is on).  If
             * data is now streaming in, then wait for some interesting stuff.
             */
            code = 0;
            while (!code && (tdc->dflags & DFFetching)
                    && tdc->validPos <= filePos) {
                /* too early: wait for DFFetching flag to vanish,
                 * or data to appear */
                afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
                           __FILE__, ICL_TYPE_INT32, __LINE__,
                           ICL_TYPE_POINTER, tdc, ICL_TYPE_INT32,
                           tdc->dflags);
                ReleaseReadLock(&tdc->lock);
                ReleaseReadLock(&avc->lock);
                code = afs_osi_SleepSig(&tdc->validPos);
                ObtainReadLock(&avc->lock);
                ObtainReadLock(&tdc->lock);
            }
            if (code) {
                error = code;
                break;
            }
            /* fetching flag gone, data is here, or we never tried
             * (BBusy for instance) */
            if (tdc->dflags & DFFetching) {
                /* still fetching, some new data is here:
                 * compute length and offset */
                offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
                len = tdc->validPos - filePos;
            } else {
                /* no longer fetching, verify data version (avoid new
                 * GetDCache call) */
                if (hsame(avc->f.m.DataVersion, tdc->f.versionNo)
                        && ((len = tdc->validPos - filePos) > 0)) {
                    offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
                } else {
                    /* don't have current data, so get it below */
                    afs_Trace3(afs_iclSetp, CM_TRACE_VERSIONNO,
                               ICL_TYPE_INT64, ICL_HANDLE_OFFSET(filePos),
                               ICL_TYPE_HYPER, &avc->f.m.DataVersion,
                               ICL_TYPE_HYPER, &tdc->f.versionNo);
                    ReleaseReadLock(&tdc->lock);
                    afs_PutDCache(tdc);
                    tdc = NULL;
                }
            }

            if (!tdc) {
                /* If we get, it was not possible to start the
                 * background daemon. With flag == 1 afs_GetDCache
                 * does the FetchData rpc synchronously.
                 */
                ReleaseReadLock(&avc->lock);
                tdc = afs_GetDCache(avc, filePos, &treq, &offset, &len, 1);
                ObtainReadLock(&avc->lock);
                if (tdc)
                    ObtainReadLock(&tdc->lock);
            }
        }

        if (!tdc) {
            error = EIO;
            break;
        }
        len = tdc->validPos - filePos;
        afs_Trace3(afs_iclSetp, CM_TRACE_VNODEREAD, ICL_TYPE_POINTER, tdc,
                   ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(offset),
                   ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(len));
        if (len > totalLength)
            len = totalLength;	/* will read len bytes */
        if (len <= 0) {		/* shouldn't get here if DFFetching is on */
            afs_Trace4(afs_iclSetp, CM_TRACE_VNODEREAD2, ICL_TYPE_POINTER,
                       tdc, ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(tdc->validPos),
                       ICL_TYPE_INT32, tdc->f.chunkBytes, ICL_TYPE_INT32,
                       tdc->dflags);
            /* read past the end of a chunk, may not be at next chunk yet, and yet
             * also not at eof, so may have to supply fake zeros */
            len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset;	/* bytes left in chunk addr space */
            if (len > totalLength)
                len = totalLength;	/* and still within xfr request */
            tlen = avc->f.m.Length - offset;	/* and still within file */
            if (len > tlen)
                len = tlen;
            if (len > AFS_ZEROS)
                len = sizeof(afs_zeros);	/* and in 0 buffer */
#ifdef AFS_DARWIN80_ENV
            trimlen = len;
            tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
            afsio_copy(auio, &tuio, tvec);
            trimlen = len;
            afsio_trim(&tuio, trimlen);
#endif
            AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
            if (code) {
                error = code;
                break;
            }
        } else {
            /* get the data from the file */
#ifdef IHINT
            if (tfile = tdc->ihint) {
                if (tdc->f.inode != tfile->inum) {
                    afs_warn("afs_UFSRead: %x hint mismatch tdc %d inum %d\n",
                             tdc, tdc->f.inode, tfile->inum);
                    osi_UFSClose(tfile);
                    tdc->ihint = tfile = 0;
                    nihints--;
                }
            }
            if (tfile != 0) {
                usedihint++;
            } else
#endif /* IHINT */
#if defined(LINUX_USE_FH)
                tfile = (struct osi_file *)osi_UFSOpen_fh(&tdc->f.fh, tdc->f.fh_type);
#else
                tfile = (struct osi_file *)osi_UFSOpen(tdc->f.inode);
#endif
#ifdef AFS_DARWIN80_ENV
            trimlen = len;
            tuiop = afsio_darwin_partialcopy(auio, trimlen);
            uio_setoffset(tuiop, offset);
#else
            /* mung uio structure to be right for this transfer */
            afsio_copy(auio, &tuio, tvec);
            trimlen = len;
            afsio_trim(&tuio, trimlen);
            tuio.afsio_offset = offset;
#endif

#if defined(AFS_AIX41_ENV)
            AFS_GUNLOCK();
            code =
                VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL,
                          NULL, afs_osi_credp);
            AFS_GLOCK();
#elif defined(AFS_AIX32_ENV)
            code =
                VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL);
            /* Flush all JFS pages now for big performance gain in big file cases
             * If we do something like this, must check to be sure that AFS file
             * isn't mmapped... see afs_gn_map() for why.
             */
            /*
            	  if (tfile->vnode->v_gnode && tfile->vnode->v_gnode->gn_seg) {
             many different ways to do similar things:
               so far, the best performing one is #2, but #1 might match it if we
               straighten out the confusion regarding which pages to flush.  It
               really does matter.
               1.	    vm_flushp(tfile->vnode->v_gnode->gn_seg, 0, len/PAGESIZE - 1);
               2.	    vm_releasep(tfile->vnode->v_gnode->gn_seg, offset/PAGESIZE,
            			(len + PAGESIZE-1)/PAGESIZE);
               3.	    vms_inactive(tfile->vnode->v_gnode->gn_seg) Doesn't work correctly
               4.  	    vms_delete(tfile->vnode->v_gnode->gn_seg) probably also fails
            	    tfile->vnode->v_gnode->gn_seg = NULL;
               5.       deletep
               6.       ipgrlse
               7.       ifreeseg
                      Unfortunately, this seems to cause frequent "cache corruption" episodes.
               	    vm_releasep(tfile->vnode->v_gnode->gn_seg, offset/PAGESIZE,
            			(len + PAGESIZE-1)/PAGESIZE);
            	  }
            */
#elif defined(AFS_AIX_ENV)
            code =
                VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, (off_t) & offset,
                          &tuio, NULL, NULL, -1);
#elif defined(AFS_SUN5_ENV)
            AFS_GUNLOCK();
#ifdef AFS_SUN510_ENV
            {
                caller_context_t ct;

                VOP_RWLOCK(tfile->vnode, 0, &ct);
                code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp, &ct);
                VOP_RWUNLOCK(tfile->vnode, 0, &ct);
            }
#else
            VOP_RWLOCK(tfile->vnode, 0);
            code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
            VOP_RWUNLOCK(tfile->vnode, 0);
#endif
            AFS_GLOCK();
#elif defined(AFS_SGI_ENV)
            AFS_GUNLOCK();
            AFS_VOP_RWLOCK(tfile->vnode, VRWLOCK_READ);
            AFS_VOP_READ(tfile->vnode, &tuio, IO_ISLOCKED, afs_osi_credp,
                         code);
            AFS_VOP_RWUNLOCK(tfile->vnode, VRWLOCK_READ);
            AFS_GLOCK();
#elif defined(AFS_OSF_ENV)
            tuio.uio_rw = UIO_READ;
            AFS_GUNLOCK();
            VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp, code);
            AFS_GLOCK();
#elif defined(AFS_HPUX100_ENV)
            AFS_GUNLOCK();
            code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, afs_osi_credp);
            AFS_GLOCK();
#elif defined(AFS_LINUX20_ENV)
            AFS_GUNLOCK();
            code = osi_rdwr(tfile, &tuio, UIO_READ);
            AFS_GLOCK();
#elif defined(AFS_DARWIN80_ENV)
            AFS_GUNLOCK();
            code = VNOP_READ(tfile->vnode, tuiop, 0, afs_osi_ctxtp);
            AFS_GLOCK();
#elif defined(AFS_DARWIN_ENV)
            AFS_GUNLOCK();
            VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, current_proc());
            code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
            VOP_UNLOCK(tfile->vnode, 0, current_proc());
            AFS_GLOCK();
#elif defined(AFS_FBSD80_ENV)
            AFS_GUNLOCK();
            VOP_LOCK(tfile->vnode, LK_EXCLUSIVE);
            code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
            VOP_UNLOCK(tfile->vnode, 0);
            AFS_GLOCK();
#elif defined(AFS_FBSD50_ENV)
            AFS_GUNLOCK();
            VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curthread);
            code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
            VOP_UNLOCK(tfile->vnode, 0, curthread);
            AFS_GLOCK();
#elif defined(AFS_XBSD_ENV)
            AFS_GUNLOCK();
            VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curproc);
            code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
            VOP_UNLOCK(tfile->vnode, 0, curproc);
            AFS_GLOCK();
#else
            code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, afs_osi_credp);
#endif

#ifdef IHINT
            if (!tdc->ihint && nihints < maxIHint) {
                tdc->ihint = tfile;
                nihints++;
            } else
#endif /* IHINT */
                osi_UFSClose(tfile);

            if (code) {
                error = code;
                break;
            }
        }
        /* otherwise we've read some, fixup length, etc and continue with next seg */
        len = len - AFS_UIO_RESID(tuiop);	/* compute amount really transferred */
        trimlen = len;
        afsio_skip(auio, trimlen);	/* update input uio structure */
        totalLength -= len;
        transferLength += len;
        filePos += len;
        if (len <= 0)
            break;		/* surprise eof */
#ifdef AFS_DARWIN80_ENV
        if (tuiop) {
            uio_free(tuiop);
            tuiop = 0;
        }
#endif
    }

    /* if we make it here with tdc non-zero, then it is the last chunk we
     * dealt with, and we have to release it when we're done.  We hold on
     * to it in case we need to do a prefetch, obviously.
     */
    if (tdc) {
        ReleaseReadLock(&tdc->lock);
#if !defined(AFS_VM_RDWR_ENV)
        /* try to queue prefetch, if needed */
        if (!noLock) {
            if (!(tdc->mflags & DFNextStarted))
                afs_PrefetchChunk(avc, tdc, acred, &treq);
        }
#endif
        afs_PutDCache(tdc);
    }
    if (!noLock)
        ReleaseReadLock(&avc->lock);

#ifdef AFS_DARWIN80_ENV
    if (tuiop)
        uio_free(tuiop);
#else
    osi_FreeSmallSpace(tvec);
#endif
    AFS_DISCON_UNLOCK();
    error = afs_CheckCode(error, &treq, 13);
    return error;
}