VMDK.cpp@ 58160

Last change on this file since 58160 was 58132, checked in by vboxsync, 9 years ago
*: Doxygen fixes.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 244.4 KB

Line
1	/* $Id: VMDK.cpp 58132 2015-10-09 00:09:37Z vboxsync $ */
2	/** @file
3	* VMDK disk image, core code.
4	*/
5
6	/*
7	* Copyright (C) 2006-2015 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18
19	/*********************************************************************************************************************************
20	* Header Files *
21	*********************************************************************************************************************************/
22	#define LOG_GROUP LOG_GROUP_VD_VMDK
23	#include <VBox/vd-plugin.h>
24	#include <VBox/err.h>
25
26	#include <VBox/log.h>
27	#include <iprt/assert.h>
28	#include <iprt/alloc.h>
29	#include <iprt/uuid.h>
30	#include <iprt/path.h>
31	#include <iprt/string.h>
32	#include <iprt/rand.h>
33	#include <iprt/zip.h>
34	#include <iprt/asm.h>
35
36	#include "VDBackends.h"
37
38
39	/*********************************************************************************************************************************
40	* Constants And Macros, Structures and Typedefs *
41	*********************************************************************************************************************************/
42
43	/** Maximum encoded string size (including NUL) we allow for VMDK images.
44	* Deliberately not set high to avoid running out of descriptor space. */
45	#define VMDK_ENCODED_COMMENT_MAX 1024
46
47	/** VMDK descriptor DDB entry for PCHS cylinders. */
48	#define VMDK_DDB_GEO_PCHS_CYLINDERS "ddb.geometry.cylinders"
49
50	/** VMDK descriptor DDB entry for PCHS heads. */
51	#define VMDK_DDB_GEO_PCHS_HEADS "ddb.geometry.heads"
52
53	/** VMDK descriptor DDB entry for PCHS sectors. */
54	#define VMDK_DDB_GEO_PCHS_SECTORS "ddb.geometry.sectors"
55
56	/** VMDK descriptor DDB entry for LCHS cylinders. */
57	#define VMDK_DDB_GEO_LCHS_CYLINDERS "ddb.geometry.biosCylinders"
58
59	/** VMDK descriptor DDB entry for LCHS heads. */
60	#define VMDK_DDB_GEO_LCHS_HEADS "ddb.geometry.biosHeads"
61
62	/** VMDK descriptor DDB entry for LCHS sectors. */
63	#define VMDK_DDB_GEO_LCHS_SECTORS "ddb.geometry.biosSectors"
64
65	/** VMDK descriptor DDB entry for image UUID. */
66	#define VMDK_DDB_IMAGE_UUID "ddb.uuid.image"
67
68	/** VMDK descriptor DDB entry for image modification UUID. */
69	#define VMDK_DDB_MODIFICATION_UUID "ddb.uuid.modification"
70
71	/** VMDK descriptor DDB entry for parent image UUID. */
72	#define VMDK_DDB_PARENT_UUID "ddb.uuid.parent"
73
74	/** VMDK descriptor DDB entry for parent image modification UUID. */
75	#define VMDK_DDB_PARENT_MODIFICATION_UUID "ddb.uuid.parentmodification"
76
77	/** No compression for streamOptimized files. */
78	#define VMDK_COMPRESSION_NONE 0
79
80	/** Deflate compression for streamOptimized files. */
81	#define VMDK_COMPRESSION_DEFLATE 1
82
83	/** Marker that the actual GD value is stored in the footer. */
84	#define VMDK_GD_AT_END 0xffffffffffffffffULL
85
86	/** Marker for end-of-stream in streamOptimized images. */
87	#define VMDK_MARKER_EOS 0
88
89	/** Marker for grain table block in streamOptimized images. */
90	#define VMDK_MARKER_GT 1
91
92	/** Marker for grain directory block in streamOptimized images. */
93	#define VMDK_MARKER_GD 2
94
95	/** Marker for footer in streamOptimized images. */
96	#define VMDK_MARKER_FOOTER 3
97
98	/** Marker for unknown purpose in streamOptimized images.
99	* Shows up in very recent images created by vSphere, but only sporadically.
100	* They "forgot" to document that one in the VMDK specification. */
101	#define VMDK_MARKER_UNSPECIFIED 4
102
103	/** Dummy marker for "don't check the marker value". */
104	#define VMDK_MARKER_IGNORE 0xffffffffU
105
106	/**
107	* Magic number for hosted images created by VMware Workstation 4, VMware
108	* Workstation 5, VMware Server or VMware Player. Not necessarily sparse.
109	*/
110	#define VMDK_SPARSE_MAGICNUMBER 0x564d444b /* 'V' 'M' 'D' 'K' */
111
112	/**
113	* VMDK hosted binary extent header. The "Sparse" is a total misnomer, as
114	* this header is also used for monolithic flat images.
115	*/
116	#pragma pack(1)
117	typedef struct SparseExtentHeader
118	{
119	uint32_t magicNumber;
120	uint32_t version;
121	uint32_t flags;
122	uint64_t capacity;
123	uint64_t grainSize;
124	uint64_t descriptorOffset;
125	uint64_t descriptorSize;
126	uint32_t numGTEsPerGT;
127	uint64_t rgdOffset;
128	uint64_t gdOffset;
129	uint64_t overHead;
130	bool uncleanShutdown;
131	char singleEndLineChar;
132	char nonEndLineChar;
133	char doubleEndLineChar1;
134	char doubleEndLineChar2;
135	uint16_t compressAlgorithm;
136	uint8_t pad[433];
137	} SparseExtentHeader;
138	#pragma pack()
139
140	/** VMDK capacity for a single chunk when 2G splitting is turned on. Should be
141	* divisible by the default grain size (64K) */
142	#define VMDK_2G_SPLIT_SIZE (2047 * 1024 * 1024)
143
144	/** VMDK streamOptimized file format marker. The type field may or may not
145	* be actually valid, but there's always data to read there. */
146	#pragma pack(1)
147	typedef struct VMDKMARKER
148	{
149	uint64_t uSector;
150	uint32_t cbSize;
151	uint32_t uType;
152	} VMDKMARKER, *PVMDKMARKER;
153	#pragma pack()
154
155
156	#ifdef VBOX_WITH_VMDK_ESX
157
158	/** @todo the ESX code is not tested, not used, and lacks error messages. */
159
160	/**
161	* Magic number for images created by VMware GSX Server 3 or ESX Server 3.
162	*/
163	#define VMDK_ESX_SPARSE_MAGICNUMBER 0x44574f43 /* 'C' 'O' 'W' 'D' */
164
165	#pragma pack(1)
166	typedef struct COWDisk_Header
167	{
168	uint32_t magicNumber;
169	uint32_t version;
170	uint32_t flags;
171	uint32_t numSectors;
172	uint32_t grainSize;
173	uint32_t gdOffset;
174	uint32_t numGDEntries;
175	uint32_t freeSector;
176	/* The spec incompletely documents quite a few further fields, but states
177	* that they are unused by the current format. Replace them by padding. */
178	char reserved1[1604];
179	uint32_t savedGeneration;
180	char reserved2[8];
181	uint32_t uncleanShutdown;
182	char padding[396];
183	} COWDisk_Header;
184	#pragma pack()
185	#endif /* VBOX_WITH_VMDK_ESX */
186
187
188	/** Convert sector number/size to byte offset/size. */
189	#define VMDK_SECTOR2BYTE(u) ((uint64_t)(u) << 9)
190
191	/** Convert byte offset/size to sector number/size. */
192	#define VMDK_BYTE2SECTOR(u) ((u) >> 9)
193
194	/**
195	* VMDK extent type.
196	*/
197	typedef enum VMDKETYPE
198	{
199	/** Hosted sparse extent. */
200	VMDKETYPE_HOSTED_SPARSE = 1,
201	/** Flat extent. */
202	VMDKETYPE_FLAT,
203	/** Zero extent. */
204	VMDKETYPE_ZERO,
205	/** VMFS extent, used by ESX. */
206	VMDKETYPE_VMFS
207	#ifdef VBOX_WITH_VMDK_ESX
208	,
209	/** ESX sparse extent. */
210	VMDKETYPE_ESX_SPARSE
211	#endif /* VBOX_WITH_VMDK_ESX */
212	} VMDKETYPE, *PVMDKETYPE;
213
214	/**
215	* VMDK access type for a extent.
216	*/
217	typedef enum VMDKACCESS
218	{
219	/** No access allowed. */
220	VMDKACCESS_NOACCESS = 0,
221	/** Read-only access. */
222	VMDKACCESS_READONLY,
223	/** Read-write access. */
224	VMDKACCESS_READWRITE
225	} VMDKACCESS, *PVMDKACCESS;
226
227	/** Forward declaration for PVMDKIMAGE. */
228	typedef struct VMDKIMAGE *PVMDKIMAGE;
229
230	/**
231	* Extents files entry. Used for opening a particular file only once.
232	*/
233	typedef struct VMDKFILE
234	{
235	/** Pointer to filename. Local copy. */
236	const char *pszFilename;
237	/** File open flags for consistency checking. */
238	unsigned fOpen;
239	/** Handle for sync/async file abstraction.*/
240	PVDIOSTORAGE pStorage;
241	/** Reference counter. */
242	unsigned uReferences;
243	/** Flag whether the file should be deleted on last close. */
244	bool fDelete;
245	/** Pointer to the image we belong to (for debugging purposes). */
246	PVMDKIMAGE pImage;
247	/** Pointer to next file descriptor. */
248	struct VMDKFILE *pNext;
249	/** Pointer to the previous file descriptor. */
250	struct VMDKFILE *pPrev;
251	} VMDKFILE, *PVMDKFILE;
252
253	/**
254	* VMDK extent data structure.
255	*/
256	typedef struct VMDKEXTENT
257	{
258	/** File handle. */
259	PVMDKFILE pFile;
260	/** Base name of the image extent. */
261	const char *pszBasename;
262	/** Full name of the image extent. */
263	const char *pszFullname;
264	/** Number of sectors in this extent. */
265	uint64_t cSectors;
266	/** Number of sectors per block (grain in VMDK speak). */
267	uint64_t cSectorsPerGrain;
268	/** Starting sector number of descriptor. */
269	uint64_t uDescriptorSector;
270	/** Size of descriptor in sectors. */
271	uint64_t cDescriptorSectors;
272	/** Starting sector number of grain directory. */
273	uint64_t uSectorGD;
274	/** Starting sector number of redundant grain directory. */
275	uint64_t uSectorRGD;
276	/** Total number of metadata sectors. */
277	uint64_t cOverheadSectors;
278	/** Nominal size (i.e. as described by the descriptor) of this extent. */
279	uint64_t cNominalSectors;
280	/** Sector offset (i.e. as described by the descriptor) of this extent. */
281	uint64_t uSectorOffset;
282	/** Number of entries in a grain table. */
283	uint32_t cGTEntries;
284	/** Number of sectors reachable via a grain directory entry. */
285	uint32_t cSectorsPerGDE;
286	/** Number of entries in the grain directory. */
287	uint32_t cGDEntries;
288	/** Pointer to the next free sector. Legacy information. Do not use. */
289	uint32_t uFreeSector;
290	/** Number of this extent in the list of images. */
291	uint32_t uExtent;
292	/** Pointer to the descriptor (NULL if no descriptor in this extent). */
293	char *pDescData;
294	/** Pointer to the grain directory. */
295	uint32_t *pGD;
296	/** Pointer to the redundant grain directory. */
297	uint32_t *pRGD;
298	/** VMDK version of this extent. 1=1.0/1.1 */
299	uint32_t uVersion;
300	/** Type of this extent. */
301	VMDKETYPE enmType;
302	/** Access to this extent. */
303	VMDKACCESS enmAccess;
304	/** Flag whether this extent is marked as unclean. */
305	bool fUncleanShutdown;
306	/** Flag whether the metadata in the extent header needs to be updated. */
307	bool fMetaDirty;
308	/** Flag whether there is a footer in this extent. */
309	bool fFooter;
310	/** Compression type for this extent. */
311	uint16_t uCompression;
312	/** Append position for writing new grain. Only for sparse extents. */
313	uint64_t uAppendPosition;
314	/** Last grain which was accessed. Only for streamOptimized extents. */
315	uint32_t uLastGrainAccess;
316	/** Starting sector corresponding to the grain buffer. */
317	uint32_t uGrainSectorAbs;
318	/** Grain number corresponding to the grain buffer. */
319	uint32_t uGrain;
320	/** Actual size of the compressed data, only valid for reading. */
321	uint32_t cbGrainStreamRead;
322	/** Size of compressed grain buffer for streamOptimized extents. */
323	size_t cbCompGrain;
324	/** Compressed grain buffer for streamOptimized extents, with marker. */
325	void *pvCompGrain;
326	/** Decompressed grain buffer for streamOptimized extents. */
327	void *pvGrain;
328	/** Reference to the image in which this extent is used. Do not use this
329	* on a regular basis to avoid passing pImage references to functions
330	* explicitly. */
331	struct VMDKIMAGE *pImage;
332	} VMDKEXTENT, *PVMDKEXTENT;
333
334	/**
335	* Grain table cache size. Allocated per image.
336	*/
337	#define VMDK_GT_CACHE_SIZE 256
338
339	/**
340	* Grain table block size. Smaller than an actual grain table block to allow
341	* more grain table blocks to be cached without having to allocate excessive
342	* amounts of memory for the cache.
343	*/
344	#define VMDK_GT_CACHELINE_SIZE 128
345
346
347	/**
348	* Maximum number of lines in a descriptor file. Not worth the effort of
349	* making it variable. Descriptor files are generally very short (~20 lines),
350	* with the exception of sparse files split in 2G chunks, which need for the
351	* maximum size (almost 2T) exactly 1025 lines for the disk database.
352	*/
353	#define VMDK_DESCRIPTOR_LINES_MAX 1100U
354
355	/**
356	* Parsed descriptor information. Allows easy access and update of the
357	* descriptor (whether separate file or not). Free form text files suck.
358	*/
359	typedef struct VMDKDESCRIPTOR
360	{
361	/** Line number of first entry of the disk descriptor. */
362	unsigned uFirstDesc;
363	/** Line number of first entry in the extent description. */
364	unsigned uFirstExtent;
365	/** Line number of first disk database entry. */
366	unsigned uFirstDDB;
367	/** Total number of lines. */
368	unsigned cLines;
369	/** Total amount of memory available for the descriptor. */
370	size_t cbDescAlloc;
371	/** Set if descriptor has been changed and not yet written to disk. */
372	bool fDirty;
373	/** Array of pointers to the data in the descriptor. */
374	char *aLines[VMDK_DESCRIPTOR_LINES_MAX];
375	/** Array of line indices pointing to the next non-comment line. */
376	unsigned aNextLines[VMDK_DESCRIPTOR_LINES_MAX];
377	} VMDKDESCRIPTOR, *PVMDKDESCRIPTOR;
378
379
380	/**
381	* Cache entry for translating extent/sector to a sector number in that
382	* extent.
383	*/
384	typedef struct VMDKGTCACHEENTRY
385	{
386	/** Extent number for which this entry is valid. */
387	uint32_t uExtent;
388	/** GT data block number. */
389	uint64_t uGTBlock;
390	/** Data part of the cache entry. */
391	uint32_t aGTData[VMDK_GT_CACHELINE_SIZE];
392	} VMDKGTCACHEENTRY, *PVMDKGTCACHEENTRY;
393
394	/**
395	* Cache data structure for blocks of grain table entries. For now this is a
396	* fixed size direct mapping cache, but this should be adapted to the size of
397	* the sparse image and maybe converted to a set-associative cache. The
398	* implementation below implements a write-through cache with write allocate.
399	*/
400	typedef struct VMDKGTCACHE
401	{
402	/** Cache entries. */
403	VMDKGTCACHEENTRY aGTCache[VMDK_GT_CACHE_SIZE];
404	/** Number of cache entries (currently unused). */
405	unsigned cEntries;
406	} VMDKGTCACHE, *PVMDKGTCACHE;
407
408	/**
409	* Complete VMDK image data structure. Mainly a collection of extents and a few
410	* extra global data fields.
411	*/
412	typedef struct VMDKIMAGE
413	{
414	/** Image name. */
415	const char *pszFilename;
416	/** Descriptor file if applicable. */
417	PVMDKFILE pFile;
418
419	/** Pointer to the per-disk VD interface list. */
420	PVDINTERFACE pVDIfsDisk;
421	/** Pointer to the per-image VD interface list. */
422	PVDINTERFACE pVDIfsImage;
423
424	/** Error interface. */
425	PVDINTERFACEERROR pIfError;
426	/** I/O interface. */
427	PVDINTERFACEIOINT pIfIo;
428
429
430	/** Pointer to the image extents. */
431	PVMDKEXTENT pExtents;
432	/** Number of image extents. */
433	unsigned cExtents;
434	/** Pointer to the files list, for opening a file referenced multiple
435	* times only once (happens mainly with raw partition access). */
436	PVMDKFILE pFiles;
437
438	/**
439	* Pointer to an array of segment entries for async I/O.
440	* This is an optimization because the task number to submit is not known
441	* and allocating/freeing an array in the read/write functions every time
442	* is too expensive.
443	*/
444	PPDMDATASEG paSegments;
445	/** Entries available in the segments array. */
446	unsigned cSegments;
447
448	/** Open flags passed by VBoxHD layer. */
449	unsigned uOpenFlags;
450	/** Image flags defined during creation or determined during open. */
451	unsigned uImageFlags;
452	/** Total size of the image. */
453	uint64_t cbSize;
454	/** Physical geometry of this image. */
455	VDGEOMETRY PCHSGeometry;
456	/** Logical geometry of this image. */
457	VDGEOMETRY LCHSGeometry;
458	/** Image UUID. */
459	RTUUID ImageUuid;
460	/** Image modification UUID. */
461	RTUUID ModificationUuid;
462	/** Parent image UUID. */
463	RTUUID ParentUuid;
464	/** Parent image modification UUID. */
465	RTUUID ParentModificationUuid;
466
467	/** Pointer to grain table cache, if this image contains sparse extents. */
468	PVMDKGTCACHE pGTCache;
469	/** Pointer to the descriptor (NULL if no separate descriptor file). */
470	char *pDescData;
471	/** Allocation size of the descriptor file. */
472	size_t cbDescAlloc;
473	/** Parsed descriptor file content. */
474	VMDKDESCRIPTOR Descriptor;
475	} VMDKIMAGE;
476
477
478	/** State for the input/output callout of the inflate reader/deflate writer. */
479	typedef struct VMDKCOMPRESSIO
480	{
481	/* Image this operation relates to. */
482	PVMDKIMAGE pImage;
483	/* Current read position. */
484	ssize_t iOffset;
485	/* Size of the compressed grain buffer (available data). */
486	size_t cbCompGrain;
487	/* Pointer to the compressed grain buffer. */
488	void *pvCompGrain;
489	} VMDKCOMPRESSIO;
490
491
492	/** Tracks async grain allocation. */
493	typedef struct VMDKGRAINALLOCASYNC
494	{
495	/** Flag whether the allocation failed. */
496	bool fIoErr;
497	/** Current number of transfers pending.
498	* If reached 0 and there is an error the old state is restored. */
499	unsigned cIoXfersPending;
500	/** Sector number */
501	uint64_t uSector;
502	/** Flag whether the grain table needs to be updated. */
503	bool fGTUpdateNeeded;
504	/** Extent the allocation happens. */
505	PVMDKEXTENT pExtent;
506	/** Position of the new grain, required for the grain table update. */
507	uint64_t uGrainOffset;
508	/** Grain table sector. */
509	uint64_t uGTSector;
510	/** Backup grain table sector. */
511	uint64_t uRGTSector;
512	} VMDKGRAINALLOCASYNC, *PVMDKGRAINALLOCASYNC;
513
514
515	/*********************************************************************************************************************************
516	* Static Variables *
517	*********************************************************************************************************************************/
518
519	/** NULL-terminated array of supported file extensions. */
520	static const VDFILEEXTENSION s_aVmdkFileExtensions[] =
521	{
522	{"vmdk", VDTYPE_HDD},
523	{NULL, VDTYPE_INVALID}
524	};
525
526
527	/*********************************************************************************************************************************
528	* Internal Functions *
529	*********************************************************************************************************************************/
530
531	static void vmdkFreeStreamBuffers(PVMDKEXTENT pExtent);
532	static int vmdkFreeExtentData(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
533	bool fDelete);
534
535	static int vmdkCreateExtents(PVMDKIMAGE pImage, unsigned cExtents);
536	static int vmdkFlushImage(PVMDKIMAGE pImage, PVDIOCTX pIoCtx);
537	static int vmdkSetImageComment(PVMDKIMAGE pImage, const char *pszComment);
538	static int vmdkFreeImage(PVMDKIMAGE pImage, bool fDelete);
539
540	static DECLCALLBACK(int) vmdkAllocGrainComplete(void *pBackendData, PVDIOCTX pIoCtx,
541	void *pvUser, int rcReq);
542
543	/**
544	* Internal: open a file (using a file descriptor cache to ensure each file
545	* is only opened once - anything else can cause locking problems).
546	*/
547	static int vmdkFileOpen(PVMDKIMAGE pImage, PVMDKFILE *ppVmdkFile,
548	const char *pszFilename, uint32_t fOpen)
549	{
550	int rc = VINF_SUCCESS;
551	PVMDKFILE pVmdkFile;
552
553	for (pVmdkFile = pImage->pFiles;
554	pVmdkFile != NULL;
555	pVmdkFile = pVmdkFile->pNext)
556	{
557	if (!strcmp(pszFilename, pVmdkFile->pszFilename))
558	{
559	Assert(fOpen == pVmdkFile->fOpen);
560	pVmdkFile->uReferences++;
561
562	*ppVmdkFile = pVmdkFile;
563
564	return rc;
565	}
566	}
567
568	/* If we get here, there's no matching entry in the cache. */
569	pVmdkFile = (PVMDKFILE)RTMemAllocZ(sizeof(VMDKFILE));
570	if (!pVmdkFile)
571	{
572	*ppVmdkFile = NULL;
573	return VERR_NO_MEMORY;
574	}
575
576	pVmdkFile->pszFilename = RTStrDup(pszFilename);
577	if (!pVmdkFile->pszFilename)
578	{
579	RTMemFree(pVmdkFile);
580	*ppVmdkFile = NULL;
581	return VERR_NO_MEMORY;
582	}
583	pVmdkFile->fOpen = fOpen;
584
585	rc = vdIfIoIntFileOpen(pImage->pIfIo, pszFilename, fOpen,
586	&pVmdkFile->pStorage);
587	if (RT_SUCCESS(rc))
588	{
589	pVmdkFile->uReferences = 1;
590	pVmdkFile->pImage = pImage;
591	pVmdkFile->pNext = pImage->pFiles;
592	if (pImage->pFiles)
593	pImage->pFiles->pPrev = pVmdkFile;
594	pImage->pFiles = pVmdkFile;
595	*ppVmdkFile = pVmdkFile;
596	}
597	else
598	{
599	RTStrFree((char )(void )pVmdkFile->pszFilename);
600	RTMemFree(pVmdkFile);
601	*ppVmdkFile = NULL;
602	}
603
604	return rc;
605	}
606
607	/**
608	* Internal: close a file, updating the file descriptor cache.
609	*/
610	static int vmdkFileClose(PVMDKIMAGE pImage, PVMDKFILE *ppVmdkFile, bool fDelete)
611	{
612	int rc = VINF_SUCCESS;
613	PVMDKFILE pVmdkFile = *ppVmdkFile;
614
615	AssertPtr(pVmdkFile);
616
617	pVmdkFile->fDelete \|= fDelete;
618	Assert(pVmdkFile->uReferences);
619	pVmdkFile->uReferences--;
620	if (pVmdkFile->uReferences == 0)
621	{
622	PVMDKFILE pPrev;
623	PVMDKFILE pNext;
624
625	/* Unchain the element from the list. */
626	pPrev = pVmdkFile->pPrev;
627	pNext = pVmdkFile->pNext;
628
629	if (pNext)
630	pNext->pPrev = pPrev;
631	if (pPrev)
632	pPrev->pNext = pNext;
633	else
634	pImage->pFiles = pNext;
635
636	rc = vdIfIoIntFileClose(pImage->pIfIo, pVmdkFile->pStorage);
637	if (RT_SUCCESS(rc) && pVmdkFile->fDelete)
638	rc = vdIfIoIntFileDelete(pImage->pIfIo, pVmdkFile->pszFilename);
639	RTStrFree((char )(void )pVmdkFile->pszFilename);
640	RTMemFree(pVmdkFile);
641	}
642
643	*ppVmdkFile = NULL;
644	return rc;
645	}
646
647	/#define VMDK_USE_BLOCK_DECOMP_API - test and enable /
648	#ifndef VMDK_USE_BLOCK_DECOMP_API
649	static DECLCALLBACK(int) vmdkFileInflateHelper(void pvUser, void pvBuf, size_t cbBuf, size_t *pcbBuf)
650	{
651	VMDKCOMPRESSIO pInflateState = (VMDKCOMPRESSIO )pvUser;
652	size_t cbInjected = 0;
653
654	Assert(cbBuf);
655	if (pInflateState->iOffset < 0)
656	{
657	(uint8_t )pvBuf = RTZIPTYPE_ZLIB;
658	pvBuf = (uint8_t *)pvBuf + 1;
659	cbBuf--;
660	cbInjected = 1;
661	pInflateState->iOffset = RT_OFFSETOF(VMDKMARKER, uType);
662	}
663	if (!cbBuf)
664	{
665	if (pcbBuf)
666	*pcbBuf = cbInjected;
667	return VINF_SUCCESS;
668	}
669	cbBuf = RT_MIN(cbBuf, pInflateState->cbCompGrain - pInflateState->iOffset);
670	memcpy(pvBuf,
671	(uint8_t *)pInflateState->pvCompGrain + pInflateState->iOffset,
672	cbBuf);
673	pInflateState->iOffset += cbBuf;
674	Assert(pcbBuf);
675	*pcbBuf = cbBuf + cbInjected;
676	return VINF_SUCCESS;
677	}
678	#endif
679
680	/**
681	* Internal: read from a file and inflate the compressed data,
682	* distinguishing between async and normal operation
683	*/
684	DECLINLINE(int) vmdkFileInflateSync(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
685	uint64_t uOffset, void *pvBuf,
686	size_t cbToRead, const void *pcvMarker,
687	uint64_t puLBA, uint32_t pcbMarkerData)
688	{
689	int rc;
690	#ifndef VMDK_USE_BLOCK_DECOMP_API
691	PRTZIPDECOMP pZip = NULL;
692	#endif
693	VMDKMARKER pMarker = (VMDKMARKER )pExtent->pvCompGrain;
694	size_t cbCompSize, cbActuallyRead;
695
696	if (!pcvMarker)
697	{
698	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
699	uOffset, pMarker, RT_OFFSETOF(VMDKMARKER, uType));
700	if (RT_FAILURE(rc))
701	return rc;
702	}
703	else
704	{
705	memcpy(pMarker, pcvMarker, RT_OFFSETOF(VMDKMARKER, uType));
706	/* pcvMarker endianness has already been partially transformed, fix it */
707	pMarker->uSector = RT_H2LE_U64(pMarker->uSector);
708	pMarker->cbSize = RT_H2LE_U32(pMarker->cbSize);
709	}
710
711	cbCompSize = RT_LE2H_U32(pMarker->cbSize);
712	if (cbCompSize == 0)
713	{
714	AssertMsgFailed(("VMDK: corrupted marker\n"));
715	return VERR_VD_VMDK_INVALID_FORMAT;
716	}
717
718	/* Sanity check - the expansion ratio should be much less than 2. */
719	Assert(cbCompSize < 2 * cbToRead);
720	if (cbCompSize >= 2 * cbToRead)
721	return VERR_VD_VMDK_INVALID_FORMAT;
722
723	/* Compressed grain marker. Data follows immediately. */
724	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
725	uOffset + RT_OFFSETOF(VMDKMARKER, uType),
726	(uint8_t *)pExtent->pvCompGrain
727	+ RT_OFFSETOF(VMDKMARKER, uType),
728	RT_ALIGN_Z( cbCompSize
729	+ RT_OFFSETOF(VMDKMARKER, uType),
730	512)
731	- RT_OFFSETOF(VMDKMARKER, uType));
732
733	if (puLBA)
734	*puLBA = RT_LE2H_U64(pMarker->uSector);
735	if (pcbMarkerData)
736	*pcbMarkerData = RT_ALIGN( cbCompSize
737	+ RT_OFFSETOF(VMDKMARKER, uType),
738	512);
739
740	#ifdef VMDK_USE_BLOCK_DECOMP_API
741	rc = RTZipBlockDecompress(RTZIPTYPE_ZLIB, 0 /fFlags/,
742	pExtent->pvCompGrain, cbCompSize + RT_OFFSETOF(VMDKMARKER, uType), NULL,
743	pvBuf, cbToRead, &cbActuallyRead);
744	#else
745	VMDKCOMPRESSIO InflateState;
746	InflateState.pImage = pImage;
747	InflateState.iOffset = -1;
748	InflateState.cbCompGrain = cbCompSize + RT_OFFSETOF(VMDKMARKER, uType);
749	InflateState.pvCompGrain = pExtent->pvCompGrain;
750
751	rc = RTZipDecompCreate(&pZip, &InflateState, vmdkFileInflateHelper);
752	if (RT_FAILURE(rc))
753	return rc;
754	rc = RTZipDecompress(pZip, pvBuf, cbToRead, &cbActuallyRead);
755	RTZipDecompDestroy(pZip);
756	#endif /* !VMDK_USE_BLOCK_DECOMP_API */
757	if (RT_FAILURE(rc))
758	{
759	if (rc == VERR_ZIP_CORRUPTED)
760	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: Compressed image is corrupted '%s'"), pExtent->pszFullname);
761	return rc;
762	}
763	if (cbActuallyRead != cbToRead)
764	rc = VERR_VD_VMDK_INVALID_FORMAT;
765	return rc;
766	}
767
768	static DECLCALLBACK(int) vmdkFileDeflateHelper(void pvUser, const void pvBuf, size_t cbBuf)
769	{
770	VMDKCOMPRESSIO pDeflateState = (VMDKCOMPRESSIO )pvUser;
771
772	Assert(cbBuf);
773	if (pDeflateState->iOffset < 0)
774	{
775	pvBuf = (const uint8_t *)pvBuf + 1;
776	cbBuf--;
777	pDeflateState->iOffset = RT_OFFSETOF(VMDKMARKER, uType);
778	}
779	if (!cbBuf)
780	return VINF_SUCCESS;
781	if (pDeflateState->iOffset + cbBuf > pDeflateState->cbCompGrain)
782	return VERR_BUFFER_OVERFLOW;
783	memcpy((uint8_t *)pDeflateState->pvCompGrain + pDeflateState->iOffset,
784	pvBuf, cbBuf);
785	pDeflateState->iOffset += cbBuf;
786	return VINF_SUCCESS;
787	}
788
789	/**
790	* Internal: deflate the uncompressed data and write to a file,
791	* distinguishing between async and normal operation
792	*/
793	DECLINLINE(int) vmdkFileDeflateSync(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
794	uint64_t uOffset, const void *pvBuf,
795	size_t cbToWrite, uint64_t uLBA,
796	uint32_t *pcbMarkerData)
797	{
798	int rc;
799	PRTZIPCOMP pZip = NULL;
800	VMDKCOMPRESSIO DeflateState;
801
802	DeflateState.pImage = pImage;
803	DeflateState.iOffset = -1;
804	DeflateState.cbCompGrain = pExtent->cbCompGrain;
805	DeflateState.pvCompGrain = pExtent->pvCompGrain;
806
807	rc = RTZipCompCreate(&pZip, &DeflateState, vmdkFileDeflateHelper,
808	RTZIPTYPE_ZLIB, RTZIPLEVEL_DEFAULT);
809	if (RT_FAILURE(rc))
810	return rc;
811	rc = RTZipCompress(pZip, pvBuf, cbToWrite);
812	if (RT_SUCCESS(rc))
813	rc = RTZipCompFinish(pZip);
814	RTZipCompDestroy(pZip);
815	if (RT_SUCCESS(rc))
816	{
817	Assert( DeflateState.iOffset > 0
818	&& (size_t)DeflateState.iOffset <= DeflateState.cbCompGrain);
819
820	/* pad with zeroes to get to a full sector size */
821	uint32_t uSize = DeflateState.iOffset;
822	if (uSize % 512)
823	{
824	uint32_t uSizeAlign = RT_ALIGN(uSize, 512);
825	memset((uint8_t *)pExtent->pvCompGrain + uSize, '\0',
826	uSizeAlign - uSize);
827	uSize = uSizeAlign;
828	}
829
830	if (pcbMarkerData)
831	*pcbMarkerData = uSize;
832
833	/* Compressed grain marker. Data follows immediately. */
834	VMDKMARKER pMarker = (VMDKMARKER )pExtent->pvCompGrain;
835	pMarker->uSector = RT_H2LE_U64(uLBA);
836	pMarker->cbSize = RT_H2LE_U32( DeflateState.iOffset
837	- RT_OFFSETOF(VMDKMARKER, uType));
838	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
839	uOffset, pMarker, uSize);
840	if (RT_FAILURE(rc))
841	return rc;
842	}
843	return rc;
844	}
845
846
847	/**
848	* Internal: check if all files are closed, prevent leaking resources.
849	*/
850	static int vmdkFileCheckAllClose(PVMDKIMAGE pImage)
851	{
852	int rc = VINF_SUCCESS, rc2;
853	PVMDKFILE pVmdkFile;
854
855	Assert(pImage->pFiles == NULL);
856	for (pVmdkFile = pImage->pFiles;
857	pVmdkFile != NULL;
858	pVmdkFile = pVmdkFile->pNext)
859	{
860	LogRel(("VMDK: leaking reference to file \"%s\"\n",
861	pVmdkFile->pszFilename));
862	pImage->pFiles = pVmdkFile->pNext;
863
864	rc2 = vmdkFileClose(pImage, &pVmdkFile, pVmdkFile->fDelete);
865
866	if (RT_SUCCESS(rc))
867	rc = rc2;
868	}
869	return rc;
870	}
871
872	/**
873	* Internal: truncate a string (at a UTF8 code point boundary) and encode the
874	* critical non-ASCII characters.
875	*/
876	static char vmdkEncodeString(const char psz)
877	{
878	char szEnc[VMDK_ENCODED_COMMENT_MAX + 3];
879	char *pszDst = szEnc;
880
881	AssertPtr(psz);
882
883	for (; *psz; psz = RTStrNextCp(psz))
884	{
885	char *pszDstPrev = pszDst;
886	RTUNICP Cp = RTStrGetCp(psz);
887	if (Cp == '\\')
888	{
889	pszDst = RTStrPutCp(pszDst, Cp);
890	pszDst = RTStrPutCp(pszDst, Cp);
891	}
892	else if (Cp == '\n')
893	{
894	pszDst = RTStrPutCp(pszDst, '\\');
895	pszDst = RTStrPutCp(pszDst, 'n');
896	}
897	else if (Cp == '\r')
898	{
899	pszDst = RTStrPutCp(pszDst, '\\');
900	pszDst = RTStrPutCp(pszDst, 'r');
901	}
902	else
903	pszDst = RTStrPutCp(pszDst, Cp);
904	if (pszDst - szEnc >= VMDK_ENCODED_COMMENT_MAX - 1)
905	{
906	pszDst = pszDstPrev;
907	break;
908	}
909	}
910	*pszDst = '\0';
911	return RTStrDup(szEnc);
912	}
913
914	/**
915	* Internal: decode a string and store it into the specified string.
916	*/
917	static int vmdkDecodeString(const char pszEncoded, char psz, size_t cb)
918	{
919	int rc = VINF_SUCCESS;
920	char szBuf[4];
921
922	if (!cb)
923	return VERR_BUFFER_OVERFLOW;
924
925	AssertPtr(psz);
926
927	for (; *pszEncoded; pszEncoded = RTStrNextCp(pszEncoded))
928	{
929	char *pszDst = szBuf;
930	RTUNICP Cp = RTStrGetCp(pszEncoded);
931	if (Cp == '\\')
932	{
933	pszEncoded = RTStrNextCp(pszEncoded);
934	RTUNICP CpQ = RTStrGetCp(pszEncoded);
935	if (CpQ == 'n')
936	RTStrPutCp(pszDst, '\n');
937	else if (CpQ == 'r')
938	RTStrPutCp(pszDst, '\r');
939	else if (CpQ == '\0')
940	{
941	rc = VERR_VD_VMDK_INVALID_HEADER;
942	break;
943	}
944	else
945	RTStrPutCp(pszDst, CpQ);
946	}
947	else
948	pszDst = RTStrPutCp(pszDst, Cp);
949
950	/* Need to leave space for terminating NUL. */
951	if ((size_t)(pszDst - szBuf) + 1 >= cb)
952	{
953	rc = VERR_BUFFER_OVERFLOW;
954	break;
955	}
956	memcpy(psz, szBuf, pszDst - szBuf);
957	psz += pszDst - szBuf;
958	}
959	*psz = '\0';
960	return rc;
961	}
962
963	/**
964	* Internal: free all buffers associated with grain directories.
965	*/
966	static void vmdkFreeGrainDirectory(PVMDKEXTENT pExtent)
967	{
968	if (pExtent->pGD)
969	{
970	RTMemFree(pExtent->pGD);
971	pExtent->pGD = NULL;
972	}
973	if (pExtent->pRGD)
974	{
975	RTMemFree(pExtent->pRGD);
976	pExtent->pRGD = NULL;
977	}
978	}
979
980	/**
981	* Internal: allocate the compressed/uncompressed buffers for streamOptimized
982	* images.
983	*/
984	static int vmdkAllocStreamBuffers(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
985	{
986	int rc = VINF_SUCCESS;
987
988	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
989	{
990	/* streamOptimized extents need a compressed grain buffer, which must
991	* be big enough to hold uncompressible data (which needs ~8 bytes
992	* more than the uncompressed data), the marker and padding. */
993	pExtent->cbCompGrain = RT_ALIGN_Z( VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain)
994	+ 8 + sizeof(VMDKMARKER), 512);
995	pExtent->pvCompGrain = RTMemAlloc(pExtent->cbCompGrain);
996	if (!pExtent->pvCompGrain)
997	{
998	rc = VERR_NO_MEMORY;
999	goto out;
1000	}
1001
1002	/* streamOptimized extents need a decompressed grain buffer. */
1003	pExtent->pvGrain = RTMemAlloc(VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
1004	if (!pExtent->pvGrain)
1005	{
1006	rc = VERR_NO_MEMORY;
1007	goto out;
1008	}
1009	}
1010
1011	out:
1012	if (RT_FAILURE(rc))
1013	vmdkFreeStreamBuffers(pExtent);
1014	return rc;
1015	}
1016
1017	/**
1018	* Internal: allocate all buffers associated with grain directories.
1019	*/
1020	static int vmdkAllocGrainDirectory(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
1021	{
1022	int rc = VINF_SUCCESS;
1023	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
1024	uint32_t pGD = NULL, pRGD = NULL;
1025
1026	pGD = (uint32_t *)RTMemAllocZ(cbGD);
1027	if (!pGD)
1028	{
1029	rc = VERR_NO_MEMORY;
1030	goto out;
1031	}
1032	pExtent->pGD = pGD;
1033
1034	if (pExtent->uSectorRGD)
1035	{
1036	pRGD = (uint32_t *)RTMemAllocZ(cbGD);
1037	if (!pRGD)
1038	{
1039	rc = VERR_NO_MEMORY;
1040	goto out;
1041	}
1042	pExtent->pRGD = pRGD;
1043	}
1044
1045	out:
1046	if (RT_FAILURE(rc))
1047	vmdkFreeGrainDirectory(pExtent);
1048	return rc;
1049	}
1050
1051	static int vmdkReadGrainDirectory(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
1052	{
1053	int rc = VINF_SUCCESS;
1054	size_t i;
1055	uint32_t pGDTmp, pRGDTmp;
1056	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
1057
1058	if (pExtent->enmType != VMDKETYPE_HOSTED_SPARSE)
1059	goto out;
1060
1061	if ( pExtent->uSectorGD == VMDK_GD_AT_END
1062	\|\| pExtent->uSectorRGD == VMDK_GD_AT_END)
1063	{
1064	rc = VERR_INTERNAL_ERROR;
1065	goto out;
1066	}
1067
1068	rc = vmdkAllocGrainDirectory(pImage, pExtent);
1069	if (RT_FAILURE(rc))
1070	goto out;
1071
1072	/* The VMDK 1.1 spec seems to talk about compressed grain directories,
1073	* but in reality they are not compressed. */
1074	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1075	VMDK_SECTOR2BYTE(pExtent->uSectorGD),
1076	pExtent->pGD, cbGD);
1077	AssertRC(rc);
1078	if (RT_FAILURE(rc))
1079	{
1080	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1081	N_("VMDK: could not read grain directory in '%s': %Rrc"), pExtent->pszFullname, rc);
1082	goto out;
1083	}
1084	for (i = 0, pGDTmp = pExtent->pGD; i < pExtent->cGDEntries; i++, pGDTmp++)
1085	pGDTmp = RT_LE2H_U32(pGDTmp);
1086
1087	if ( pExtent->uSectorRGD
1088	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_SKIP_CONSISTENCY_CHECKS))
1089	{
1090	/* The VMDK 1.1 spec seems to talk about compressed grain directories,
1091	* but in reality they are not compressed. */
1092	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1093	VMDK_SECTOR2BYTE(pExtent->uSectorRGD),
1094	pExtent->pRGD, cbGD);
1095	AssertRC(rc);
1096	if (RT_FAILURE(rc))
1097	{
1098	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1099	N_("VMDK: could not read redundant grain directory in '%s'"), pExtent->pszFullname);
1100	goto out;
1101	}
1102	for (i = 0, pRGDTmp = pExtent->pRGD; i < pExtent->cGDEntries; i++, pRGDTmp++)
1103	pRGDTmp = RT_LE2H_U32(pRGDTmp);
1104
1105	/* Check grain table and redundant grain table for consistency. */
1106	size_t cbGT = pExtent->cGTEntries * sizeof(uint32_t);
1107	size_t cbGTBuffers = cbGT; /* Start with space for one GT. */
1108	size_t cbGTBuffersMax = _1M;
1109
1110	uint32_t pTmpGT1 = (uint32_t )RTMemAlloc(cbGTBuffers);
1111	uint32_t pTmpGT2 = (uint32_t )RTMemAlloc(cbGTBuffers);
1112
1113	if ( !pTmpGT1
1114	\|\| !pTmpGT2)
1115	rc = VERR_NO_MEMORY;
1116
1117	i = 0;
1118	pGDTmp = pExtent->pGD;
1119	pRGDTmp = pExtent->pRGD;
1120
1121	/* Loop through all entries. */
1122	while (i < pExtent->cGDEntries)
1123	{
1124	uint32_t uGTStart = *pGDTmp;
1125	uint32_t uRGTStart = *pRGDTmp;
1126	size_t cbGTRead = cbGT;
1127
1128	/* If no grain table is allocated skip the entry. */
1129	if (pGDTmp == 0 && pRGDTmp == 0)
1130	{
1131	i++;
1132	continue;
1133	}
1134
1135	if (pGDTmp == 0 \|\| pRGDTmp == 0 \|\| pGDTmp == pRGDTmp)
1136	{
1137	/* Just one grain directory entry refers to a not yet allocated
1138	* grain table or both grain directory copies refer to the same
1139	* grain table. Not allowed. */
1140	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: inconsistent references to grain directory in '%s'"), pExtent->pszFullname);
1141	break;
1142	}
1143
1144	i++;
1145	pGDTmp++;
1146	pRGDTmp++;
1147
1148	/*
1149	* Read a few tables at once if adjacent to decrease the number
1150	* of I/O requests. Read at maximum 1MB at once.
1151	*/
1152	while ( i < pExtent->cGDEntries
1153	&& cbGTRead < cbGTBuffersMax)
1154	{
1155	/* If no grain table is allocated skip the entry. */
1156	if (pGDTmp == 0 && pRGDTmp == 0)
1157	continue;
1158
1159	if (pGDTmp == 0 \|\| pRGDTmp == 0 \|\| pGDTmp == pRGDTmp)
1160	{
1161	/* Just one grain directory entry refers to a not yet allocated
1162	* grain table or both grain directory copies refer to the same
1163	* grain table. Not allowed. */
1164	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: inconsistent references to grain directory in '%s'"), pExtent->pszFullname);
1165	break;
1166	}
1167
1168	/* Check that the start offsets are adjacent.*/
1169	if ( VMDK_SECTOR2BYTE(uGTStart) + cbGTRead != VMDK_SECTOR2BYTE(*pGDTmp)
1170	\|\| VMDK_SECTOR2BYTE(uRGTStart) + cbGTRead != VMDK_SECTOR2BYTE(*pRGDTmp))
1171	break;
1172
1173	i++;
1174	pGDTmp++;
1175	pRGDTmp++;
1176	cbGTRead += cbGT;
1177	}
1178
1179	/* Increase buffers if required. */
1180	if ( RT_SUCCESS(rc)
1181	&& cbGTBuffers < cbGTRead)
1182	{
1183	uint32_t *pTmp;
1184	pTmp = (uint32_t *)RTMemRealloc(pTmpGT1, cbGTRead);
1185	if (pTmp)
1186	{
1187	pTmpGT1 = pTmp;
1188	pTmp = (uint32_t *)RTMemRealloc(pTmpGT2, cbGTRead);
1189	if (pTmp)
1190	pTmpGT2 = pTmp;
1191	else
1192	rc = VERR_NO_MEMORY;
1193	}
1194	else
1195	rc = VERR_NO_MEMORY;
1196
1197	if (rc == VERR_NO_MEMORY)
1198	{
1199	/* Reset to the old values. */
1200	rc = VINF_SUCCESS;
1201	i -= cbGTRead / cbGT;
1202	cbGTRead = cbGT;
1203
1204	/* Don't try to increase the buffer again in the next run. */
1205	cbGTBuffersMax = cbGTBuffers;
1206	}
1207	}
1208
1209	if (RT_SUCCESS(rc))
1210	{
1211	/* The VMDK 1.1 spec seems to talk about compressed grain tables,
1212	* but in reality they are not compressed. */
1213	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1214	VMDK_SECTOR2BYTE(uGTStart),
1215	pTmpGT1, cbGTRead);
1216	if (RT_FAILURE(rc))
1217	{
1218	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1219	N_("VMDK: error reading grain table in '%s'"), pExtent->pszFullname);
1220	break;
1221	}
1222	/* The VMDK 1.1 spec seems to talk about compressed grain tables,
1223	* but in reality they are not compressed. */
1224	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1225	VMDK_SECTOR2BYTE(uRGTStart),
1226	pTmpGT2, cbGTRead);
1227	if (RT_FAILURE(rc))
1228	{
1229	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1230	N_("VMDK: error reading backup grain table in '%s'"), pExtent->pszFullname);
1231	break;
1232	}
1233	if (memcmp(pTmpGT1, pTmpGT2, cbGTRead))
1234	{
1235	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1236	N_("VMDK: inconsistency between grain table and backup grain table in '%s'"), pExtent->pszFullname);
1237	break;
1238	}
1239	}
1240	} /* while (i < pExtent->cGDEntries) */
1241
1242	/** @todo figure out what to do for unclean VMDKs. */
1243	if (pTmpGT1)
1244	RTMemFree(pTmpGT1);
1245	if (pTmpGT2)
1246	RTMemFree(pTmpGT2);
1247	}
1248
1249	out:
1250	if (RT_FAILURE(rc))
1251	vmdkFreeGrainDirectory(pExtent);
1252	return rc;
1253	}
1254
1255	static int vmdkCreateGrainDirectory(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
1256	uint64_t uStartSector, bool fPreAlloc)
1257	{
1258	int rc = VINF_SUCCESS;
1259	unsigned i;
1260	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
1261	size_t cbGDRounded = RT_ALIGN_64(cbGD, 512);
1262	size_t cbGTRounded;
1263	uint64_t cbOverhead;
1264
1265	if (fPreAlloc)
1266	{
1267	cbGTRounded = RT_ALIGN_64(pExtent->cGDEntries * pExtent->cGTEntries * sizeof(uint32_t), 512);
1268	cbOverhead = VMDK_SECTOR2BYTE(uStartSector) + cbGDRounded
1269	+ cbGTRounded;
1270	}
1271	else
1272	{
1273	/* Use a dummy start sector for layout computation. */
1274	if (uStartSector == VMDK_GD_AT_END)
1275	uStartSector = 1;
1276	cbGTRounded = 0;
1277	cbOverhead = VMDK_SECTOR2BYTE(uStartSector) + cbGDRounded;
1278	}
1279
1280	/* For streamOptimized extents there is only one grain directory,
1281	* and for all others take redundant grain directory into account. */
1282	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
1283	{
1284	cbOverhead = RT_ALIGN_64(cbOverhead,
1285	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
1286	}
1287	else
1288	{
1289	cbOverhead += cbGDRounded + cbGTRounded;
1290	cbOverhead = RT_ALIGN_64(cbOverhead,
1291	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
1292	rc = vdIfIoIntFileSetSize(pImage->pIfIo, pExtent->pFile->pStorage, cbOverhead);
1293	}
1294	if (RT_FAILURE(rc))
1295	goto out;
1296	pExtent->uAppendPosition = cbOverhead;
1297	pExtent->cOverheadSectors = VMDK_BYTE2SECTOR(cbOverhead);
1298
1299	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
1300	{
1301	pExtent->uSectorRGD = 0;
1302	pExtent->uSectorGD = uStartSector;
1303	}
1304	else
1305	{
1306	pExtent->uSectorRGD = uStartSector;
1307	pExtent->uSectorGD = uStartSector + VMDK_BYTE2SECTOR(cbGDRounded + cbGTRounded);
1308	}
1309
1310	rc = vmdkAllocStreamBuffers(pImage, pExtent);
1311	if (RT_FAILURE(rc))
1312	goto out;
1313
1314	rc = vmdkAllocGrainDirectory(pImage, pExtent);
1315	if (RT_FAILURE(rc))
1316	goto out;
1317
1318	if (fPreAlloc)
1319	{
1320	uint32_t uGTSectorLE;
1321	uint64_t uOffsetSectors;
1322
1323	if (pExtent->pRGD)
1324	{
1325	uOffsetSectors = pExtent->uSectorRGD + VMDK_BYTE2SECTOR(cbGDRounded);
1326	for (i = 0; i < pExtent->cGDEntries; i++)
1327	{
1328	pExtent->pRGD[i] = uOffsetSectors;
1329	uGTSectorLE = RT_H2LE_U64(uOffsetSectors);
1330	/* Write the redundant grain directory entry to disk. */
1331	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
1332	VMDK_SECTOR2BYTE(pExtent->uSectorRGD) + i * sizeof(uGTSectorLE),
1333	&uGTSectorLE, sizeof(uGTSectorLE));
1334	if (RT_FAILURE(rc))
1335	{
1336	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write new redundant grain directory entry in '%s'"), pExtent->pszFullname);
1337	goto out;
1338	}
1339	uOffsetSectors += VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
1340	}
1341	}
1342
1343	uOffsetSectors = pExtent->uSectorGD + VMDK_BYTE2SECTOR(cbGDRounded);
1344	for (i = 0; i < pExtent->cGDEntries; i++)
1345	{
1346	pExtent->pGD[i] = uOffsetSectors;
1347	uGTSectorLE = RT_H2LE_U64(uOffsetSectors);
1348	/* Write the grain directory entry to disk. */
1349	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
1350	VMDK_SECTOR2BYTE(pExtent->uSectorGD) + i * sizeof(uGTSectorLE),
1351	&uGTSectorLE, sizeof(uGTSectorLE));
1352	if (RT_FAILURE(rc))
1353	{
1354	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write new grain directory entry in '%s'"), pExtent->pszFullname);
1355	goto out;
1356	}
1357	uOffsetSectors += VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
1358	}
1359	}
1360
1361	out:
1362	if (RT_FAILURE(rc))
1363	vmdkFreeGrainDirectory(pExtent);
1364	return rc;
1365	}
1366
1367	static int vmdkStringUnquote(PVMDKIMAGE pImage, const char *pszStr,
1368	char ppszUnquoted, char ppszNext)
1369	{
1370	char *pszQ;
1371	char *pszUnquoted;
1372
1373	/* Skip over whitespace. */
1374	while (pszStr == ' ' \|\| pszStr == '\t')
1375	pszStr++;
1376
1377	if (*pszStr != '"')
1378	{
1379	pszQ = (char *)pszStr;
1380	while (pszQ && pszQ != ' ' && *pszQ != '\t')
1381	pszQ++;
1382	}
1383	else
1384	{
1385	pszStr++;
1386	pszQ = (char *)strchr(pszStr, '"');
1387	if (pszQ == NULL)
1388	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrectly quoted value in descriptor in '%s'"), pImage->pszFilename);
1389	}
1390
1391	pszUnquoted = (char *)RTMemTmpAlloc(pszQ - pszStr + 1);
1392	if (!pszUnquoted)
1393	return VERR_NO_MEMORY;
1394	memcpy(pszUnquoted, pszStr, pszQ - pszStr);
1395	pszUnquoted[pszQ - pszStr] = '\0';
1396	*ppszUnquoted = pszUnquoted;
1397	if (ppszNext)
1398	*ppszNext = pszQ + 1;
1399	return VINF_SUCCESS;
1400	}
1401
1402	static int vmdkDescInitStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1403	const char *pszLine)
1404	{
1405	char *pEnd = pDescriptor->aLines[pDescriptor->cLines];
1406	ssize_t cbDiff = strlen(pszLine) + 1;
1407
1408	if ( pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1
1409	&& pEnd - pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff)
1410	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1411
1412	memcpy(pEnd, pszLine, cbDiff);
1413	pDescriptor->cLines++;
1414	pDescriptor->aLines[pDescriptor->cLines] = pEnd + cbDiff;
1415	pDescriptor->fDirty = true;
1416
1417	return VINF_SUCCESS;
1418	}
1419
1420	static bool vmdkDescGetStr(PVMDKDESCRIPTOR pDescriptor, unsigned uStart,
1421	const char pszKey, const char *ppszValue)
1422	{
1423	size_t cbKey = strlen(pszKey);
1424	const char *pszValue;
1425
1426	while (uStart != 0)
1427	{
1428	if (!strncmp(pDescriptor->aLines[uStart], pszKey, cbKey))
1429	{
1430	/* Key matches, check for a '=' (preceded by whitespace). */
1431	pszValue = pDescriptor->aLines[uStart] + cbKey;
1432	while (pszValue == ' ' \|\| pszValue == '\t')
1433	pszValue++;
1434	if (*pszValue == '=')
1435	{
1436	*ppszValue = pszValue + 1;
1437	break;
1438	}
1439	}
1440	uStart = pDescriptor->aNextLines[uStart];
1441	}
1442	return !!uStart;
1443	}
1444
1445	static int vmdkDescSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1446	unsigned uStart,
1447	const char pszKey, const char pszValue)
1448	{
1449	char *pszTmp;
1450	size_t cbKey = strlen(pszKey);
1451	unsigned uLast = 0;
1452
1453	while (uStart != 0)
1454	{
1455	if (!strncmp(pDescriptor->aLines[uStart], pszKey, cbKey))
1456	{
1457	/* Key matches, check for a '=' (preceded by whitespace). */
1458	pszTmp = pDescriptor->aLines[uStart] + cbKey;
1459	while (pszTmp == ' ' \|\| pszTmp == '\t')
1460	pszTmp++;
1461	if (*pszTmp == '=')
1462	{
1463	pszTmp++;
1464	while (pszTmp == ' ' \|\| pszTmp == '\t')
1465	pszTmp++;
1466	break;
1467	}
1468	}
1469	if (!pDescriptor->aNextLines[uStart])
1470	uLast = uStart;
1471	uStart = pDescriptor->aNextLines[uStart];
1472	}
1473	if (uStart)
1474	{
1475	if (pszValue)
1476	{
1477	/* Key already exists, replace existing value. */
1478	size_t cbOldVal = strlen(pszTmp);
1479	size_t cbNewVal = strlen(pszValue);
1480	ssize_t cbDiff = cbNewVal - cbOldVal;
1481	/* Check for buffer overflow. */
1482	if ( pDescriptor->aLines[pDescriptor->cLines]
1483	- pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff)
1484	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1485
1486	memmove(pszTmp + cbNewVal, pszTmp + cbOldVal,
1487	pDescriptor->aLines[pDescriptor->cLines] - pszTmp - cbOldVal);
1488	memcpy(pszTmp, pszValue, cbNewVal + 1);
1489	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1490	pDescriptor->aLines[i] += cbDiff;
1491	}
1492	else
1493	{
1494	memmove(pDescriptor->aLines[uStart], pDescriptor->aLines[uStart+1],
1495	pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[uStart+1] + 1);
1496	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1497	{
1498	pDescriptor->aLines[i-1] = pDescriptor->aLines[i];
1499	if (pDescriptor->aNextLines[i])
1500	pDescriptor->aNextLines[i-1] = pDescriptor->aNextLines[i] - 1;
1501	else
1502	pDescriptor->aNextLines[i-1] = 0;
1503	}
1504	pDescriptor->cLines--;
1505	/* Adjust starting line numbers of following descriptor sections. */
1506	if (uStart < pDescriptor->uFirstExtent)
1507	pDescriptor->uFirstExtent--;
1508	if (uStart < pDescriptor->uFirstDDB)
1509	pDescriptor->uFirstDDB--;
1510	}
1511	}
1512	else
1513	{
1514	/* Key doesn't exist, append after the last entry in this category. */
1515	if (!pszValue)
1516	{
1517	/* Key doesn't exist, and it should be removed. Simply a no-op. */
1518	return VINF_SUCCESS;
1519	}
1520	cbKey = strlen(pszKey);
1521	size_t cbValue = strlen(pszValue);
1522	ssize_t cbDiff = cbKey + 1 + cbValue + 1;
1523	/* Check for buffer overflow. */
1524	if ( (pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1)
1525	\|\| ( pDescriptor->aLines[pDescriptor->cLines]
1526	- pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff))
1527	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1528	for (unsigned i = pDescriptor->cLines + 1; i > uLast + 1; i--)
1529	{
1530	pDescriptor->aLines[i] = pDescriptor->aLines[i - 1];
1531	if (pDescriptor->aNextLines[i - 1])
1532	pDescriptor->aNextLines[i] = pDescriptor->aNextLines[i - 1] + 1;
1533	else
1534	pDescriptor->aNextLines[i] = 0;
1535	}
1536	uStart = uLast + 1;
1537	pDescriptor->aNextLines[uLast] = uStart;
1538	pDescriptor->aNextLines[uStart] = 0;
1539	pDescriptor->cLines++;
1540	pszTmp = pDescriptor->aLines[uStart];
1541	memmove(pszTmp + cbDiff, pszTmp,
1542	pDescriptor->aLines[pDescriptor->cLines] - pszTmp);
1543	memcpy(pDescriptor->aLines[uStart], pszKey, cbKey);
1544	pDescriptor->aLines[uStart][cbKey] = '=';
1545	memcpy(pDescriptor->aLines[uStart] + cbKey + 1, pszValue, cbValue + 1);
1546	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1547	pDescriptor->aLines[i] += cbDiff;
1548
1549	/* Adjust starting line numbers of following descriptor sections. */
1550	if (uStart <= pDescriptor->uFirstExtent)
1551	pDescriptor->uFirstExtent++;
1552	if (uStart <= pDescriptor->uFirstDDB)
1553	pDescriptor->uFirstDDB++;
1554	}
1555	pDescriptor->fDirty = true;
1556	return VINF_SUCCESS;
1557	}
1558
1559	static int vmdkDescBaseGetU32(PVMDKDESCRIPTOR pDescriptor, const char *pszKey,
1560	uint32_t *puValue)
1561	{
1562	const char *pszValue;
1563
1564	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDesc, pszKey,
1565	&pszValue))
1566	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1567	return RTStrToUInt32Ex(pszValue, NULL, 10, puValue);
1568	}
1569
1570	static int vmdkDescBaseGetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1571	const char pszKey, const char *ppszValue)
1572	{
1573	const char *pszValue;
1574	char *pszValueUnquoted;
1575
1576	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDesc, pszKey,
1577	&pszValue))
1578	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1579	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1580	if (RT_FAILURE(rc))
1581	return rc;
1582	*ppszValue = pszValueUnquoted;
1583	return rc;
1584	}
1585
1586	static int vmdkDescBaseSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1587	const char pszKey, const char pszValue)
1588	{
1589	char *pszValueQuoted;
1590
1591	RTStrAPrintf(&pszValueQuoted, "\"%s\"", pszValue);
1592	if (!pszValueQuoted)
1593	return VERR_NO_STR_MEMORY;
1594	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc, pszKey,
1595	pszValueQuoted);
1596	RTStrFree(pszValueQuoted);
1597	return rc;
1598	}
1599
1600	static void vmdkDescExtRemoveDummy(PVMDKIMAGE pImage,
1601	PVMDKDESCRIPTOR pDescriptor)
1602	{
1603	unsigned uEntry = pDescriptor->uFirstExtent;
1604	ssize_t cbDiff;
1605
1606	if (!uEntry)
1607	return;
1608
1609	cbDiff = strlen(pDescriptor->aLines[uEntry]) + 1;
1610	/* Move everything including \0 in the entry marking the end of buffer. */
1611	memmove(pDescriptor->aLines[uEntry], pDescriptor->aLines[uEntry + 1],
1612	pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[uEntry + 1] + 1);
1613	for (unsigned i = uEntry + 1; i <= pDescriptor->cLines; i++)
1614	{
1615	pDescriptor->aLines[i - 1] = pDescriptor->aLines[i] - cbDiff;
1616	if (pDescriptor->aNextLines[i])
1617	pDescriptor->aNextLines[i - 1] = pDescriptor->aNextLines[i] - 1;
1618	else
1619	pDescriptor->aNextLines[i - 1] = 0;
1620	}
1621	pDescriptor->cLines--;
1622	if (pDescriptor->uFirstDDB)
1623	pDescriptor->uFirstDDB--;
1624
1625	return;
1626	}
1627
1628	static int vmdkDescExtInsert(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1629	VMDKACCESS enmAccess, uint64_t cNominalSectors,
1630	VMDKETYPE enmType, const char *pszBasename,
1631	uint64_t uSectorOffset)
1632	{
1633	static const char *apszAccess[] = { "NOACCESS", "RDONLY", "RW" };
1634	static const char *apszType[] = { "", "SPARSE", "FLAT", "ZERO", "VMFS" };
1635	char *pszTmp;
1636	unsigned uStart = pDescriptor->uFirstExtent, uLast = 0;
1637	char szExt[1024];
1638	ssize_t cbDiff;
1639
1640	Assert((unsigned)enmAccess < RT_ELEMENTS(apszAccess));
1641	Assert((unsigned)enmType < RT_ELEMENTS(apszType));
1642
1643	/* Find last entry in extent description. */
1644	while (uStart)
1645	{
1646	if (!pDescriptor->aNextLines[uStart])
1647	uLast = uStart;
1648	uStart = pDescriptor->aNextLines[uStart];
1649	}
1650
1651	if (enmType == VMDKETYPE_ZERO)
1652	{
1653	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s ", apszAccess[enmAccess],
1654	cNominalSectors, apszType[enmType]);
1655	}
1656	else if (enmType == VMDKETYPE_FLAT)
1657	{
1658	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s \"%s\" %llu",
1659	apszAccess[enmAccess], cNominalSectors,
1660	apszType[enmType], pszBasename, uSectorOffset);
1661	}
1662	else
1663	{
1664	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s \"%s\"",
1665	apszAccess[enmAccess], cNominalSectors,
1666	apszType[enmType], pszBasename);
1667	}
1668	cbDiff = strlen(szExt) + 1;
1669
1670	/* Check for buffer overflow. */
1671	if ( (pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1)
1672	\|\| ( pDescriptor->aLines[pDescriptor->cLines]
1673	- pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff))
1674	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1675
1676	for (unsigned i = pDescriptor->cLines + 1; i > uLast + 1; i--)
1677	{
1678	pDescriptor->aLines[i] = pDescriptor->aLines[i - 1];
1679	if (pDescriptor->aNextLines[i - 1])
1680	pDescriptor->aNextLines[i] = pDescriptor->aNextLines[i - 1] + 1;
1681	else
1682	pDescriptor->aNextLines[i] = 0;
1683	}
1684	uStart = uLast + 1;
1685	pDescriptor->aNextLines[uLast] = uStart;
1686	pDescriptor->aNextLines[uStart] = 0;
1687	pDescriptor->cLines++;
1688	pszTmp = pDescriptor->aLines[uStart];
1689	memmove(pszTmp + cbDiff, pszTmp,
1690	pDescriptor->aLines[pDescriptor->cLines] - pszTmp);
1691	memcpy(pDescriptor->aLines[uStart], szExt, cbDiff);
1692	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1693	pDescriptor->aLines[i] += cbDiff;
1694
1695	/* Adjust starting line numbers of following descriptor sections. */
1696	if (uStart <= pDescriptor->uFirstDDB)
1697	pDescriptor->uFirstDDB++;
1698
1699	pDescriptor->fDirty = true;
1700	return VINF_SUCCESS;
1701	}
1702
1703	static int vmdkDescDDBGetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1704	const char pszKey, const char *ppszValue)
1705	{
1706	const char *pszValue;
1707	char *pszValueUnquoted;
1708
1709	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1710	&pszValue))
1711	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1712	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1713	if (RT_FAILURE(rc))
1714	return rc;
1715	*ppszValue = pszValueUnquoted;
1716	return rc;
1717	}
1718
1719	static int vmdkDescDDBGetU32(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1720	const char pszKey, uint32_t puValue)
1721	{
1722	const char *pszValue;
1723	char *pszValueUnquoted;
1724
1725	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1726	&pszValue))
1727	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1728	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1729	if (RT_FAILURE(rc))
1730	return rc;
1731	rc = RTStrToUInt32Ex(pszValueUnquoted, NULL, 10, puValue);
1732	RTMemTmpFree(pszValueUnquoted);
1733	return rc;
1734	}
1735
1736	static int vmdkDescDDBGetUuid(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1737	const char *pszKey, PRTUUID pUuid)
1738	{
1739	const char *pszValue;
1740	char *pszValueUnquoted;
1741
1742	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1743	&pszValue))
1744	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1745	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1746	if (RT_FAILURE(rc))
1747	return rc;
1748	rc = RTUuidFromStr(pUuid, pszValueUnquoted);
1749	RTMemTmpFree(pszValueUnquoted);
1750	return rc;
1751	}
1752
1753	static int vmdkDescDDBSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1754	const char pszKey, const char pszVal)
1755	{
1756	int rc;
1757	char *pszValQuoted;
1758
1759	if (pszVal)
1760	{
1761	RTStrAPrintf(&pszValQuoted, "\"%s\"", pszVal);
1762	if (!pszValQuoted)
1763	return VERR_NO_STR_MEMORY;
1764	}
1765	else
1766	pszValQuoted = NULL;
1767	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1768	pszValQuoted);
1769	if (pszValQuoted)
1770	RTStrFree(pszValQuoted);
1771	return rc;
1772	}
1773
1774	static int vmdkDescDDBSetUuid(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1775	const char *pszKey, PCRTUUID pUuid)
1776	{
1777	char *pszUuid;
1778
1779	RTStrAPrintf(&pszUuid, "\"%RTuuid\"", pUuid);
1780	if (!pszUuid)
1781	return VERR_NO_STR_MEMORY;
1782	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1783	pszUuid);
1784	RTStrFree(pszUuid);
1785	return rc;
1786	}
1787
1788	static int vmdkDescDDBSetU32(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1789	const char *pszKey, uint32_t uValue)
1790	{
1791	char *pszValue;
1792
1793	RTStrAPrintf(&pszValue, "\"%d\"", uValue);
1794	if (!pszValue)
1795	return VERR_NO_STR_MEMORY;
1796	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1797	pszValue);
1798	RTStrFree(pszValue);
1799	return rc;
1800	}
1801
1802	static int vmdkPreprocessDescriptor(PVMDKIMAGE pImage, char *pDescData,
1803	size_t cbDescData,
1804	PVMDKDESCRIPTOR pDescriptor)
1805	{
1806	int rc = VINF_SUCCESS;
1807	unsigned cLine = 0, uLastNonEmptyLine = 0;
1808	char *pTmp = pDescData;
1809
1810	pDescriptor->cbDescAlloc = cbDescData;
1811	while (*pTmp != '\0')
1812	{
1813	pDescriptor->aLines[cLine++] = pTmp;
1814	if (cLine >= VMDK_DESCRIPTOR_LINES_MAX)
1815	{
1816	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1817	goto out;
1818	}
1819
1820	while (pTmp != '\0' && pTmp != '\n')
1821	{
1822	if (*pTmp == '\r')
1823	{
1824	if (*(pTmp + 1) != '\n')
1825	{
1826	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: unsupported end of line in descriptor in '%s'"), pImage->pszFilename);
1827	goto out;
1828	}
1829	else
1830	{
1831	/* Get rid of CR character. */
1832	*pTmp = '\0';
1833	}
1834	}
1835	pTmp++;
1836	}
1837	/* Get rid of LF character. */
1838	if (*pTmp == '\n')
1839	{
1840	*pTmp = '\0';
1841	pTmp++;
1842	}
1843	}
1844	pDescriptor->cLines = cLine;
1845	/* Pointer right after the end of the used part of the buffer. */
1846	pDescriptor->aLines[cLine] = pTmp;
1847
1848	if ( strcmp(pDescriptor->aLines[0], "# Disk DescriptorFile")
1849	&& strcmp(pDescriptor->aLines[0], "# Disk Descriptor File"))
1850	{
1851	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor does not start as expected in '%s'"), pImage->pszFilename);
1852	goto out;
1853	}
1854
1855	/* Initialize those, because we need to be able to reopen an image. */
1856	pDescriptor->uFirstDesc = 0;
1857	pDescriptor->uFirstExtent = 0;
1858	pDescriptor->uFirstDDB = 0;
1859	for (unsigned i = 0; i < cLine; i++)
1860	{
1861	if (pDescriptor->aLines[i] != '#' && pDescriptor->aLines[i] != '\0')
1862	{
1863	if ( !strncmp(pDescriptor->aLines[i], "RW", 2)
1864	\|\| !strncmp(pDescriptor->aLines[i], "RDONLY", 6)
1865	\|\| !strncmp(pDescriptor->aLines[i], "NOACCESS", 8) )
1866	{
1867	/* An extent descriptor. */
1868	if (!pDescriptor->uFirstDesc \|\| pDescriptor->uFirstDDB)
1869	{
1870	/* Incorrect ordering of entries. */
1871	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
1872	goto out;
1873	}
1874	if (!pDescriptor->uFirstExtent)
1875	{
1876	pDescriptor->uFirstExtent = i;
1877	uLastNonEmptyLine = 0;
1878	}
1879	}
1880	else if (!strncmp(pDescriptor->aLines[i], "ddb.", 4))
1881	{
1882	/* A disk database entry. */
1883	if (!pDescriptor->uFirstDesc \|\| !pDescriptor->uFirstExtent)
1884	{
1885	/* Incorrect ordering of entries. */
1886	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
1887	goto out;
1888	}
1889	if (!pDescriptor->uFirstDDB)
1890	{
1891	pDescriptor->uFirstDDB = i;
1892	uLastNonEmptyLine = 0;
1893	}
1894	}
1895	else
1896	{
1897	/* A normal entry. */
1898	if (pDescriptor->uFirstExtent \|\| pDescriptor->uFirstDDB)
1899	{
1900	/* Incorrect ordering of entries. */
1901	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
1902	goto out;
1903	}
1904	if (!pDescriptor->uFirstDesc)
1905	{
1906	pDescriptor->uFirstDesc = i;
1907	uLastNonEmptyLine = 0;
1908	}
1909	}
1910	if (uLastNonEmptyLine)
1911	pDescriptor->aNextLines[uLastNonEmptyLine] = i;
1912	uLastNonEmptyLine = i;
1913	}
1914	}
1915
1916	out:
1917	return rc;
1918	}
1919
1920	static int vmdkDescSetPCHSGeometry(PVMDKIMAGE pImage,
1921	PCVDGEOMETRY pPCHSGeometry)
1922	{
1923	int rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1924	VMDK_DDB_GEO_PCHS_CYLINDERS,
1925	pPCHSGeometry->cCylinders);
1926	if (RT_FAILURE(rc))
1927	return rc;
1928	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1929	VMDK_DDB_GEO_PCHS_HEADS,
1930	pPCHSGeometry->cHeads);
1931	if (RT_FAILURE(rc))
1932	return rc;
1933	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1934	VMDK_DDB_GEO_PCHS_SECTORS,
1935	pPCHSGeometry->cSectors);
1936	return rc;
1937	}
1938
1939	static int vmdkDescSetLCHSGeometry(PVMDKIMAGE pImage,
1940	PCVDGEOMETRY pLCHSGeometry)
1941	{
1942	int rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1943	VMDK_DDB_GEO_LCHS_CYLINDERS,
1944	pLCHSGeometry->cCylinders);
1945	if (RT_FAILURE(rc))
1946	return rc;
1947	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1948	VMDK_DDB_GEO_LCHS_HEADS,
1949
1950	pLCHSGeometry->cHeads);
1951	if (RT_FAILURE(rc))
1952	return rc;
1953	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1954	VMDK_DDB_GEO_LCHS_SECTORS,
1955	pLCHSGeometry->cSectors);
1956	return rc;
1957	}
1958
1959	static int vmdkCreateDescriptor(PVMDKIMAGE pImage, char *pDescData,
1960	size_t cbDescData, PVMDKDESCRIPTOR pDescriptor)
1961	{
1962	int rc;
1963
1964	pDescriptor->uFirstDesc = 0;
1965	pDescriptor->uFirstExtent = 0;
1966	pDescriptor->uFirstDDB = 0;
1967	pDescriptor->cLines = 0;
1968	pDescriptor->cbDescAlloc = cbDescData;
1969	pDescriptor->fDirty = false;
1970	pDescriptor->aLines[pDescriptor->cLines] = pDescData;
1971	memset(pDescriptor->aNextLines, '\0', sizeof(pDescriptor->aNextLines));
1972
1973	rc = vmdkDescInitStr(pImage, pDescriptor, "# Disk DescriptorFile");
1974	if (RT_FAILURE(rc))
1975	goto out;
1976	rc = vmdkDescInitStr(pImage, pDescriptor, "version=1");
1977	if (RT_FAILURE(rc))
1978	goto out;
1979	pDescriptor->uFirstDesc = pDescriptor->cLines - 1;
1980	rc = vmdkDescInitStr(pImage, pDescriptor, "");
1981	if (RT_FAILURE(rc))
1982	goto out;
1983	rc = vmdkDescInitStr(pImage, pDescriptor, "# Extent description");
1984	if (RT_FAILURE(rc))
1985	goto out;
1986	rc = vmdkDescInitStr(pImage, pDescriptor, "NOACCESS 0 ZERO ");
1987	if (RT_FAILURE(rc))
1988	goto out;
1989	pDescriptor->uFirstExtent = pDescriptor->cLines - 1;
1990	rc = vmdkDescInitStr(pImage, pDescriptor, "");
1991	if (RT_FAILURE(rc))
1992	goto out;
1993	/* The trailing space is created by VMware, too. */
1994	rc = vmdkDescInitStr(pImage, pDescriptor, "# The disk Data Base ");
1995	if (RT_FAILURE(rc))
1996	goto out;
1997	rc = vmdkDescInitStr(pImage, pDescriptor, "#DDB");
1998	if (RT_FAILURE(rc))
1999	goto out;
2000	rc = vmdkDescInitStr(pImage, pDescriptor, "");
2001	if (RT_FAILURE(rc))
2002	goto out;
2003	rc = vmdkDescInitStr(pImage, pDescriptor, "ddb.virtualHWVersion = \"4\"");
2004	if (RT_FAILURE(rc))
2005	goto out;
2006	pDescriptor->uFirstDDB = pDescriptor->cLines - 1;
2007
2008	/* Now that the framework is in place, use the normal functions to insert
2009	* the remaining keys. */
2010	char szBuf[9];
2011	RTStrPrintf(szBuf, sizeof(szBuf), "%08x", RTRandU32());
2012	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc,
2013	"CID", szBuf);
2014	if (RT_FAILURE(rc))
2015	goto out;
2016	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc,
2017	"parentCID", "ffffffff");
2018	if (RT_FAILURE(rc))
2019	goto out;
2020
2021	rc = vmdkDescDDBSetStr(pImage, pDescriptor, "ddb.adapterType", "ide");
2022	if (RT_FAILURE(rc))
2023	goto out;
2024
2025	out:
2026	return rc;
2027	}
2028
2029	static int vmdkParseDescriptor(PVMDKIMAGE pImage, char *pDescData,
2030	size_t cbDescData)
2031	{
2032	int rc;
2033	unsigned cExtents;
2034	unsigned uLine;
2035	unsigned i;
2036
2037	rc = vmdkPreprocessDescriptor(pImage, pDescData, cbDescData,
2038	&pImage->Descriptor);
2039	if (RT_FAILURE(rc))
2040	return rc;
2041
2042	/* Check version, must be 1. */
2043	uint32_t uVersion;
2044	rc = vmdkDescBaseGetU32(&pImage->Descriptor, "version", &uVersion);
2045	if (RT_FAILURE(rc))
2046	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error finding key 'version' in descriptor in '%s'"), pImage->pszFilename);
2047	if (uVersion != 1)
2048	return vdIfError(pImage->pIfError, VERR_VD_VMDK_UNSUPPORTED_VERSION, RT_SRC_POS, N_("VMDK: unsupported format version in descriptor in '%s'"), pImage->pszFilename);
2049
2050	/* Get image creation type and determine image flags. */
2051	const char pszCreateType = NULL; / initialized to make gcc shut up */
2052	rc = vmdkDescBaseGetStr(pImage, &pImage->Descriptor, "createType",
2053	&pszCreateType);
2054	if (RT_FAILURE(rc))
2055	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot get image type from descriptor in '%s'"), pImage->pszFilename);
2056	if ( !strcmp(pszCreateType, "twoGbMaxExtentSparse")
2057	\|\| !strcmp(pszCreateType, "twoGbMaxExtentFlat"))
2058	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_SPLIT_2G;
2059	else if ( !strcmp(pszCreateType, "partitionedDevice")
2060	\|\| !strcmp(pszCreateType, "fullDevice"))
2061	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_RAWDISK;
2062	else if (!strcmp(pszCreateType, "streamOptimized"))
2063	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED;
2064	else if (!strcmp(pszCreateType, "vmfs"))
2065	pImage->uImageFlags \|= VD_IMAGE_FLAGS_FIXED \| VD_VMDK_IMAGE_FLAGS_ESX;
2066	RTStrFree((char )(void )pszCreateType);
2067
2068	/* Count the number of extent config entries. */
2069	for (uLine = pImage->Descriptor.uFirstExtent, cExtents = 0;
2070	uLine != 0;
2071	uLine = pImage->Descriptor.aNextLines[uLine], cExtents++)
2072	/* nothing */;
2073
2074	if (!pImage->pDescData && cExtents != 1)
2075	{
2076	/* Monolithic image, must have only one extent (already opened). */
2077	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: monolithic image may only have one extent in '%s'"), pImage->pszFilename);
2078	}
2079
2080	if (pImage->pDescData)
2081	{
2082	/* Non-monolithic image, extents need to be allocated. */
2083	rc = vmdkCreateExtents(pImage, cExtents);
2084	if (RT_FAILURE(rc))
2085	return rc;
2086	}
2087
2088	for (i = 0, uLine = pImage->Descriptor.uFirstExtent;
2089	i < cExtents; i++, uLine = pImage->Descriptor.aNextLines[uLine])
2090	{
2091	char *pszLine = pImage->Descriptor.aLines[uLine];
2092
2093	/* Access type of the extent. */
2094	if (!strncmp(pszLine, "RW", 2))
2095	{
2096	pImage->pExtents[i].enmAccess = VMDKACCESS_READWRITE;
2097	pszLine += 2;
2098	}
2099	else if (!strncmp(pszLine, "RDONLY", 6))
2100	{
2101	pImage->pExtents[i].enmAccess = VMDKACCESS_READONLY;
2102	pszLine += 6;
2103	}
2104	else if (!strncmp(pszLine, "NOACCESS", 8))
2105	{
2106	pImage->pExtents[i].enmAccess = VMDKACCESS_NOACCESS;
2107	pszLine += 8;
2108	}
2109	else
2110	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2111	if (*pszLine++ != ' ')
2112	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2113
2114	/* Nominal size of the extent. */
2115	rc = RTStrToUInt64Ex(pszLine, &pszLine, 10,
2116	&pImage->pExtents[i].cNominalSectors);
2117	if (RT_FAILURE(rc))
2118	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2119	if (*pszLine++ != ' ')
2120	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2121
2122	/* Type of the extent. */
2123	#ifdef VBOX_WITH_VMDK_ESX
2124	/** @todo Add the ESX extent types. Not necessary for now because
2125	* the ESX extent types are only used inside an ESX server. They are
2126	* automatically converted if the VMDK is exported. */
2127	#endif /* VBOX_WITH_VMDK_ESX */
2128	if (!strncmp(pszLine, "SPARSE", 6))
2129	{
2130	pImage->pExtents[i].enmType = VMDKETYPE_HOSTED_SPARSE;
2131	pszLine += 6;
2132	}
2133	else if (!strncmp(pszLine, "FLAT", 4))
2134	{
2135	pImage->pExtents[i].enmType = VMDKETYPE_FLAT;
2136	pszLine += 4;
2137	}
2138	else if (!strncmp(pszLine, "ZERO", 4))
2139	{
2140	pImage->pExtents[i].enmType = VMDKETYPE_ZERO;
2141	pszLine += 4;
2142	}
2143	else if (!strncmp(pszLine, "VMFS", 4))
2144	{
2145	pImage->pExtents[i].enmType = VMDKETYPE_VMFS;
2146	pszLine += 4;
2147	}
2148	else
2149	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2150
2151	if (pImage->pExtents[i].enmType == VMDKETYPE_ZERO)
2152	{
2153	/* This one has no basename or offset. */
2154	if (*pszLine == ' ')
2155	pszLine++;
2156	if (*pszLine != '\0')
2157	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2158	pImage->pExtents[i].pszBasename = NULL;
2159	}
2160	else
2161	{
2162	/* All other extent types have basename and optional offset. */
2163	if (*pszLine++ != ' ')
2164	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2165
2166	/* Basename of the image. Surrounded by quotes. */
2167	char *pszBasename;
2168	rc = vmdkStringUnquote(pImage, pszLine, &pszBasename, &pszLine);
2169	if (RT_FAILURE(rc))
2170	return rc;
2171	pImage->pExtents[i].pszBasename = pszBasename;
2172	if (*pszLine == ' ')
2173	{
2174	pszLine++;
2175	if (*pszLine != '\0')
2176	{
2177	/* Optional offset in extent specified. */
2178	rc = RTStrToUInt64Ex(pszLine, &pszLine, 10,
2179	&pImage->pExtents[i].uSectorOffset);
2180	if (RT_FAILURE(rc))
2181	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2182	}
2183	}
2184
2185	if (*pszLine != '\0')
2186	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2187	}
2188	}
2189
2190	/* Determine PCHS geometry (autogenerate if necessary). */
2191	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2192	VMDK_DDB_GEO_PCHS_CYLINDERS,
2193	&pImage->PCHSGeometry.cCylinders);
2194	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2195	pImage->PCHSGeometry.cCylinders = 0;
2196	else if (RT_FAILURE(rc))
2197	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2198	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2199	VMDK_DDB_GEO_PCHS_HEADS,
2200	&pImage->PCHSGeometry.cHeads);
2201	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2202	pImage->PCHSGeometry.cHeads = 0;
2203	else if (RT_FAILURE(rc))
2204	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2205	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2206	VMDK_DDB_GEO_PCHS_SECTORS,
2207	&pImage->PCHSGeometry.cSectors);
2208	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2209	pImage->PCHSGeometry.cSectors = 0;
2210	else if (RT_FAILURE(rc))
2211	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2212	if ( pImage->PCHSGeometry.cCylinders == 0
2213	\|\| pImage->PCHSGeometry.cHeads == 0
2214	\|\| pImage->PCHSGeometry.cHeads > 16
2215	\|\| pImage->PCHSGeometry.cSectors == 0
2216	\|\| pImage->PCHSGeometry.cSectors > 63)
2217	{
2218	/* Mark PCHS geometry as not yet valid (can't do the calculation here
2219	* as the total image size isn't known yet). */
2220	pImage->PCHSGeometry.cCylinders = 0;
2221	pImage->PCHSGeometry.cHeads = 16;
2222	pImage->PCHSGeometry.cSectors = 63;
2223	}
2224
2225	/* Determine LCHS geometry (set to 0 if not specified). */
2226	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2227	VMDK_DDB_GEO_LCHS_CYLINDERS,
2228	&pImage->LCHSGeometry.cCylinders);
2229	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2230	pImage->LCHSGeometry.cCylinders = 0;
2231	else if (RT_FAILURE(rc))
2232	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2233	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2234	VMDK_DDB_GEO_LCHS_HEADS,
2235	&pImage->LCHSGeometry.cHeads);
2236	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2237	pImage->LCHSGeometry.cHeads = 0;
2238	else if (RT_FAILURE(rc))
2239	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2240	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2241	VMDK_DDB_GEO_LCHS_SECTORS,
2242	&pImage->LCHSGeometry.cSectors);
2243	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2244	pImage->LCHSGeometry.cSectors = 0;
2245	else if (RT_FAILURE(rc))
2246	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2247	if ( pImage->LCHSGeometry.cCylinders == 0
2248	\|\| pImage->LCHSGeometry.cHeads == 0
2249	\|\| pImage->LCHSGeometry.cSectors == 0)
2250	{
2251	pImage->LCHSGeometry.cCylinders = 0;
2252	pImage->LCHSGeometry.cHeads = 0;
2253	pImage->LCHSGeometry.cSectors = 0;
2254	}
2255
2256	/* Get image UUID. */
2257	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor, VMDK_DDB_IMAGE_UUID,
2258	&pImage->ImageUuid);
2259	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2260	{
2261	/* Image without UUID. Probably created by VMware and not yet used
2262	* by VirtualBox. Can only be added for images opened in read/write
2263	* mode, so don't bother producing a sensible UUID otherwise. */
2264	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2265	RTUuidClear(&pImage->ImageUuid);
2266	else
2267	{
2268	rc = RTUuidCreate(&pImage->ImageUuid);
2269	if (RT_FAILURE(rc))
2270	return rc;
2271	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2272	VMDK_DDB_IMAGE_UUID, &pImage->ImageUuid);
2273	if (RT_FAILURE(rc))
2274	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image UUID in descriptor in '%s'"), pImage->pszFilename);
2275	}
2276	}
2277	else if (RT_FAILURE(rc))
2278	return rc;
2279
2280	/* Get image modification UUID. */
2281	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor,
2282	VMDK_DDB_MODIFICATION_UUID,
2283	&pImage->ModificationUuid);
2284	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2285	{
2286	/* Image without UUID. Probably created by VMware and not yet used
2287	* by VirtualBox. Can only be added for images opened in read/write
2288	* mode, so don't bother producing a sensible UUID otherwise. */
2289	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2290	RTUuidClear(&pImage->ModificationUuid);
2291	else
2292	{
2293	rc = RTUuidCreate(&pImage->ModificationUuid);
2294	if (RT_FAILURE(rc))
2295	return rc;
2296	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2297	VMDK_DDB_MODIFICATION_UUID,
2298	&pImage->ModificationUuid);
2299	if (RT_FAILURE(rc))
2300	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image modification UUID in descriptor in '%s'"), pImage->pszFilename);
2301	}
2302	}
2303	else if (RT_FAILURE(rc))
2304	return rc;
2305
2306	/* Get UUID of parent image. */
2307	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor, VMDK_DDB_PARENT_UUID,
2308	&pImage->ParentUuid);
2309	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2310	{
2311	/* Image without UUID. Probably created by VMware and not yet used
2312	* by VirtualBox. Can only be added for images opened in read/write
2313	* mode, so don't bother producing a sensible UUID otherwise. */
2314	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2315	RTUuidClear(&pImage->ParentUuid);
2316	else
2317	{
2318	rc = RTUuidClear(&pImage->ParentUuid);
2319	if (RT_FAILURE(rc))
2320	return rc;
2321	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2322	VMDK_DDB_PARENT_UUID, &pImage->ParentUuid);
2323	if (RT_FAILURE(rc))
2324	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent UUID in descriptor in '%s'"), pImage->pszFilename);
2325	}
2326	}
2327	else if (RT_FAILURE(rc))
2328	return rc;
2329
2330	/* Get parent image modification UUID. */
2331	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor,
2332	VMDK_DDB_PARENT_MODIFICATION_UUID,
2333	&pImage->ParentModificationUuid);
2334	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2335	{
2336	/* Image without UUID. Probably created by VMware and not yet used
2337	* by VirtualBox. Can only be added for images opened in read/write
2338	* mode, so don't bother producing a sensible UUID otherwise. */
2339	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2340	RTUuidClear(&pImage->ParentModificationUuid);
2341	else
2342	{
2343	RTUuidClear(&pImage->ParentModificationUuid);
2344	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2345	VMDK_DDB_PARENT_MODIFICATION_UUID,
2346	&pImage->ParentModificationUuid);
2347	if (RT_FAILURE(rc))
2348	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent modification UUID in descriptor in '%s'"), pImage->pszFilename);
2349	}
2350	}
2351	else if (RT_FAILURE(rc))
2352	return rc;
2353
2354	return VINF_SUCCESS;
2355	}
2356
2357	/**
2358	* Internal : Prepares the descriptor to write to the image.
2359	*/
2360	static int vmdkDescriptorPrepare(PVMDKIMAGE pImage, uint64_t cbLimit,
2361	void *ppvData, size_t pcbData)
2362	{
2363	int rc = VINF_SUCCESS;
2364
2365	/*
2366	* Allocate temporary descriptor buffer.
2367	* In case there is no limit allocate a default
2368	* and increase if required.
2369	*/
2370	size_t cbDescriptor = cbLimit ? cbLimit : 4 * _1K;
2371	char pszDescriptor = (char )RTMemAllocZ(cbDescriptor);
2372	size_t offDescriptor = 0;
2373
2374	if (!pszDescriptor)
2375	return VERR_NO_MEMORY;
2376
2377	for (unsigned i = 0; i < pImage->Descriptor.cLines; i++)
2378	{
2379	const char *psz = pImage->Descriptor.aLines[i];
2380	size_t cb = strlen(psz);
2381
2382	/*
2383	* Increase the descriptor if there is no limit and
2384	* there is not enough room left for this line.
2385	*/
2386	if (offDescriptor + cb + 1 > cbDescriptor)
2387	{
2388	if (cbLimit)
2389	{
2390	rc = vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too long in '%s'"), pImage->pszFilename);
2391	break;
2392	}
2393	else
2394	{
2395	char *pszDescriptorNew = NULL;
2396	LogFlow(("Increasing descriptor cache\n"));
2397
2398	pszDescriptorNew = (char )RTMemRealloc(pszDescriptor, cbDescriptor + cb + 4 _1K);
2399	if (!pszDescriptorNew)
2400	{
2401	rc = VERR_NO_MEMORY;
2402	break;
2403	}
2404	pszDescriptor = pszDescriptorNew;
2405	cbDescriptor += cb + 4 * _1K;
2406	}
2407	}
2408
2409	if (cb > 0)
2410	{
2411	memcpy(pszDescriptor + offDescriptor, psz, cb);
2412	offDescriptor += cb;
2413	}
2414
2415	memcpy(pszDescriptor + offDescriptor, "\n", 1);
2416	offDescriptor++;
2417	}
2418
2419	if (RT_SUCCESS(rc))
2420	{
2421	*ppvData = pszDescriptor;
2422	*pcbData = offDescriptor;
2423	}
2424	else if (pszDescriptor)
2425	RTMemFree(pszDescriptor);
2426
2427	return rc;
2428	}
2429
2430	/**
2431	* Internal: write/update the descriptor part of the image.
2432	*/
2433	static int vmdkWriteDescriptor(PVMDKIMAGE pImage, PVDIOCTX pIoCtx)
2434	{
2435	int rc = VINF_SUCCESS;
2436	uint64_t cbLimit;
2437	uint64_t uOffset;
2438	PVMDKFILE pDescFile;
2439	void *pvDescriptor = NULL;
2440	size_t cbDescriptor;
2441
2442	if (pImage->pDescData)
2443	{
2444	/* Separate descriptor file. */
2445	uOffset = 0;
2446	cbLimit = 0;
2447	pDescFile = pImage->pFile;
2448	}
2449	else
2450	{
2451	/* Embedded descriptor file. */
2452	uOffset = VMDK_SECTOR2BYTE(pImage->pExtents[0].uDescriptorSector);
2453	cbLimit = VMDK_SECTOR2BYTE(pImage->pExtents[0].cDescriptorSectors);
2454	pDescFile = pImage->pExtents[0].pFile;
2455	}
2456	/* Bail out if there is no file to write to. */
2457	if (pDescFile == NULL)
2458	return VERR_INVALID_PARAMETER;
2459
2460	rc = vmdkDescriptorPrepare(pImage, cbLimit, &pvDescriptor, &cbDescriptor);
2461	if (RT_SUCCESS(rc))
2462	{
2463	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pDescFile->pStorage,
2464	uOffset, pvDescriptor,
2465	cbLimit ? cbLimit : cbDescriptor,
2466	pIoCtx, NULL, NULL);
2467	if ( RT_FAILURE(rc)
2468	&& rc != VERR_VD_ASYNC_IO_IN_PROGRESS)
2469	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error writing descriptor in '%s'"), pImage->pszFilename);
2470	}
2471
2472	if (RT_SUCCESS(rc) && !cbLimit)
2473	{
2474	rc = vdIfIoIntFileSetSize(pImage->pIfIo, pDescFile->pStorage, cbDescriptor);
2475	if (RT_FAILURE(rc))
2476	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error truncating descriptor in '%s'"), pImage->pszFilename);
2477	}
2478
2479	if (RT_SUCCESS(rc))
2480	pImage->Descriptor.fDirty = false;
2481
2482	if (pvDescriptor)
2483	RTMemFree(pvDescriptor);
2484	return rc;
2485
2486	}
2487
2488	/**
2489	* Internal: validate the consistency check values in a binary header.
2490	*/
2491	static int vmdkValidateHeader(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, const SparseExtentHeader *pHeader)
2492	{
2493	int rc = VINF_SUCCESS;
2494	if (RT_LE2H_U32(pHeader->magicNumber) != VMDK_SPARSE_MAGICNUMBER)
2495	{
2496	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrect magic in sparse extent header in '%s'"), pExtent->pszFullname);
2497	return rc;
2498	}
2499	if (RT_LE2H_U32(pHeader->version) != 1 && RT_LE2H_U32(pHeader->version) != 3)
2500	{
2501	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_UNSUPPORTED_VERSION, RT_SRC_POS, N_("VMDK: incorrect version in sparse extent header in '%s', not a VMDK 1.0/1.1 conforming file"), pExtent->pszFullname);
2502	return rc;
2503	}
2504	if ( (RT_LE2H_U32(pHeader->flags) & 1)
2505	&& ( pHeader->singleEndLineChar != '\n'
2506	\|\| pHeader->nonEndLineChar != ' '
2507	\|\| pHeader->doubleEndLineChar1 != '\r'
2508	\|\| pHeader->doubleEndLineChar2 != '\n') )
2509	{
2510	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: corrupted by CR/LF translation in '%s'"), pExtent->pszFullname);
2511	return rc;
2512	}
2513	return rc;
2514	}
2515
2516	/**
2517	* Internal: read metadata belonging to an extent with binary header, i.e.
2518	* as found in monolithic files.
2519	*/
2520	static int vmdkReadBinaryMetaExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
2521	bool fMagicAlreadyRead)
2522	{
2523	SparseExtentHeader Header;
2524	uint64_t cSectorsPerGDE;
2525	uint64_t cbFile = 0;
2526	int rc;
2527
2528	if (!fMagicAlreadyRead)
2529	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage, 0,
2530	&Header, sizeof(Header));
2531	else
2532	{
2533	Header.magicNumber = RT_H2LE_U32(VMDK_SPARSE_MAGICNUMBER);
2534	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
2535	RT_OFFSETOF(SparseExtentHeader, version),
2536	&Header.version,
2537	sizeof(Header)
2538	- RT_OFFSETOF(SparseExtentHeader, version));
2539	}
2540	AssertRC(rc);
2541	if (RT_FAILURE(rc))
2542	{
2543	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading extent header in '%s'"), pExtent->pszFullname);
2544	rc = VERR_VD_VMDK_INVALID_HEADER;
2545	goto out;
2546	}
2547	rc = vmdkValidateHeader(pImage, pExtent, &Header);
2548	if (RT_FAILURE(rc))
2549	goto out;
2550
2551	if ( (RT_LE2H_U32(Header.flags) & RT_BIT(17))
2552	&& RT_LE2H_U64(Header.gdOffset) == VMDK_GD_AT_END)
2553	pExtent->fFooter = true;
2554
2555	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2556	\|\| ( pExtent->fFooter
2557	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2558	{
2559	rc = vdIfIoIntFileGetSize(pImage->pIfIo, pExtent->pFile->pStorage, &cbFile);
2560	AssertRC(rc);
2561	if (RT_FAILURE(rc))
2562	{
2563	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot get size of '%s'"), pExtent->pszFullname);
2564	goto out;
2565	}
2566	}
2567
2568	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
2569	pExtent->uAppendPosition = RT_ALIGN_64(cbFile, 512);
2570
2571	if ( pExtent->fFooter
2572	&& ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2573	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2574	{
2575	/* Read the footer, which comes before the end-of-stream marker. */
2576	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
2577	cbFile - 2*512, &Header,
2578	sizeof(Header));
2579	AssertRC(rc);
2580	if (RT_FAILURE(rc))
2581	{
2582	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading extent footer in '%s'"), pExtent->pszFullname);
2583	rc = VERR_VD_VMDK_INVALID_HEADER;
2584	goto out;
2585	}
2586	rc = vmdkValidateHeader(pImage, pExtent, &Header);
2587	if (RT_FAILURE(rc))
2588	goto out;
2589	/* Prohibit any writes to this extent. */
2590	pExtent->uAppendPosition = 0;
2591	}
2592
2593	pExtent->uVersion = RT_LE2H_U32(Header.version);
2594	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE; /* Just dummy value, changed later. */
2595	pExtent->cSectors = RT_LE2H_U64(Header.capacity);
2596	pExtent->cSectorsPerGrain = RT_LE2H_U64(Header.grainSize);
2597	pExtent->uDescriptorSector = RT_LE2H_U64(Header.descriptorOffset);
2598	pExtent->cDescriptorSectors = RT_LE2H_U64(Header.descriptorSize);
2599	if (pExtent->uDescriptorSector && !pExtent->cDescriptorSectors)
2600	{
2601	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: inconsistent embedded descriptor config in '%s'"), pExtent->pszFullname);
2602	goto out;
2603	}
2604	pExtent->cGTEntries = RT_LE2H_U32(Header.numGTEsPerGT);
2605	if (RT_LE2H_U32(Header.flags) & RT_BIT(1))
2606	{
2607	pExtent->uSectorRGD = RT_LE2H_U64(Header.rgdOffset);
2608	pExtent->uSectorGD = RT_LE2H_U64(Header.gdOffset);
2609	}
2610	else
2611	{
2612	pExtent->uSectorGD = RT_LE2H_U64(Header.gdOffset);
2613	pExtent->uSectorRGD = 0;
2614	}
2615	if ( ( pExtent->uSectorGD == VMDK_GD_AT_END
2616	\|\| pExtent->uSectorRGD == VMDK_GD_AT_END)
2617	&& ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2618	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2619	{
2620	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: cannot resolve grain directory offset in '%s'"), pExtent->pszFullname);
2621	goto out;
2622	}
2623	pExtent->cOverheadSectors = RT_LE2H_U64(Header.overHead);
2624	pExtent->fUncleanShutdown = !!Header.uncleanShutdown;
2625	pExtent->uCompression = RT_LE2H_U16(Header.compressAlgorithm);
2626	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
2627	if (!cSectorsPerGDE \|\| cSectorsPerGDE > UINT32_MAX)
2628	{
2629	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrect grain directory size in '%s'"), pExtent->pszFullname);
2630	goto out;
2631	}
2632	pExtent->cSectorsPerGDE = cSectorsPerGDE;
2633	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
2634
2635	/* Fix up the number of descriptor sectors, as some flat images have
2636	* really just one, and this causes failures when inserting the UUID
2637	* values and other extra information. */
2638	if (pExtent->cDescriptorSectors != 0 && pExtent->cDescriptorSectors < 4)
2639	{
2640	/* Do it the easy way - just fix it for flat images which have no
2641	* other complicated metadata which needs space too. */
2642	if ( pExtent->uDescriptorSector + 4 < pExtent->cOverheadSectors
2643	&& pExtent->cGTEntries * pExtent->cGDEntries == 0)
2644	pExtent->cDescriptorSectors = 4;
2645	}
2646
2647	out:
2648	if (RT_FAILURE(rc))
2649	vmdkFreeExtentData(pImage, pExtent, false);
2650
2651	return rc;
2652	}
2653
2654	/**
2655	* Internal: read additional metadata belonging to an extent. For those
2656	* extents which have no additional metadata just verify the information.
2657	*/
2658	static int vmdkReadMetaExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
2659	{
2660	int rc = VINF_SUCCESS;
2661
2662	/* disabled the check as there are too many truncated vmdk images out there */
2663	#ifdef VBOX_WITH_VMDK_STRICT_SIZE_CHECK
2664	uint64_t cbExtentSize;
2665	/* The image must be a multiple of a sector in size and contain the data
2666	* area (flat images only). If not, it means the image is at least
2667	* truncated, or even seriously garbled. */
2668	rc = vdIfIoIntFileGetSize(pImage->pIfIo, pExtent->pFile->pStorage, &cbExtentSize);
2669	if (RT_FAILURE(rc))
2670	{
2671	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting size in '%s'"), pExtent->pszFullname);
2672	goto out;
2673	}
2674	if ( cbExtentSize != RT_ALIGN_64(cbExtentSize, 512)
2675	&& (pExtent->enmType != VMDKETYPE_FLAT \|\| pExtent->cNominalSectors + pExtent->uSectorOffset > VMDK_BYTE2SECTOR(cbExtentSize)))
2676	{
2677	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: file size is not a multiple of 512 in '%s', file is truncated or otherwise garbled"), pExtent->pszFullname);
2678	goto out;
2679	}
2680	#endif /* VBOX_WITH_VMDK_STRICT_SIZE_CHECK */
2681	if (pExtent->enmType != VMDKETYPE_HOSTED_SPARSE)
2682	goto out;
2683
2684	/* The spec says that this must be a power of two and greater than 8,
2685	* but probably they meant not less than 8. */
2686	if ( (pExtent->cSectorsPerGrain & (pExtent->cSectorsPerGrain - 1))
2687	\|\| pExtent->cSectorsPerGrain < 8)
2688	{
2689	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: invalid extent grain size %u in '%s'"), pExtent->cSectorsPerGrain, pExtent->pszFullname);
2690	goto out;
2691	}
2692
2693	/* This code requires that a grain table must hold a power of two multiple
2694	* of the number of entries per GT cache entry. */
2695	if ( (pExtent->cGTEntries & (pExtent->cGTEntries - 1))
2696	\|\| pExtent->cGTEntries < VMDK_GT_CACHELINE_SIZE)
2697	{
2698	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: grain table cache size problem in '%s'"), pExtent->pszFullname);
2699	goto out;
2700	}
2701
2702	rc = vmdkAllocStreamBuffers(pImage, pExtent);
2703	if (RT_FAILURE(rc))
2704	goto out;
2705
2706	/* Prohibit any writes to this streamOptimized extent. */
2707	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2708	pExtent->uAppendPosition = 0;
2709
2710	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2711	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2712	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
2713	rc = vmdkReadGrainDirectory(pImage, pExtent);
2714	else
2715	{
2716	pExtent->uGrainSectorAbs = pExtent->cOverheadSectors;
2717	pExtent->cbGrainStreamRead = 0;
2718	}
2719
2720	out:
2721	if (RT_FAILURE(rc))
2722	vmdkFreeExtentData(pImage, pExtent, false);
2723
2724	return rc;
2725	}
2726
2727	/**
2728	* Internal: write/update the metadata for a sparse extent.
2729	*/
2730	static int vmdkWriteMetaSparseExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
2731	uint64_t uOffset, PVDIOCTX pIoCtx)
2732	{
2733	SparseExtentHeader Header;
2734
2735	memset(&Header, '\0', sizeof(Header));
2736	Header.magicNumber = RT_H2LE_U32(VMDK_SPARSE_MAGICNUMBER);
2737	Header.version = RT_H2LE_U32(pExtent->uVersion);
2738	Header.flags = RT_H2LE_U32(RT_BIT(0));
2739	if (pExtent->pRGD)
2740	Header.flags \|= RT_H2LE_U32(RT_BIT(1));
2741	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2742	Header.flags \|= RT_H2LE_U32(RT_BIT(16) \| RT_BIT(17));
2743	Header.capacity = RT_H2LE_U64(pExtent->cSectors);
2744	Header.grainSize = RT_H2LE_U64(pExtent->cSectorsPerGrain);
2745	Header.descriptorOffset = RT_H2LE_U64(pExtent->uDescriptorSector);
2746	Header.descriptorSize = RT_H2LE_U64(pExtent->cDescriptorSectors);
2747	Header.numGTEsPerGT = RT_H2LE_U32(pExtent->cGTEntries);
2748	if (pExtent->fFooter && uOffset == 0)
2749	{
2750	if (pExtent->pRGD)
2751	{
2752	Assert(pExtent->uSectorRGD);
2753	Header.rgdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2754	Header.gdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2755	}
2756	else
2757	{
2758	Header.gdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2759	}
2760	}
2761	else
2762	{
2763	if (pExtent->pRGD)
2764	{
2765	Assert(pExtent->uSectorRGD);
2766	Header.rgdOffset = RT_H2LE_U64(pExtent->uSectorRGD);
2767	Header.gdOffset = RT_H2LE_U64(pExtent->uSectorGD);
2768	}
2769	else
2770	{
2771	Header.gdOffset = RT_H2LE_U64(pExtent->uSectorGD);
2772	}
2773	}
2774	Header.overHead = RT_H2LE_U64(pExtent->cOverheadSectors);
2775	Header.uncleanShutdown = pExtent->fUncleanShutdown;
2776	Header.singleEndLineChar = '\n';
2777	Header.nonEndLineChar = ' ';
2778	Header.doubleEndLineChar1 = '\r';
2779	Header.doubleEndLineChar2 = '\n';
2780	Header.compressAlgorithm = RT_H2LE_U16(pExtent->uCompression);
2781
2782	int rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
2783	uOffset, &Header, sizeof(Header),
2784	pIoCtx, NULL, NULL);
2785	if (RT_FAILURE(rc) && (rc != VERR_VD_ASYNC_IO_IN_PROGRESS))
2786	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error writing extent header in '%s'"), pExtent->pszFullname);
2787	return rc;
2788	}
2789
2790	#ifdef VBOX_WITH_VMDK_ESX
2791	/**
2792	* Internal: unused code to read the metadata of a sparse ESX extent.
2793	*
2794	* Such extents never leave ESX server, so this isn't ever used.
2795	*/
2796	static int vmdkReadMetaESXSparseExtent(PVMDKEXTENT pExtent)
2797	{
2798	COWDisk_Header Header;
2799	uint64_t cSectorsPerGDE;
2800
2801	int rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage, 0,
2802	&Header, sizeof(Header));
2803	AssertRC(rc);
2804	if (RT_FAILURE(rc))
2805	{
2806	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading ESX sparse extent header in '%s'"), pExtent->pszFullname);
2807	rc = VERR_VD_VMDK_INVALID_HEADER;
2808	goto out;
2809	}
2810	if ( RT_LE2H_U32(Header.magicNumber) != VMDK_ESX_SPARSE_MAGICNUMBER
2811	\|\| RT_LE2H_U32(Header.version) != 1
2812	\|\| RT_LE2H_U32(Header.flags) != 3)
2813	{
2814	rc = VERR_VD_VMDK_INVALID_HEADER;
2815	goto out;
2816	}
2817	pExtent->enmType = VMDKETYPE_ESX_SPARSE;
2818	pExtent->cSectors = RT_LE2H_U32(Header.numSectors);
2819	pExtent->cSectorsPerGrain = RT_LE2H_U32(Header.grainSize);
2820	/* The spec says that this must be between 1 sector and 1MB. This code
2821	* assumes it's a power of two, so check that requirement, too. */
2822	if ( (pExtent->cSectorsPerGrain & (pExtent->cSectorsPerGrain - 1))
2823	\|\| pExtent->cSectorsPerGrain == 0
2824	\|\| pExtent->cSectorsPerGrain > 2048)
2825	{
2826	rc = VERR_VD_VMDK_INVALID_HEADER;
2827	goto out;
2828	}
2829	pExtent->uDescriptorSector = 0;
2830	pExtent->cDescriptorSectors = 0;
2831	pExtent->uSectorGD = RT_LE2H_U32(Header.gdOffset);
2832	pExtent->uSectorRGD = 0;
2833	pExtent->cOverheadSectors = 0;
2834	pExtent->cGTEntries = 4096;
2835	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
2836	if (!cSectorsPerGDE \|\| cSectorsPerGDE > UINT32_MAX)
2837	{
2838	rc = VERR_VD_VMDK_INVALID_HEADER;
2839	goto out;
2840	}
2841	pExtent->cSectorsPerGDE = cSectorsPerGDE;
2842	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
2843	if (pExtent->cGDEntries != RT_LE2H_U32(Header.numGDEntries))
2844	{
2845	/* Inconsistency detected. Computed number of GD entries doesn't match
2846	* stored value. Better be safe than sorry. */
2847	rc = VERR_VD_VMDK_INVALID_HEADER;
2848	goto out;
2849	}
2850	pExtent->uFreeSector = RT_LE2H_U32(Header.freeSector);
2851	pExtent->fUncleanShutdown = !!Header.uncleanShutdown;
2852
2853	rc = vmdkReadGrainDirectory(pImage, pExtent);
2854
2855	out:
2856	if (RT_FAILURE(rc))
2857	vmdkFreeExtentData(pImage, pExtent, false);
2858
2859	return rc;
2860	}
2861	#endif /* VBOX_WITH_VMDK_ESX */
2862
2863	/**
2864	* Internal: free the buffers used for streamOptimized images.
2865	*/
2866	static void vmdkFreeStreamBuffers(PVMDKEXTENT pExtent)
2867	{
2868	if (pExtent->pvCompGrain)
2869	{
2870	RTMemFree(pExtent->pvCompGrain);
2871	pExtent->pvCompGrain = NULL;
2872	}
2873	if (pExtent->pvGrain)
2874	{
2875	RTMemFree(pExtent->pvGrain);
2876	pExtent->pvGrain = NULL;
2877	}
2878	}
2879
2880	/**
2881	* Internal: free the memory used by the extent data structure, optionally
2882	* deleting the referenced files.
2883	*
2884	* @returns VBox status code.
2885	* @param pImage Pointer to the image instance data.
2886	* @param pExtent The extent to free.
2887	* @param fDelete Flag whether to delete the backing storage.
2888	*/
2889	static int vmdkFreeExtentData(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
2890	bool fDelete)
2891	{
2892	int rc = VINF_SUCCESS;
2893
2894	vmdkFreeGrainDirectory(pExtent);
2895	if (pExtent->pDescData)
2896	{
2897	RTMemFree(pExtent->pDescData);
2898	pExtent->pDescData = NULL;
2899	}
2900	if (pExtent->pFile != NULL)
2901	{
2902	/* Do not delete raw extents, these have full and base names equal. */
2903	rc = vmdkFileClose(pImage, &pExtent->pFile,
2904	fDelete
2905	&& pExtent->pszFullname
2906	&& pExtent->pszBasename
2907	&& strcmp(pExtent->pszFullname, pExtent->pszBasename));
2908	}
2909	if (pExtent->pszBasename)
2910	{
2911	RTMemTmpFree((void *)pExtent->pszBasename);
2912	pExtent->pszBasename = NULL;
2913	}
2914	if (pExtent->pszFullname)
2915	{
2916	RTStrFree((char )(void )pExtent->pszFullname);
2917	pExtent->pszFullname = NULL;
2918	}
2919	vmdkFreeStreamBuffers(pExtent);
2920
2921	return rc;
2922	}
2923
2924	/**
2925	* Internal: allocate grain table cache if necessary for this image.
2926	*/
2927	static int vmdkAllocateGrainTableCache(PVMDKIMAGE pImage)
2928	{
2929	PVMDKEXTENT pExtent;
2930
2931	/* Allocate grain table cache if any sparse extent is present. */
2932	for (unsigned i = 0; i < pImage->cExtents; i++)
2933	{
2934	pExtent = &pImage->pExtents[i];
2935	if ( pExtent->enmType == VMDKETYPE_HOSTED_SPARSE
2936	#ifdef VBOX_WITH_VMDK_ESX
2937	\|\| pExtent->enmType == VMDKETYPE_ESX_SPARSE
2938	#endif /* VBOX_WITH_VMDK_ESX */
2939	)
2940	{
2941	/* Allocate grain table cache. */
2942	pImage->pGTCache = (PVMDKGTCACHE)RTMemAllocZ(sizeof(VMDKGTCACHE));
2943	if (!pImage->pGTCache)
2944	return VERR_NO_MEMORY;
2945	for (unsigned j = 0; j < VMDK_GT_CACHE_SIZE; j++)
2946	{
2947	PVMDKGTCACHEENTRY pGCE = &pImage->pGTCache->aGTCache[j];
2948	pGCE->uExtent = UINT32_MAX;
2949	}
2950	pImage->pGTCache->cEntries = VMDK_GT_CACHE_SIZE;
2951	break;
2952	}
2953	}
2954
2955	return VINF_SUCCESS;
2956	}
2957
2958	/**
2959	* Internal: allocate the given number of extents.
2960	*/
2961	static int vmdkCreateExtents(PVMDKIMAGE pImage, unsigned cExtents)
2962	{
2963	int rc = VINF_SUCCESS;
2964	PVMDKEXTENT pExtents = (PVMDKEXTENT)RTMemAllocZ(cExtents * sizeof(VMDKEXTENT));
2965	if (pExtents)
2966	{
2967	for (unsigned i = 0; i < cExtents; i++)
2968	{
2969	pExtents[i].pFile = NULL;
2970	pExtents[i].pszBasename = NULL;
2971	pExtents[i].pszFullname = NULL;
2972	pExtents[i].pGD = NULL;
2973	pExtents[i].pRGD = NULL;
2974	pExtents[i].pDescData = NULL;
2975	pExtents[i].uVersion = 1;
2976	pExtents[i].uCompression = VMDK_COMPRESSION_NONE;
2977	pExtents[i].uExtent = i;
2978	pExtents[i].pImage = pImage;
2979	}
2980	pImage->pExtents = pExtents;
2981	pImage->cExtents = cExtents;
2982	}
2983	else
2984	rc = VERR_NO_MEMORY;
2985
2986	return rc;
2987	}
2988
2989	/**
2990	* Internal: Open an image, constructing all necessary data structures.
2991	*/
2992	static int vmdkOpenImage(PVMDKIMAGE pImage, unsigned uOpenFlags)
2993	{
2994	int rc;
2995	uint32_t u32Magic;
2996	PVMDKFILE pFile;
2997	PVMDKEXTENT pExtent;
2998
2999	pImage->uOpenFlags = uOpenFlags;
3000
3001	pImage->pIfError = VDIfErrorGet(pImage->pVDIfsDisk);
3002	pImage->pIfIo = VDIfIoIntGet(pImage->pVDIfsImage);
3003	AssertPtrReturn(pImage->pIfIo, VERR_INVALID_PARAMETER);
3004
3005	/*
3006	* Open the image.
3007	* We don't have to check for asynchronous access because
3008	* we only support raw access and the opened file is a description
3009	* file were no data is stored.
3010	*/
3011
3012	rc = vmdkFileOpen(pImage, &pFile, pImage->pszFilename,
3013	VDOpenFlagsToFileOpenFlags(uOpenFlags, false /* fCreate */));
3014	if (RT_FAILURE(rc))
3015	{
3016	/* Do NOT signal an appropriate error here, as the VD layer has the
3017	* choice of retrying the open if it failed. */
3018	goto out;
3019	}
3020	pImage->pFile = pFile;
3021
3022	/* Read magic (if present). */
3023	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pFile->pStorage, 0,
3024	&u32Magic, sizeof(u32Magic));
3025	if (RT_FAILURE(rc))
3026	{
3027	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading the magic number in '%s'"), pImage->pszFilename);
3028	rc = VERR_VD_VMDK_INVALID_HEADER;
3029	goto out;
3030	}
3031
3032	/* Handle the file according to its magic number. */
3033	if (RT_LE2H_U32(u32Magic) == VMDK_SPARSE_MAGICNUMBER)
3034	{
3035	/* It's a hosted single-extent image. */
3036	rc = vmdkCreateExtents(pImage, 1);
3037	if (RT_FAILURE(rc))
3038	goto out;
3039	/* The opened file is passed to the extent. No separate descriptor
3040	* file, so no need to keep anything open for the image. */
3041	pExtent = &pImage->pExtents[0];
3042	pExtent->pFile = pFile;
3043	pImage->pFile = NULL;
3044	pExtent->pszFullname = RTPathAbsDup(pImage->pszFilename);
3045	if (!pExtent->pszFullname)
3046	{
3047	rc = VERR_NO_MEMORY;
3048	goto out;
3049	}
3050	rc = vmdkReadBinaryMetaExtent(pImage, pExtent, true /* fMagicAlreadyRead */);
3051	if (RT_FAILURE(rc))
3052	goto out;
3053
3054	/* As we're dealing with a monolithic image here, there must
3055	* be a descriptor embedded in the image file. */
3056	if (!pExtent->uDescriptorSector \|\| !pExtent->cDescriptorSectors)
3057	{
3058	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: monolithic image without descriptor in '%s'"), pImage->pszFilename);
3059	goto out;
3060	}
3061	/* HACK: extend the descriptor if it is unusually small and it fits in
3062	* the unused space after the image header. Allows opening VMDK files
3063	* with extremely small descriptor in read/write mode. */
3064	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3065	&& pExtent->cDescriptorSectors < 3
3066	&& (int64_t)pExtent->uSectorGD - pExtent->uDescriptorSector >= 4
3067	&& (!pExtent->uSectorRGD \|\| (int64_t)pExtent->uSectorRGD - pExtent->uDescriptorSector >= 4))
3068	{
3069	pExtent->cDescriptorSectors = 4;
3070	pExtent->fMetaDirty = true;
3071	}
3072	/* Read the descriptor from the extent. */
3073	pExtent->pDescData = (char *)RTMemAllocZ(VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
3074	if (!pExtent->pDescData)
3075	{
3076	rc = VERR_NO_MEMORY;
3077	goto out;
3078	}
3079	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
3080	VMDK_SECTOR2BYTE(pExtent->uDescriptorSector),
3081	pExtent->pDescData,
3082	VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
3083	AssertRC(rc);
3084	if (RT_FAILURE(rc))
3085	{
3086	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: read error for descriptor in '%s'"), pExtent->pszFullname);
3087	goto out;
3088	}
3089
3090	rc = vmdkParseDescriptor(pImage, pExtent->pDescData,
3091	VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
3092	if (RT_FAILURE(rc))
3093	goto out;
3094
3095	if ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
3096	&& uOpenFlags & VD_OPEN_FLAGS_ASYNC_IO)
3097	{
3098	rc = VERR_NOT_SUPPORTED;
3099	goto out;
3100	}
3101
3102	rc = vmdkReadMetaExtent(pImage, pExtent);
3103	if (RT_FAILURE(rc))
3104	goto out;
3105
3106	/* Mark the extent as unclean if opened in read-write mode. */
3107	if ( !(uOpenFlags & VD_OPEN_FLAGS_READONLY)
3108	&& !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
3109	{
3110	pExtent->fUncleanShutdown = true;
3111	pExtent->fMetaDirty = true;
3112	}
3113	}
3114	else
3115	{
3116	/* Allocate at least 10K, and make sure that there is 5K free space
3117	* in case new entries need to be added to the descriptor. Never
3118	* allocate more than 128K, because that's no valid descriptor file
3119	* and will result in the correct "truncated read" error handling. */
3120	uint64_t cbFileSize;
3121	rc = vdIfIoIntFileGetSize(pImage->pIfIo, pFile->pStorage, &cbFileSize);
3122	if (RT_FAILURE(rc))
3123	goto out;
3124
3125	/* If the descriptor file is shorter than 50 bytes it can't be valid. */
3126	if (cbFileSize < 50)
3127	{
3128	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor in '%s' is too short"), pImage->pszFilename);
3129	goto out;
3130	}
3131
3132	uint64_t cbSize = cbFileSize;
3133	if (cbSize % VMDK_SECTOR2BYTE(10))
3134	cbSize += VMDK_SECTOR2BYTE(20) - cbSize % VMDK_SECTOR2BYTE(10);
3135	else
3136	cbSize += VMDK_SECTOR2BYTE(10);
3137	cbSize = RT_MIN(cbSize, _128K);
3138	pImage->cbDescAlloc = RT_MAX(VMDK_SECTOR2BYTE(20), cbSize);
3139	pImage->pDescData = (char *)RTMemAllocZ(pImage->cbDescAlloc);
3140	if (!pImage->pDescData)
3141	{
3142	rc = VERR_NO_MEMORY;
3143	goto out;
3144	}
3145
3146	/* Don't reread the place where the magic would live in a sparse
3147	* image if it's a descriptor based one. */
3148	memcpy(pImage->pDescData, &u32Magic, sizeof(u32Magic));
3149	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pFile->pStorage, sizeof(u32Magic),
3150	pImage->pDescData + sizeof(u32Magic),
3151	RT_MIN(pImage->cbDescAlloc - sizeof(u32Magic),
3152	cbFileSize - sizeof(u32Magic)));
3153	if (RT_FAILURE(rc))
3154	{
3155	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: read error for descriptor in '%s'"), pImage->pszFilename);
3156	goto out;
3157	}
3158
3159	#if 0 /** @todo: Revisit */
3160	cbRead += sizeof(u32Magic);
3161	if (cbRead == pImage->cbDescAlloc)
3162	{
3163	/* Likely the read is truncated. Better fail a bit too early
3164	* (normally the descriptor is much smaller than our buffer). */
3165	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: cannot read descriptor in '%s'"), pImage->pszFilename);
3166	goto out;
3167	}
3168	#endif
3169
3170	rc = vmdkParseDescriptor(pImage, pImage->pDescData,
3171	pImage->cbDescAlloc);
3172	if (RT_FAILURE(rc))
3173	goto out;
3174
3175	/*
3176	* We have to check for the asynchronous open flag. The
3177	* extents are parsed and the type of all are known now.
3178	* Check if every extent is either FLAT or ZERO.
3179	*/
3180	if (uOpenFlags & VD_OPEN_FLAGS_ASYNC_IO)
3181	{
3182	unsigned cFlatExtents = 0;
3183
3184	for (unsigned i = 0; i < pImage->cExtents; i++)
3185	{
3186	pExtent = &pImage->pExtents[i];
3187
3188	if (( pExtent->enmType != VMDKETYPE_FLAT
3189	&& pExtent->enmType != VMDKETYPE_ZERO
3190	&& pExtent->enmType != VMDKETYPE_VMFS)
3191	\|\| ((pImage->pExtents[i].enmType == VMDKETYPE_FLAT) && (cFlatExtents > 0)))
3192	{
3193	/*
3194	* Opened image contains at least one none flat or zero extent.
3195	* Return error but don't set error message as the caller
3196	* has the chance to open in non async I/O mode.
3197	*/
3198	rc = VERR_NOT_SUPPORTED;
3199	goto out;
3200	}
3201	if (pExtent->enmType == VMDKETYPE_FLAT)
3202	cFlatExtents++;
3203	}
3204	}
3205
3206	for (unsigned i = 0; i < pImage->cExtents; i++)
3207	{
3208	pExtent = &pImage->pExtents[i];
3209
3210	if (pExtent->pszBasename)
3211	{
3212	/* Hack to figure out whether the specified name in the
3213	* extent descriptor is absolute. Doesn't always work, but
3214	* should be good enough for now. */
3215	char *pszFullname;
3216	/** @todo implement proper path absolute check. */
3217	if (pExtent->pszBasename[0] == RTPATH_SLASH)
3218	{
3219	pszFullname = RTStrDup(pExtent->pszBasename);
3220	if (!pszFullname)
3221	{
3222	rc = VERR_NO_MEMORY;
3223	goto out;
3224	}
3225	}
3226	else
3227	{
3228	char *pszDirname = RTStrDup(pImage->pszFilename);
3229	if (!pszDirname)
3230	{
3231	rc = VERR_NO_MEMORY;
3232	goto out;
3233	}
3234	RTPathStripFilename(pszDirname);
3235	pszFullname = RTPathJoinA(pszDirname, pExtent->pszBasename);
3236	RTStrFree(pszDirname);
3237	if (!pszFullname)
3238	{
3239	rc = VERR_NO_STR_MEMORY;
3240	goto out;
3241	}
3242	}
3243	pExtent->pszFullname = pszFullname;
3244	}
3245	else
3246	pExtent->pszFullname = NULL;
3247
3248	switch (pExtent->enmType)
3249	{
3250	case VMDKETYPE_HOSTED_SPARSE:
3251	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszFullname,
3252	VDOpenFlagsToFileOpenFlags(uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
3253	false /* fCreate */));
3254	if (RT_FAILURE(rc))
3255	{
3256	/* Do NOT signal an appropriate error here, as the VD
3257	* layer has the choice of retrying the open if it
3258	* failed. */
3259	goto out;
3260	}
3261	rc = vmdkReadBinaryMetaExtent(pImage, pExtent,
3262	false /* fMagicAlreadyRead */);
3263	if (RT_FAILURE(rc))
3264	goto out;
3265	rc = vmdkReadMetaExtent(pImage, pExtent);
3266	if (RT_FAILURE(rc))
3267	goto out;
3268
3269	/* Mark extent as unclean if opened in read-write mode. */
3270	if (!(uOpenFlags & VD_OPEN_FLAGS_READONLY))
3271	{
3272	pExtent->fUncleanShutdown = true;
3273	pExtent->fMetaDirty = true;
3274	}
3275	break;
3276	case VMDKETYPE_VMFS:
3277	case VMDKETYPE_FLAT:
3278	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszFullname,
3279	VDOpenFlagsToFileOpenFlags(uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
3280	false /* fCreate */));
3281	if (RT_FAILURE(rc))
3282	{
3283	/* Do NOT signal an appropriate error here, as the VD
3284	* layer has the choice of retrying the open if it
3285	* failed. */
3286	goto out;
3287	}
3288	break;
3289	case VMDKETYPE_ZERO:
3290	/* Nothing to do. */
3291	break;
3292	default:
3293	AssertMsgFailed(("unknown vmdk extent type %d\n", pExtent->enmType));
3294	}
3295	}
3296	}
3297
3298	/* Make sure this is not reached accidentally with an error status. */
3299	AssertRC(rc);
3300
3301	/* Determine PCHS geometry if not set. */
3302	if (pImage->PCHSGeometry.cCylinders == 0)
3303	{
3304	uint64_t cCylinders = VMDK_BYTE2SECTOR(pImage->cbSize)
3305	/ pImage->PCHSGeometry.cHeads
3306	/ pImage->PCHSGeometry.cSectors;
3307	pImage->PCHSGeometry.cCylinders = (unsigned)RT_MIN(cCylinders, 16383);
3308	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3309	&& !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
3310	{
3311	rc = vmdkDescSetPCHSGeometry(pImage, &pImage->PCHSGeometry);
3312	AssertRC(rc);
3313	}
3314	}
3315
3316	/* Update the image metadata now in case has changed. */
3317	rc = vmdkFlushImage(pImage, NULL);
3318	if (RT_FAILURE(rc))
3319	goto out;
3320
3321	/* Figure out a few per-image constants from the extents. */
3322	pImage->cbSize = 0;
3323	for (unsigned i = 0; i < pImage->cExtents; i++)
3324	{
3325	pExtent = &pImage->pExtents[i];
3326	if ( pExtent->enmType == VMDKETYPE_HOSTED_SPARSE
3327	#ifdef VBOX_WITH_VMDK_ESX
3328	\|\| pExtent->enmType == VMDKETYPE_ESX_SPARSE
3329	#endif /* VBOX_WITH_VMDK_ESX */
3330	)
3331	{
3332	/* Here used to be a check whether the nominal size of an extent
3333	* is a multiple of the grain size. The spec says that this is
3334	* always the case, but unfortunately some files out there in the
3335	* wild violate the spec (e.g. ReactOS 0.3.1). */
3336	}
3337	pImage->cbSize += VMDK_SECTOR2BYTE(pExtent->cNominalSectors);
3338	}
3339
3340	for (unsigned i = 0; i < pImage->cExtents; i++)
3341	{
3342	pExtent = &pImage->pExtents[i];
3343	if ( pImage->pExtents[i].enmType == VMDKETYPE_FLAT
3344	\|\| pImage->pExtents[i].enmType == VMDKETYPE_ZERO)
3345	{
3346	pImage->uImageFlags \|= VD_IMAGE_FLAGS_FIXED;
3347	break;
3348	}
3349	}
3350
3351	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
3352	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3353	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
3354	rc = vmdkAllocateGrainTableCache(pImage);
3355
3356	out:
3357	if (RT_FAILURE(rc))
3358	vmdkFreeImage(pImage, false);
3359	return rc;
3360	}
3361
3362	/**
3363	* Internal: create VMDK images for raw disk/partition access.
3364	*/
3365	static int vmdkCreateRawImage(PVMDKIMAGE pImage, const PVBOXHDDRAW pRaw,
3366	uint64_t cbSize)
3367	{
3368	int rc = VINF_SUCCESS;
3369	PVMDKEXTENT pExtent;
3370
3371	if (pRaw->uFlags & VBOXHDDRAW_DISK)
3372	{
3373	/* Full raw disk access. This requires setting up a descriptor
3374	* file and open the (flat) raw disk. */
3375	rc = vmdkCreateExtents(pImage, 1);
3376	if (RT_FAILURE(rc))
3377	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
3378	pExtent = &pImage->pExtents[0];
3379	/* Create raw disk descriptor file. */
3380	rc = vmdkFileOpen(pImage, &pImage->pFile, pImage->pszFilename,
3381	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
3382	true /* fCreate */));
3383	if (RT_FAILURE(rc))
3384	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pImage->pszFilename);
3385
3386	/* Set up basename for extent description. Cannot use StrDup. */
3387	size_t cbBasename = strlen(pRaw->pszRawDisk) + 1;
3388	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
3389	if (!pszBasename)
3390	return VERR_NO_MEMORY;
3391	memcpy(pszBasename, pRaw->pszRawDisk, cbBasename);
3392	pExtent->pszBasename = pszBasename;
3393	/* For raw disks the full name is identical to the base name. */
3394	pExtent->pszFullname = RTStrDup(pszBasename);
3395	if (!pExtent->pszFullname)
3396	return VERR_NO_MEMORY;
3397	pExtent->enmType = VMDKETYPE_FLAT;
3398	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
3399	pExtent->uSectorOffset = 0;
3400	pExtent->enmAccess = (pRaw->uFlags & VBOXHDDRAW_READONLY) ? VMDKACCESS_READONLY : VMDKACCESS_READWRITE;
3401	pExtent->fMetaDirty = false;
3402
3403	/* Open flat image, the raw disk. */
3404	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszFullname,
3405	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
3406	false /* fCreate */));
3407	if (RT_FAILURE(rc))
3408	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not open raw disk file '%s'"), pExtent->pszFullname);
3409	}
3410	else
3411	{
3412	/* Raw partition access. This requires setting up a descriptor
3413	* file, write the partition information to a flat extent and
3414	* open all the (flat) raw disk partitions. */
3415
3416	/* First pass over the partition data areas to determine how many
3417	* extents we need. One data area can require up to 2 extents, as
3418	* it might be necessary to skip over unpartitioned space. */
3419	unsigned cExtents = 0;
3420	uint64_t uStart = 0;
3421	for (unsigned i = 0; i < pRaw->cPartDescs; i++)
3422	{
3423	PVBOXHDDRAWPARTDESC pPart = &pRaw->pPartDescs[i];
3424	if (uStart > pPart->uStart)
3425	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS, N_("VMDK: incorrect partition data area ordering set up by the caller in '%s'"), pImage->pszFilename);
3426
3427	if (uStart < pPart->uStart)
3428	cExtents++;
3429	uStart = pPart->uStart + pPart->cbData;
3430	cExtents++;
3431	}
3432	/* Another extent for filling up the rest of the image. */
3433	if (uStart != cbSize)
3434	cExtents++;
3435
3436	rc = vmdkCreateExtents(pImage, cExtents);
3437	if (RT_FAILURE(rc))
3438	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
3439
3440	/* Create raw partition descriptor file. */
3441	rc = vmdkFileOpen(pImage, &pImage->pFile, pImage->pszFilename,
3442	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
3443	true /* fCreate */));
3444	if (RT_FAILURE(rc))
3445	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pImage->pszFilename);
3446
3447	/* Create base filename for the partition table extent. */
3448	/** @todo remove fixed buffer without creating memory leaks. */
3449	char pszPartition[1024];
3450	const char *pszBase = RTPathFilename(pImage->pszFilename);
3451	const char *pszSuff = RTPathSuffix(pszBase);
3452	if (pszSuff == NULL)
3453	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: invalid filename '%s'"), pImage->pszFilename);
3454	char *pszBaseBase = RTStrDup(pszBase);
3455	if (!pszBaseBase)
3456	return VERR_NO_MEMORY;
3457	RTPathStripSuffix(pszBaseBase);
3458	RTStrPrintf(pszPartition, sizeof(pszPartition), "%s-pt%s",
3459	pszBaseBase, pszSuff);
3460	RTStrFree(pszBaseBase);
3461
3462	/* Second pass over the partitions, now define all extents. */
3463	uint64_t uPartOffset = 0;
3464	cExtents = 0;
3465	uStart = 0;
3466	for (unsigned i = 0; i < pRaw->cPartDescs; i++)
3467	{
3468	PVBOXHDDRAWPARTDESC pPart = &pRaw->pPartDescs[i];
3469	pExtent = &pImage->pExtents[cExtents++];
3470
3471	if (uStart < pPart->uStart)
3472	{
3473	pExtent->pszBasename = NULL;
3474	pExtent->pszFullname = NULL;
3475	pExtent->enmType = VMDKETYPE_ZERO;
3476	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->uStart - uStart);
3477	pExtent->uSectorOffset = 0;
3478	pExtent->enmAccess = VMDKACCESS_READWRITE;
3479	pExtent->fMetaDirty = false;
3480	/* go to next extent */
3481	pExtent = &pImage->pExtents[cExtents++];
3482	}
3483	uStart = pPart->uStart + pPart->cbData;
3484
3485	if (pPart->pvPartitionData)
3486	{
3487	/* Set up basename for extent description. Can't use StrDup. */
3488	size_t cbBasename = strlen(pszPartition) + 1;
3489	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
3490	if (!pszBasename)
3491	return VERR_NO_MEMORY;
3492	memcpy(pszBasename, pszPartition, cbBasename);
3493	pExtent->pszBasename = pszBasename;
3494
3495	/* Set up full name for partition extent. */
3496	char *pszDirname = RTStrDup(pImage->pszFilename);
3497	if (!pszDirname)
3498	return VERR_NO_STR_MEMORY;
3499	RTPathStripFilename(pszDirname);
3500	char *pszFullname = RTPathJoinA(pszDirname, pExtent->pszBasename);
3501	RTStrFree(pszDirname);
3502	if (!pszDirname)
3503	return VERR_NO_STR_MEMORY;
3504	pExtent->pszFullname = pszFullname;
3505	pExtent->enmType = VMDKETYPE_FLAT;
3506	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
3507	pExtent->uSectorOffset = uPartOffset;
3508	pExtent->enmAccess = VMDKACCESS_READWRITE;
3509	pExtent->fMetaDirty = false;
3510
3511	/* Create partition table flat image. */
3512	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszFullname,
3513	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
3514	true /* fCreate */));
3515	if (RT_FAILURE(rc))
3516	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new partition data file '%s'"), pExtent->pszFullname);
3517	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
3518	VMDK_SECTOR2BYTE(uPartOffset),
3519	pPart->pvPartitionData,
3520	pPart->cbData);
3521	if (RT_FAILURE(rc))
3522	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not write partition data to '%s'"), pExtent->pszFullname);
3523	uPartOffset += VMDK_BYTE2SECTOR(pPart->cbData);
3524	}
3525	else
3526	{
3527	if (pPart->pszRawDevice)
3528	{
3529	/* Set up basename for extent descr. Can't use StrDup. */
3530	size_t cbBasename = strlen(pPart->pszRawDevice) + 1;
3531	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
3532	if (!pszBasename)
3533	return VERR_NO_MEMORY;
3534	memcpy(pszBasename, pPart->pszRawDevice, cbBasename);
3535	pExtent->pszBasename = pszBasename;
3536	/* For raw disks full name is identical to base name. */
3537	pExtent->pszFullname = RTStrDup(pszBasename);
3538	if (!pExtent->pszFullname)
3539	return VERR_NO_MEMORY;
3540	pExtent->enmType = VMDKETYPE_FLAT;
3541	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
3542	pExtent->uSectorOffset = VMDK_BYTE2SECTOR(pPart->uStartOffset);
3543	pExtent->enmAccess = (pPart->uFlags & VBOXHDDRAW_READONLY) ? VMDKACCESS_READONLY : VMDKACCESS_READWRITE;
3544	pExtent->fMetaDirty = false;
3545
3546	/* Open flat image, the raw partition. */
3547	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszFullname,
3548	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
3549	false /* fCreate */));
3550	if (RT_FAILURE(rc))
3551	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not open raw partition file '%s'"), pExtent->pszFullname);
3552	}
3553	else
3554	{
3555	pExtent->pszBasename = NULL;
3556	pExtent->pszFullname = NULL;
3557	pExtent->enmType = VMDKETYPE_ZERO;
3558	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
3559	pExtent->uSectorOffset = 0;
3560	pExtent->enmAccess = VMDKACCESS_READWRITE;
3561	pExtent->fMetaDirty = false;
3562	}
3563	}
3564	}
3565	/* Another extent for filling up the rest of the image. */
3566	if (uStart != cbSize)
3567	{
3568	pExtent = &pImage->pExtents[cExtents++];
3569	pExtent->pszBasename = NULL;
3570	pExtent->pszFullname = NULL;
3571	pExtent->enmType = VMDKETYPE_ZERO;
3572	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize - uStart);
3573	pExtent->uSectorOffset = 0;
3574	pExtent->enmAccess = VMDKACCESS_READWRITE;
3575	pExtent->fMetaDirty = false;
3576	}
3577	}
3578
3579	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
3580	(pRaw->uFlags & VBOXHDDRAW_DISK) ?
3581	"fullDevice" : "partitionedDevice");
3582	if (RT_FAILURE(rc))
3583	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
3584	return rc;
3585	}
3586
3587	/**
3588	* Internal: create a regular (i.e. file-backed) VMDK image.
3589	*/
3590	static int vmdkCreateRegularImage(PVMDKIMAGE pImage, uint64_t cbSize,
3591	unsigned uImageFlags,
3592	PFNVDPROGRESS pfnProgress, void *pvUser,
3593	unsigned uPercentStart, unsigned uPercentSpan)
3594	{
3595	int rc = VINF_SUCCESS;
3596	unsigned cExtents = 1;
3597	uint64_t cbOffset = 0;
3598	uint64_t cbRemaining = cbSize;
3599
3600	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
3601	{
3602	cExtents = cbSize / VMDK_2G_SPLIT_SIZE;
3603	/* Do proper extent computation: need one smaller extent if the total
3604	* size isn't evenly divisible by the split size. */
3605	if (cbSize % VMDK_2G_SPLIT_SIZE)
3606	cExtents++;
3607	}
3608	rc = vmdkCreateExtents(pImage, cExtents);
3609	if (RT_FAILURE(rc))
3610	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
3611
3612	/* Basename strings needed for constructing the extent names. */
3613	char *pszBasenameSubstr = RTPathFilename(pImage->pszFilename);
3614	AssertPtr(pszBasenameSubstr);
3615	size_t cbBasenameSubstr = strlen(pszBasenameSubstr) + 1;
3616
3617	/* Create separate descriptor file if necessary. */
3618	if (cExtents != 1 \|\| (uImageFlags & VD_IMAGE_FLAGS_FIXED))
3619	{
3620	rc = vmdkFileOpen(pImage, &pImage->pFile, pImage->pszFilename,
3621	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
3622	true /* fCreate */));
3623	if (RT_FAILURE(rc))
3624	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new sparse descriptor file '%s'"), pImage->pszFilename);
3625	}
3626	else
3627	pImage->pFile = NULL;
3628
3629	/* Set up all extents. */
3630	for (unsigned i = 0; i < cExtents; i++)
3631	{
3632	PVMDKEXTENT pExtent = &pImage->pExtents[i];
3633	uint64_t cbExtent = cbRemaining;
3634
3635	/* Set up fullname/basename for extent description. Cannot use StrDup
3636	* for basename, as it is not guaranteed that the memory can be freed
3637	* with RTMemTmpFree, which must be used as in other code paths
3638	* StrDup is not usable. */
3639	if (cExtents == 1 && !(uImageFlags & VD_IMAGE_FLAGS_FIXED))
3640	{
3641	char pszBasename = (char )RTMemTmpAlloc(cbBasenameSubstr);
3642	if (!pszBasename)
3643	return VERR_NO_MEMORY;
3644	memcpy(pszBasename, pszBasenameSubstr, cbBasenameSubstr);
3645	pExtent->pszBasename = pszBasename;
3646	}
3647	else
3648	{
3649	char *pszBasenameSuff = RTPathSuffix(pszBasenameSubstr);
3650	char *pszBasenameBase = RTStrDup(pszBasenameSubstr);
3651	RTPathStripSuffix(pszBasenameBase);
3652	char *pszTmp;
3653	size_t cbTmp;
3654	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
3655	{
3656	if (cExtents == 1)
3657	RTStrAPrintf(&pszTmp, "%s-flat%s", pszBasenameBase,
3658	pszBasenameSuff);
3659	else
3660	RTStrAPrintf(&pszTmp, "%s-f%03d%s", pszBasenameBase,
3661	i+1, pszBasenameSuff);
3662	}
3663	else
3664	RTStrAPrintf(&pszTmp, "%s-s%03d%s", pszBasenameBase, i+1,
3665	pszBasenameSuff);
3666	RTStrFree(pszBasenameBase);
3667	if (!pszTmp)
3668	return VERR_NO_STR_MEMORY;
3669	cbTmp = strlen(pszTmp) + 1;
3670	char pszBasename = (char )RTMemTmpAlloc(cbTmp);
3671	if (!pszBasename)
3672	return VERR_NO_MEMORY;
3673	memcpy(pszBasename, pszTmp, cbTmp);
3674	RTStrFree(pszTmp);
3675	pExtent->pszBasename = pszBasename;
3676	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
3677	cbExtent = RT_MIN(cbRemaining, VMDK_2G_SPLIT_SIZE);
3678	}
3679	char *pszBasedirectory = RTStrDup(pImage->pszFilename);
3680	if (!pszBasedirectory)
3681	return VERR_NO_STR_MEMORY;
3682	RTPathStripFilename(pszBasedirectory);
3683	char *pszFullname = RTPathJoinA(pszBasedirectory, pExtent->pszBasename);
3684	RTStrFree(pszBasedirectory);
3685	if (!pszFullname)
3686	return VERR_NO_STR_MEMORY;
3687	pExtent->pszFullname = pszFullname;
3688
3689	/* Create file for extent. */
3690	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszFullname,
3691	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
3692	true /* fCreate */));
3693	if (RT_FAILURE(rc))
3694	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pExtent->pszFullname);
3695	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
3696	{
3697	rc = vdIfIoIntFileSetSize(pImage->pIfIo, pExtent->pFile->pStorage, cbExtent);
3698	if (RT_FAILURE(rc))
3699	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
3700
3701	/* Fill image with zeroes. We do this for every fixed-size image since on some systems
3702	* (for example Windows Vista), it takes ages to write a block near the end of a sparse
3703	* file and the guest could complain about an ATA timeout. */
3704
3705	/** @todo Starting with Linux 2.6.23, there is an fallocate() system call.
3706	* Currently supported file systems are ext4 and ocfs2. */
3707
3708	/* Allocate a temporary zero-filled buffer. Use a bigger block size to optimize writing */
3709	const size_t cbBuf = 128 * _1K;
3710	void *pvBuf = RTMemTmpAllocZ(cbBuf);
3711	if (!pvBuf)
3712	return VERR_NO_MEMORY;
3713
3714	uint64_t uOff = 0;
3715	/* Write data to all image blocks. */
3716	while (uOff < cbExtent)
3717	{
3718	unsigned cbChunk = (unsigned)RT_MIN(cbExtent, cbBuf);
3719
3720	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
3721	uOff, pvBuf, cbChunk);
3722	if (RT_FAILURE(rc))
3723	{
3724	RTMemFree(pvBuf);
3725	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: writing block failed for '%s'"), pImage->pszFilename);
3726	}
3727
3728	uOff += cbChunk;
3729
3730	if (pfnProgress)
3731	{
3732	rc = pfnProgress(pvUser,
3733	uPercentStart + (cbOffset + uOff) * uPercentSpan / cbSize);
3734	if (RT_FAILURE(rc))
3735	{
3736	RTMemFree(pvBuf);
3737	return rc;
3738	}
3739	}
3740	}
3741	RTMemTmpFree(pvBuf);
3742	}
3743
3744	/* Place descriptor file information (where integrated). */
3745	if (cExtents == 1 && !(uImageFlags & VD_IMAGE_FLAGS_FIXED))
3746	{
3747	pExtent->uDescriptorSector = 1;
3748	pExtent->cDescriptorSectors = VMDK_BYTE2SECTOR(pImage->cbDescAlloc);
3749	/* The descriptor is part of the (only) extent. */
3750	pExtent->pDescData = pImage->pDescData;
3751	pImage->pDescData = NULL;
3752	}
3753
3754	if (!(uImageFlags & VD_IMAGE_FLAGS_FIXED))
3755	{
3756	uint64_t cSectorsPerGDE, cSectorsPerGD;
3757	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE;
3758	pExtent->cSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64(cbExtent, _64K));
3759	pExtent->cSectorsPerGrain = VMDK_BYTE2SECTOR(_64K);
3760	pExtent->cGTEntries = 512;
3761	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
3762	pExtent->cSectorsPerGDE = cSectorsPerGDE;
3763	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
3764	cSectorsPerGD = (pExtent->cGDEntries + (512 / sizeof(uint32_t) - 1)) / (512 / sizeof(uint32_t));
3765	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
3766	{
3767	/* The spec says version is 1 for all VMDKs, but the vast
3768	* majority of streamOptimized VMDKs actually contain
3769	* version 3 - so go with the majority. Both are accepted. */
3770	pExtent->uVersion = 3;
3771	pExtent->uCompression = VMDK_COMPRESSION_DEFLATE;
3772	}
3773	}
3774	else
3775	{
3776	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
3777	pExtent->enmType = VMDKETYPE_VMFS;
3778	else
3779	pExtent->enmType = VMDKETYPE_FLAT;
3780	}
3781
3782	pExtent->enmAccess = VMDKACCESS_READWRITE;
3783	pExtent->fUncleanShutdown = true;
3784	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbExtent);
3785	pExtent->uSectorOffset = 0;
3786	pExtent->fMetaDirty = true;
3787
3788	if (!(uImageFlags & VD_IMAGE_FLAGS_FIXED))
3789	{
3790	/* fPreAlloc should never be false because VMware can't use such images. */
3791	rc = vmdkCreateGrainDirectory(pImage, pExtent,
3792	RT_MAX( pExtent->uDescriptorSector
3793	+ pExtent->cDescriptorSectors,
3794	1),
3795	true /* fPreAlloc */);
3796	if (RT_FAILURE(rc))
3797	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new grain directory in '%s'"), pExtent->pszFullname);
3798	}
3799
3800	cbOffset += cbExtent;
3801
3802	if (RT_SUCCESS(rc) && pfnProgress)
3803	pfnProgress(pvUser, uPercentStart + cbOffset * uPercentSpan / cbSize);
3804
3805	cbRemaining -= cbExtent;
3806	}
3807
3808	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
3809	{
3810	/* VirtualBox doesn't care, but VMWare ESX freaks out if the wrong
3811	* controller type is set in an image. */
3812	rc = vmdkDescDDBSetStr(pImage, &pImage->Descriptor, "ddb.adapterType", "lsilogic");
3813	if (RT_FAILURE(rc))
3814	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set controller type to lsilogic in '%s'"), pImage->pszFilename);
3815	}
3816
3817	const char *pszDescType = NULL;
3818	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
3819	{
3820	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
3821	pszDescType = "vmfs";
3822	else
3823	pszDescType = (cExtents == 1)
3824	? "monolithicFlat" : "twoGbMaxExtentFlat";
3825	}
3826	else
3827	{
3828	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
3829	pszDescType = "streamOptimized";
3830	else
3831	{
3832	pszDescType = (cExtents == 1)
3833	? "monolithicSparse" : "twoGbMaxExtentSparse";
3834	}
3835	}
3836	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
3837	pszDescType);
3838	if (RT_FAILURE(rc))
3839	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
3840	return rc;
3841	}
3842
3843	/**
3844	* Internal: Create a real stream optimized VMDK using only linear writes.
3845	*/
3846	static int vmdkCreateStreamImage(PVMDKIMAGE pImage, uint64_t cbSize,
3847	unsigned uImageFlags,
3848	PFNVDPROGRESS pfnProgress, void *pvUser,
3849	unsigned uPercentStart, unsigned uPercentSpan)
3850	{
3851	int rc;
3852
3853	rc = vmdkCreateExtents(pImage, 1);
3854	if (RT_FAILURE(rc))
3855	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
3856
3857	/* Basename strings needed for constructing the extent names. */
3858	const char *pszBasenameSubstr = RTPathFilename(pImage->pszFilename);
3859	AssertPtr(pszBasenameSubstr);
3860	size_t cbBasenameSubstr = strlen(pszBasenameSubstr) + 1;
3861
3862	/* No separate descriptor file. */
3863	pImage->pFile = NULL;
3864
3865	/* Set up all extents. */
3866	PVMDKEXTENT pExtent = &pImage->pExtents[0];
3867
3868	/* Set up fullname/basename for extent description. Cannot use StrDup
3869	* for basename, as it is not guaranteed that the memory can be freed
3870	* with RTMemTmpFree, which must be used as in other code paths
3871	* StrDup is not usable. */
3872	char pszBasename = (char )RTMemTmpAlloc(cbBasenameSubstr);
3873	if (!pszBasename)
3874	return VERR_NO_MEMORY;
3875	memcpy(pszBasename, pszBasenameSubstr, cbBasenameSubstr);
3876	pExtent->pszBasename = pszBasename;
3877
3878	char *pszBasedirectory = RTStrDup(pImage->pszFilename);
3879	RTPathStripFilename(pszBasedirectory);
3880	char *pszFullname = RTPathJoinA(pszBasedirectory, pExtent->pszBasename);
3881	RTStrFree(pszBasedirectory);
3882	if (!pszFullname)
3883	return VERR_NO_STR_MEMORY;
3884	pExtent->pszFullname = pszFullname;
3885
3886	/* Create file for extent. Make it write only, no reading allowed. */
3887	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszFullname,
3888	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
3889	true /* fCreate */)
3890	& ~RTFILE_O_READ);
3891	if (RT_FAILURE(rc))
3892	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pExtent->pszFullname);
3893
3894	/* Place descriptor file information. */
3895	pExtent->uDescriptorSector = 1;
3896	pExtent->cDescriptorSectors = VMDK_BYTE2SECTOR(pImage->cbDescAlloc);
3897	/* The descriptor is part of the (only) extent. */
3898	pExtent->pDescData = pImage->pDescData;
3899	pImage->pDescData = NULL;
3900
3901	uint64_t cSectorsPerGDE, cSectorsPerGD;
3902	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE;
3903	pExtent->cSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64(cbSize, _64K));
3904	pExtent->cSectorsPerGrain = VMDK_BYTE2SECTOR(_64K);
3905	pExtent->cGTEntries = 512;
3906	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
3907	pExtent->cSectorsPerGDE = cSectorsPerGDE;
3908	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
3909	cSectorsPerGD = (pExtent->cGDEntries + (512 / sizeof(uint32_t) - 1)) / (512 / sizeof(uint32_t));
3910
3911	/* The spec says version is 1 for all VMDKs, but the vast
3912	* majority of streamOptimized VMDKs actually contain
3913	* version 3 - so go with the majority. Both are accepted. */
3914	pExtent->uVersion = 3;
3915	pExtent->uCompression = VMDK_COMPRESSION_DEFLATE;
3916	pExtent->fFooter = true;
3917
3918	pExtent->enmAccess = VMDKACCESS_READONLY;
3919	pExtent->fUncleanShutdown = false;
3920	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
3921	pExtent->uSectorOffset = 0;
3922	pExtent->fMetaDirty = true;
3923
3924	/* Create grain directory, without preallocating it straight away. It will
3925	* be constructed on the fly when writing out the data and written when
3926	* closing the image. The end effect is that the full grain directory is
3927	* allocated, which is a requirement of the VMDK specs. */
3928	rc = vmdkCreateGrainDirectory(pImage, pExtent, VMDK_GD_AT_END,
3929	false /* fPreAlloc */);
3930	if (RT_FAILURE(rc))
3931	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new grain directory in '%s'"), pExtent->pszFullname);
3932
3933	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
3934	"streamOptimized");
3935	if (RT_FAILURE(rc))
3936	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
3937
3938	return rc;
3939	}
3940
3941	/**
3942	* Internal: The actual code for creating any VMDK variant currently in
3943	* existence on hosted environments.
3944	*/
3945	static int vmdkCreateImage(PVMDKIMAGE pImage, uint64_t cbSize,
3946	unsigned uImageFlags, const char *pszComment,
3947	PCVDGEOMETRY pPCHSGeometry,
3948	PCVDGEOMETRY pLCHSGeometry, PCRTUUID pUuid,
3949	PFNVDPROGRESS pfnProgress, void *pvUser,
3950	unsigned uPercentStart, unsigned uPercentSpan)
3951	{
3952	int rc;
3953
3954	pImage->uImageFlags = uImageFlags;
3955
3956	pImage->pIfError = VDIfErrorGet(pImage->pVDIfsDisk);
3957	pImage->pIfIo = VDIfIoIntGet(pImage->pVDIfsImage);
3958	AssertPtrReturn(pImage->pIfIo, VERR_INVALID_PARAMETER);
3959
3960	rc = vmdkCreateDescriptor(pImage, pImage->pDescData, pImage->cbDescAlloc,
3961	&pImage->Descriptor);
3962	if (RT_FAILURE(rc))
3963	{
3964	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new descriptor in '%s'"), pImage->pszFilename);
3965	goto out;
3966	}
3967
3968	if ( (uImageFlags & VD_IMAGE_FLAGS_FIXED)
3969	&& (uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK))
3970	{
3971	/* Raw disk image (includes raw partition). */
3972	const PVBOXHDDRAW pRaw = (const PVBOXHDDRAW)pszComment;
3973	/* As the comment is misused, zap it so that no garbage comment
3974	* is set below. */
3975	pszComment = NULL;
3976	rc = vmdkCreateRawImage(pImage, pRaw, cbSize);
3977	}
3978	else
3979	{
3980	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
3981	{
3982	/* Stream optimized sparse image (monolithic). */
3983	rc = vmdkCreateStreamImage(pImage, cbSize, uImageFlags,
3984	pfnProgress, pvUser, uPercentStart,
3985	uPercentSpan * 95 / 100);
3986	}
3987	else
3988	{
3989	/* Regular fixed or sparse image (monolithic or split). */
3990	rc = vmdkCreateRegularImage(pImage, cbSize, uImageFlags,
3991	pfnProgress, pvUser, uPercentStart,
3992	uPercentSpan * 95 / 100);
3993	}
3994	}
3995
3996	if (RT_FAILURE(rc))
3997	goto out;
3998
3999	if (RT_SUCCESS(rc) && pfnProgress)
4000	pfnProgress(pvUser, uPercentStart + uPercentSpan * 98 / 100);
4001
4002	pImage->cbSize = cbSize;
4003
4004	for (unsigned i = 0; i < pImage->cExtents; i++)
4005	{
4006	PVMDKEXTENT pExtent = &pImage->pExtents[i];
4007
4008	rc = vmdkDescExtInsert(pImage, &pImage->Descriptor, pExtent->enmAccess,
4009	pExtent->cNominalSectors, pExtent->enmType,
4010	pExtent->pszBasename, pExtent->uSectorOffset);
4011	if (RT_FAILURE(rc))
4012	{
4013	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not insert the extent list into descriptor in '%s'"), pImage->pszFilename);
4014	goto out;
4015	}
4016	}
4017	vmdkDescExtRemoveDummy(pImage, &pImage->Descriptor);
4018
4019	if ( pPCHSGeometry->cCylinders != 0
4020	&& pPCHSGeometry->cHeads != 0
4021	&& pPCHSGeometry->cSectors != 0)
4022	{
4023	rc = vmdkDescSetPCHSGeometry(pImage, pPCHSGeometry);
4024	if (RT_FAILURE(rc))
4025	goto out;
4026	}
4027	if ( pLCHSGeometry->cCylinders != 0
4028	&& pLCHSGeometry->cHeads != 0
4029	&& pLCHSGeometry->cSectors != 0)
4030	{
4031	rc = vmdkDescSetLCHSGeometry(pImage, pLCHSGeometry);
4032	if (RT_FAILURE(rc))
4033	goto out;
4034	}
4035
4036	pImage->LCHSGeometry = *pLCHSGeometry;
4037	pImage->PCHSGeometry = *pPCHSGeometry;
4038
4039	pImage->ImageUuid = *pUuid;
4040	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
4041	VMDK_DDB_IMAGE_UUID, &pImage->ImageUuid);
4042	if (RT_FAILURE(rc))
4043	{
4044	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image UUID in new descriptor in '%s'"), pImage->pszFilename);
4045	goto out;
4046	}
4047	RTUuidClear(&pImage->ParentUuid);
4048	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
4049	VMDK_DDB_PARENT_UUID, &pImage->ParentUuid);
4050	if (RT_FAILURE(rc))
4051	{
4052	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent image UUID in new descriptor in '%s'"), pImage->pszFilename);
4053	goto out;
4054	}
4055	RTUuidClear(&pImage->ModificationUuid);
4056	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
4057	VMDK_DDB_MODIFICATION_UUID,
4058	&pImage->ModificationUuid);
4059	if (RT_FAILURE(rc))
4060	{
4061	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing modification UUID in new descriptor in '%s'"), pImage->pszFilename);
4062	goto out;
4063	}
4064	RTUuidClear(&pImage->ParentModificationUuid);
4065	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
4066	VMDK_DDB_PARENT_MODIFICATION_UUID,
4067	&pImage->ParentModificationUuid);
4068	if (RT_FAILURE(rc))
4069	{
4070	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent modification UUID in new descriptor in '%s'"), pImage->pszFilename);
4071	goto out;
4072	}
4073
4074	rc = vmdkAllocateGrainTableCache(pImage);
4075	if (RT_FAILURE(rc))
4076	goto out;
4077
4078	rc = vmdkSetImageComment(pImage, pszComment);
4079	if (RT_FAILURE(rc))
4080	{
4081	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot set image comment in '%s'"), pImage->pszFilename);
4082	goto out;
4083	}
4084
4085	if (RT_SUCCESS(rc) && pfnProgress)
4086	pfnProgress(pvUser, uPercentStart + uPercentSpan * 99 / 100);
4087
4088	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
4089	{
4090	/* streamOptimized is a bit special, we cannot trigger the flush
4091	* until all data has been written. So we write the necessary
4092	* information explicitly. */
4093	pImage->pExtents[0].cDescriptorSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64( pImage->Descriptor.aLines[pImage->Descriptor.cLines]
4094	- pImage->Descriptor.aLines[0], 512));
4095	rc = vmdkWriteMetaSparseExtent(pImage, &pImage->pExtents[0], 0, NULL);
4096	if (RT_FAILURE(rc))
4097	{
4098	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write VMDK header in '%s'"), pImage->pszFilename);
4099	goto out;
4100	}
4101
4102	rc = vmdkWriteDescriptor(pImage, NULL);
4103	if (RT_FAILURE(rc))
4104	{
4105	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write VMDK descriptor in '%s'"), pImage->pszFilename);
4106	goto out;
4107	}
4108	}
4109	else
4110	rc = vmdkFlushImage(pImage, NULL);
4111
4112	out:
4113	if (RT_SUCCESS(rc) && pfnProgress)
4114	pfnProgress(pvUser, uPercentStart + uPercentSpan);
4115
4116	if (RT_FAILURE(rc))
4117	vmdkFreeImage(pImage, rc != VERR_ALREADY_EXISTS);
4118	return rc;
4119	}
4120
4121	/**
4122	* Internal: Update image comment.
4123	*/
4124	static int vmdkSetImageComment(PVMDKIMAGE pImage, const char *pszComment)
4125	{
4126	char *pszCommentEncoded;
4127	if (pszComment)
4128	{
4129	pszCommentEncoded = vmdkEncodeString(pszComment);
4130	if (!pszCommentEncoded)
4131	return VERR_NO_MEMORY;
4132	}
4133	else
4134	pszCommentEncoded = NULL;
4135	int rc = vmdkDescDDBSetStr(pImage, &pImage->Descriptor,
4136	"ddb.comment", pszCommentEncoded);
4137	if (pszComment)
4138	RTStrFree(pszCommentEncoded);
4139	if (RT_FAILURE(rc))
4140	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image comment in descriptor in '%s'"), pImage->pszFilename);
4141	return VINF_SUCCESS;
4142	}
4143
4144	/**
4145	* Internal. Clear the grain table buffer for real stream optimized writing.
4146	*/
4147	static void vmdkStreamClearGT(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
4148	{
4149	uint32_t cCacheLines = RT_ALIGN(pExtent->cGTEntries, VMDK_GT_CACHELINE_SIZE) / VMDK_GT_CACHELINE_SIZE;
4150	for (uint32_t i = 0; i < cCacheLines; i++)
4151	memset(&pImage->pGTCache->aGTCache[i].aGTData[0], '\0',
4152	VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t));
4153	}
4154
4155	/**
4156	* Internal. Flush the grain table buffer for real stream optimized writing.
4157	*/
4158	static int vmdkStreamFlushGT(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
4159	uint32_t uGDEntry)
4160	{
4161	int rc = VINF_SUCCESS;
4162	uint32_t cCacheLines = RT_ALIGN(pExtent->cGTEntries, VMDK_GT_CACHELINE_SIZE) / VMDK_GT_CACHELINE_SIZE;
4163
4164	/* VMware does not write out completely empty grain tables in the case
4165	* of streamOptimized images, which according to my interpretation of
4166	* the VMDK 1.1 spec is bending the rules. Since they do it and we can
4167	* handle it without problems do it the same way and save some bytes. */
4168	bool fAllZero = true;
4169	for (uint32_t i = 0; i < cCacheLines; i++)
4170	{
4171	/* Convert the grain table to little endian in place, as it will not
4172	* be used at all after this function has been called. */
4173	uint32_t *pGTTmp = &pImage->pGTCache->aGTCache[i].aGTData[0];
4174	for (uint32_t j = 0; j < VMDK_GT_CACHELINE_SIZE; j++, pGTTmp++)
4175	if (*pGTTmp)
4176	{
4177	fAllZero = false;
4178	break;
4179	}
4180	if (!fAllZero)
4181	break;
4182	}
4183	if (fAllZero)
4184	return VINF_SUCCESS;
4185
4186	uint64_t uFileOffset = pExtent->uAppendPosition;
4187	if (!uFileOffset)
4188	return VERR_INTERNAL_ERROR;
4189	/* Align to sector, as the previous write could have been any size. */
4190	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
4191
4192	/* Grain table marker. */
4193	uint8_t aMarker[512];
4194	PVMDKMARKER pMarker = (PVMDKMARKER)&aMarker[0];
4195	memset(pMarker, '\0', sizeof(aMarker));
4196	pMarker->uSector = RT_H2LE_U64(VMDK_BYTE2SECTOR((uint64_t)pExtent->cGTEntries * sizeof(uint32_t)));
4197	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_GT);
4198	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
4199	aMarker, sizeof(aMarker));
4200	AssertRC(rc);
4201	uFileOffset += 512;
4202
4203	if (!pExtent->pGD \|\| pExtent->pGD[uGDEntry])
4204	return VERR_INTERNAL_ERROR;
4205
4206	pExtent->pGD[uGDEntry] = VMDK_BYTE2SECTOR(uFileOffset);
4207
4208	for (uint32_t i = 0; i < cCacheLines; i++)
4209	{
4210	/* Convert the grain table to little endian in place, as it will not
4211	* be used at all after this function has been called. */
4212	uint32_t *pGTTmp = &pImage->pGTCache->aGTCache[i].aGTData[0];
4213	for (uint32_t j = 0; j < VMDK_GT_CACHELINE_SIZE; j++, pGTTmp++)
4214	pGTTmp = RT_H2LE_U32(pGTTmp);
4215
4216	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
4217	&pImage->pGTCache->aGTCache[i].aGTData[0],
4218	VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t));
4219	uFileOffset += VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t);
4220	if (RT_FAILURE(rc))
4221	break;
4222	}
4223	Assert(!(uFileOffset % 512));
4224	pExtent->uAppendPosition = RT_ALIGN_64(uFileOffset, 512);
4225	return rc;
4226	}
4227
4228	/**
4229	* Internal. Free all allocated space for representing an image, and optionally
4230	* delete the image from disk.
4231	*/
4232	static int vmdkFreeImage(PVMDKIMAGE pImage, bool fDelete)
4233	{
4234	int rc = VINF_SUCCESS;
4235
4236	/* Freeing a never allocated image (e.g. because the open failed) is
4237	* not signalled as an error. After all nothing bad happens. */
4238	if (pImage)
4239	{
4240	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
4241	{
4242	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
4243	{
4244	/* Check if all extents are clean. */
4245	for (unsigned i = 0; i < pImage->cExtents; i++)
4246	{
4247	Assert(!pImage->pExtents[i].fUncleanShutdown);
4248	}
4249	}
4250	else
4251	{
4252	/* Mark all extents as clean. */
4253	for (unsigned i = 0; i < pImage->cExtents; i++)
4254	{
4255	if ( ( pImage->pExtents[i].enmType == VMDKETYPE_HOSTED_SPARSE
4256	#ifdef VBOX_WITH_VMDK_ESX
4257	\|\| pImage->pExtents[i].enmType == VMDKETYPE_ESX_SPARSE
4258	#endif /* VBOX_WITH_VMDK_ESX */
4259	)
4260	&& pImage->pExtents[i].fUncleanShutdown)
4261	{
4262	pImage->pExtents[i].fUncleanShutdown = false;
4263	pImage->pExtents[i].fMetaDirty = true;
4264	}
4265
4266	/* From now on it's not safe to append any more data. */
4267	pImage->pExtents[i].uAppendPosition = 0;
4268	}
4269	}
4270	}
4271
4272	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
4273	{
4274	/* No need to write any pending data if the file will be deleted
4275	* or if the new file wasn't successfully created. */
4276	if ( !fDelete && pImage->pExtents
4277	&& pImage->pExtents[0].cGTEntries
4278	&& pImage->pExtents[0].uAppendPosition)
4279	{
4280	PVMDKEXTENT pExtent = &pImage->pExtents[0];
4281	uint32_t uLastGDEntry = pExtent->uLastGrainAccess / pExtent->cGTEntries;
4282	rc = vmdkStreamFlushGT(pImage, pExtent, uLastGDEntry);
4283	AssertRC(rc);
4284	vmdkStreamClearGT(pImage, pExtent);
4285	for (uint32_t i = uLastGDEntry + 1; i < pExtent->cGDEntries; i++)
4286	{
4287	rc = vmdkStreamFlushGT(pImage, pExtent, i);
4288	AssertRC(rc);
4289	}
4290
4291	uint64_t uFileOffset = pExtent->uAppendPosition;
4292	if (!uFileOffset)
4293	return VERR_INTERNAL_ERROR;
4294	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
4295
4296	/* From now on it's not safe to append any more data. */
4297	pExtent->uAppendPosition = 0;
4298
4299	/* Grain directory marker. */
4300	uint8_t aMarker[512];
4301	PVMDKMARKER pMarker = (PVMDKMARKER)&aMarker[0];
4302	memset(pMarker, '\0', sizeof(aMarker));
4303	pMarker->uSector = VMDK_BYTE2SECTOR(RT_ALIGN_64(RT_H2LE_U64((uint64_t)pExtent->cGDEntries * sizeof(uint32_t)), 512));
4304	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_GD);
4305	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
4306	aMarker, sizeof(aMarker));
4307	AssertRC(rc);
4308	uFileOffset += 512;
4309
4310	/* Write grain directory in little endian style. The array will
4311	* not be used after this, so convert in place. */
4312	uint32_t *pGDTmp = pExtent->pGD;
4313	for (uint32_t i = 0; i < pExtent->cGDEntries; i++, pGDTmp++)
4314	pGDTmp = RT_H2LE_U32(pGDTmp);
4315	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
4316	uFileOffset, pExtent->pGD,
4317	pExtent->cGDEntries * sizeof(uint32_t));
4318	AssertRC(rc);
4319
4320	pExtent->uSectorGD = VMDK_BYTE2SECTOR(uFileOffset);
4321	pExtent->uSectorRGD = VMDK_BYTE2SECTOR(uFileOffset);
4322	uFileOffset = RT_ALIGN_64( uFileOffset
4323	+ pExtent->cGDEntries * sizeof(uint32_t),
4324	512);
4325
4326	/* Footer marker. */
4327	memset(pMarker, '\0', sizeof(aMarker));
4328	pMarker->uSector = VMDK_BYTE2SECTOR(512);
4329	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_FOOTER);
4330	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
4331	uFileOffset, aMarker, sizeof(aMarker));
4332	AssertRC(rc);
4333
4334	uFileOffset += 512;
4335	rc = vmdkWriteMetaSparseExtent(pImage, pExtent, uFileOffset, NULL);
4336	AssertRC(rc);
4337
4338	uFileOffset += 512;
4339	/* End-of-stream marker. */
4340	memset(pMarker, '\0', sizeof(aMarker));
4341	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
4342	uFileOffset, aMarker, sizeof(aMarker));
4343	AssertRC(rc);
4344	}
4345	}
4346	else
4347	vmdkFlushImage(pImage, NULL);
4348
4349	if (pImage->pExtents != NULL)
4350	{
4351	for (unsigned i = 0 ; i < pImage->cExtents; i++)
4352	{
4353	int rc2 = vmdkFreeExtentData(pImage, &pImage->pExtents[i], fDelete);
4354	if (RT_SUCCESS(rc))
4355	rc = rc2; /* Propogate any error when closing the file. */
4356	}
4357	RTMemFree(pImage->pExtents);
4358	pImage->pExtents = NULL;
4359	}
4360	pImage->cExtents = 0;
4361	if (pImage->pFile != NULL)
4362	{
4363	int rc2 = vmdkFileClose(pImage, &pImage->pFile, fDelete);
4364	if (RT_SUCCESS(rc))
4365	rc = rc2; /* Propogate any error when closing the file. */
4366	}
4367	int rc2 = vmdkFileCheckAllClose(pImage);
4368	if (RT_SUCCESS(rc))
4369	rc = rc2; /* Propogate any error when closing the file. */
4370
4371	if (pImage->pGTCache)
4372	{
4373	RTMemFree(pImage->pGTCache);
4374	pImage->pGTCache = NULL;
4375	}
4376	if (pImage->pDescData)
4377	{
4378	RTMemFree(pImage->pDescData);
4379	pImage->pDescData = NULL;
4380	}
4381	}
4382
4383	LogFlowFunc(("returns %Rrc\n", rc));
4384	return rc;
4385	}
4386
4387	/**
4388	* Internal. Flush image data (and metadata) to disk.
4389	*/
4390	static int vmdkFlushImage(PVMDKIMAGE pImage, PVDIOCTX pIoCtx)
4391	{
4392	PVMDKEXTENT pExtent;
4393	int rc = VINF_SUCCESS;
4394
4395	/* Update descriptor if changed. */
4396	if (pImage->Descriptor.fDirty)
4397	{
4398	rc = vmdkWriteDescriptor(pImage, pIoCtx);
4399	if (RT_FAILURE(rc))
4400	goto out;
4401	}
4402
4403	for (unsigned i = 0; i < pImage->cExtents; i++)
4404	{
4405	pExtent = &pImage->pExtents[i];
4406	if (pExtent->pFile != NULL && pExtent->fMetaDirty)
4407	{
4408	switch (pExtent->enmType)
4409	{
4410	case VMDKETYPE_HOSTED_SPARSE:
4411	if (!pExtent->fFooter)
4412	{
4413	rc = vmdkWriteMetaSparseExtent(pImage, pExtent, 0, pIoCtx);
4414	if (RT_FAILURE(rc))
4415	goto out;
4416	}
4417	else
4418	{
4419	uint64_t uFileOffset = pExtent->uAppendPosition;
4420	/* Simply skip writing anything if the streamOptimized
4421	* image hasn't been just created. */
4422	if (!uFileOffset)
4423	break;
4424	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
4425	rc = vmdkWriteMetaSparseExtent(pImage, pExtent,
4426	uFileOffset, pIoCtx);
4427	if (RT_FAILURE(rc))
4428	goto out;
4429	}
4430	break;
4431	#ifdef VBOX_WITH_VMDK_ESX
4432	case VMDKETYPE_ESX_SPARSE:
4433	/** @todo update the header. */
4434	break;
4435	#endif /* VBOX_WITH_VMDK_ESX */
4436	case VMDKETYPE_VMFS:
4437	case VMDKETYPE_FLAT:
4438	/* Nothing to do. */
4439	break;
4440	case VMDKETYPE_ZERO:
4441	default:
4442	AssertMsgFailed(("extent with type %d marked as dirty\n",
4443	pExtent->enmType));
4444	break;
4445	}
4446	}
4447	switch (pExtent->enmType)
4448	{
4449	case VMDKETYPE_HOSTED_SPARSE:
4450	#ifdef VBOX_WITH_VMDK_ESX
4451	case VMDKETYPE_ESX_SPARSE:
4452	#endif /* VBOX_WITH_VMDK_ESX */
4453	case VMDKETYPE_VMFS:
4454	case VMDKETYPE_FLAT:
4455	/** @todo implement proper path absolute check. */
4456	if ( pExtent->pFile != NULL
4457	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
4458	&& !(pExtent->pszBasename[0] == RTPATH_SLASH))
4459	rc = vdIfIoIntFileFlush(pImage->pIfIo, pExtent->pFile->pStorage, pIoCtx,
4460	NULL, NULL);
4461	break;
4462	case VMDKETYPE_ZERO:
4463	/* No need to do anything for this extent. */
4464	break;
4465	default:
4466	AssertMsgFailed(("unknown extent type %d\n", pExtent->enmType));
4467	break;
4468	}
4469	}
4470
4471	out:
4472	return rc;
4473	}
4474
4475	/**
4476	* Internal. Find extent corresponding to the sector number in the disk.
4477	*/
4478	static int vmdkFindExtent(PVMDKIMAGE pImage, uint64_t offSector,
4479	PVMDKEXTENT ppExtent, uint64_t puSectorInExtent)
4480	{
4481	PVMDKEXTENT pExtent = NULL;
4482	int rc = VINF_SUCCESS;
4483
4484	for (unsigned i = 0; i < pImage->cExtents; i++)
4485	{
4486	if (offSector < pImage->pExtents[i].cNominalSectors)
4487	{
4488	pExtent = &pImage->pExtents[i];
4489	*puSectorInExtent = offSector + pImage->pExtents[i].uSectorOffset;
4490	break;
4491	}
4492	offSector -= pImage->pExtents[i].cNominalSectors;
4493	}
4494
4495	if (pExtent)
4496	*ppExtent = pExtent;
4497	else
4498	rc = VERR_IO_SECTOR_NOT_FOUND;
4499
4500	return rc;
4501	}
4502
4503	/**
4504	* Internal. Hash function for placing the grain table hash entries.
4505	*/
4506	static uint32_t vmdkGTCacheHash(PVMDKGTCACHE pCache, uint64_t uSector,
4507	unsigned uExtent)
4508	{
4509	/** @todo this hash function is quite simple, maybe use a better one which
4510	* scrambles the bits better. */
4511	return (uSector + uExtent) % pCache->cEntries;
4512	}
4513
4514	/**
4515	* Internal. Get sector number in the extent file from the relative sector
4516	* number in the extent.
4517	*/
4518	static int vmdkGetSector(PVMDKIMAGE pImage, PVDIOCTX pIoCtx,
4519	PVMDKEXTENT pExtent, uint64_t uSector,
4520	uint64_t *puExtentSector)
4521	{
4522	PVMDKGTCACHE pCache = pImage->pGTCache;
4523	uint64_t uGDIndex, uGTSector, uGTBlock;
4524	uint32_t uGTHash, uGTBlockIndex;
4525	PVMDKGTCACHEENTRY pGTCacheEntry;
4526	uint32_t aGTDataTmp[VMDK_GT_CACHELINE_SIZE];
4527	int rc;
4528
4529	/* For newly created and readonly/sequentially opened streamOptimized
4530	* images this must be a no-op, as the grain directory is not there. */
4531	if ( ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
4532	&& pExtent->uAppendPosition)
4533	\|\| ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
4534	&& pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY
4535	&& pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
4536	{
4537	*puExtentSector = 0;
4538	return VINF_SUCCESS;
4539	}
4540
4541	uGDIndex = uSector / pExtent->cSectorsPerGDE;
4542	if (uGDIndex >= pExtent->cGDEntries)
4543	return VERR_OUT_OF_RANGE;
4544	uGTSector = pExtent->pGD[uGDIndex];
4545	if (!uGTSector)
4546	{
4547	/* There is no grain table referenced by this grain directory
4548	* entry. So there is absolutely no data in this area. */
4549	*puExtentSector = 0;
4550	return VINF_SUCCESS;
4551	}
4552
4553	uGTBlock = uSector / (pExtent->cSectorsPerGrain * VMDK_GT_CACHELINE_SIZE);
4554	uGTHash = vmdkGTCacheHash(pCache, uGTBlock, pExtent->uExtent);
4555	pGTCacheEntry = &pCache->aGTCache[uGTHash];
4556	if ( pGTCacheEntry->uExtent != pExtent->uExtent
4557	\|\| pGTCacheEntry->uGTBlock != uGTBlock)
4558	{
4559	/* Cache miss, fetch data from disk. */
4560	PVDMETAXFER pMetaXfer;
4561	rc = vdIfIoIntFileReadMeta(pImage->pIfIo, pExtent->pFile->pStorage,
4562	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
4563	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx, &pMetaXfer, NULL, NULL);
4564	if (RT_FAILURE(rc))
4565	return rc;
4566	/* We can release the metadata transfer immediately. */
4567	vdIfIoIntMetaXferRelease(pImage->pIfIo, pMetaXfer);
4568	pGTCacheEntry->uExtent = pExtent->uExtent;
4569	pGTCacheEntry->uGTBlock = uGTBlock;
4570	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
4571	pGTCacheEntry->aGTData[i] = RT_LE2H_U32(aGTDataTmp[i]);
4572	}
4573	uGTBlockIndex = (uSector / pExtent->cSectorsPerGrain) % VMDK_GT_CACHELINE_SIZE;
4574	uint32_t uGrainSector = pGTCacheEntry->aGTData[uGTBlockIndex];
4575	if (uGrainSector)
4576	*puExtentSector = uGrainSector + uSector % pExtent->cSectorsPerGrain;
4577	else
4578	*puExtentSector = 0;
4579	return VINF_SUCCESS;
4580	}
4581
4582	/**
4583	* Internal. Writes the grain and also if necessary the grain tables.
4584	* Uses the grain table cache as a true grain table.
4585	*/
4586	static int vmdkStreamAllocGrain(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
4587	uint64_t uSector, PVDIOCTX pIoCtx,
4588	uint64_t cbWrite)
4589	{
4590	uint32_t uGrain;
4591	uint32_t uGDEntry, uLastGDEntry;
4592	uint32_t cbGrain = 0;
4593	uint32_t uCacheLine, uCacheEntry;
4594	const void *pData;
4595	int rc;
4596
4597	/* Very strict requirements: always write at least one full grain, with
4598	* proper alignment. Everything else would require reading of already
4599	* written data, which we don't support for obvious reasons. The only
4600	* exception is the last grain, and only if the image size specifies
4601	* that only some portion holds data. In any case the write must be
4602	* within the image limits, no "overshoot" allowed. */
4603	if ( cbWrite == 0
4604	\|\| ( cbWrite < VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain)
4605	&& pExtent->cNominalSectors - uSector >= pExtent->cSectorsPerGrain)
4606	\|\| uSector % pExtent->cSectorsPerGrain
4607	\|\| uSector + VMDK_BYTE2SECTOR(cbWrite) > pExtent->cNominalSectors)
4608	return VERR_INVALID_PARAMETER;
4609
4610	/* Clip write range to at most the rest of the grain. */
4611	cbWrite = RT_MIN(cbWrite, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain - uSector % pExtent->cSectorsPerGrain));
4612
4613	/* Do not allow to go back. */
4614	uGrain = uSector / pExtent->cSectorsPerGrain;
4615	uCacheLine = uGrain % pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE;
4616	uCacheEntry = uGrain % VMDK_GT_CACHELINE_SIZE;
4617	uGDEntry = uGrain / pExtent->cGTEntries;
4618	uLastGDEntry = pExtent->uLastGrainAccess / pExtent->cGTEntries;
4619	if (uGrain < pExtent->uLastGrainAccess)
4620	return VERR_VD_VMDK_INVALID_WRITE;
4621
4622	/* Zero byte write optimization. Since we don't tell VBoxHDD that we need
4623	* to allocate something, we also need to detect the situation ourself. */
4624	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_HONOR_ZEROES)
4625	&& vdIfIoIntIoCtxIsZero(pImage->pIfIo, pIoCtx, cbWrite, true /* fAdvance */))
4626	return VINF_SUCCESS;
4627
4628	if (uGDEntry != uLastGDEntry)
4629	{
4630	rc = vmdkStreamFlushGT(pImage, pExtent, uLastGDEntry);
4631	if (RT_FAILURE(rc))
4632	return rc;
4633	vmdkStreamClearGT(pImage, pExtent);
4634	for (uint32_t i = uLastGDEntry + 1; i < uGDEntry; i++)
4635	{
4636	rc = vmdkStreamFlushGT(pImage, pExtent, i);
4637	if (RT_FAILURE(rc))
4638	return rc;
4639	}
4640	}
4641
4642	uint64_t uFileOffset;
4643	uFileOffset = pExtent->uAppendPosition;
4644	if (!uFileOffset)
4645	return VERR_INTERNAL_ERROR;
4646	/* Align to sector, as the previous write could have been any size. */
4647	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
4648
4649	/* Paranoia check: extent type, grain table buffer presence and
4650	* grain table buffer space. Also grain table entry must be clear. */
4651	if ( pExtent->enmType != VMDKETYPE_HOSTED_SPARSE
4652	\|\| !pImage->pGTCache
4653	\|\| pExtent->cGTEntries > VMDK_GT_CACHE_SIZE * VMDK_GT_CACHELINE_SIZE
4654	\|\| pImage->pGTCache->aGTCache[uCacheLine].aGTData[uCacheEntry])
4655	return VERR_INTERNAL_ERROR;
4656
4657	/* Update grain table entry. */
4658	pImage->pGTCache->aGTCache[uCacheLine].aGTData[uCacheEntry] = VMDK_BYTE2SECTOR(uFileOffset);
4659
4660	if (cbWrite != VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
4661	{
4662	vdIfIoIntIoCtxCopyFrom(pImage->pIfIo, pIoCtx, pExtent->pvGrain, cbWrite);
4663	memset((char *)pExtent->pvGrain + cbWrite, '\0',
4664	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain) - cbWrite);
4665	pData = pExtent->pvGrain;
4666	}
4667	else
4668	{
4669	RTSGSEG Segment;
4670	unsigned cSegments = 1;
4671	size_t cbSeg = 0;
4672
4673	cbSeg = vdIfIoIntIoCtxSegArrayCreate(pImage->pIfIo, pIoCtx, &Segment,
4674	&cSegments, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
4675	Assert(cbSeg == VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
4676	pData = Segment.pvSeg;
4677	}
4678	rc = vmdkFileDeflateSync(pImage, pExtent, uFileOffset, pData,
4679	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
4680	uSector, &cbGrain);
4681	if (RT_FAILURE(rc))
4682	{
4683	pExtent->uGrainSectorAbs = 0;
4684	AssertRC(rc);
4685	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write compressed data block in '%s'"), pExtent->pszFullname);
4686	}
4687	pExtent->uLastGrainAccess = uGrain;
4688	pExtent->uAppendPosition += cbGrain;
4689
4690	return rc;
4691	}
4692
4693	/**
4694	* Internal: Updates the grain table during grain allocation.
4695	*/
4696	static int vmdkAllocGrainGTUpdate(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, PVDIOCTX pIoCtx,
4697	PVMDKGRAINALLOCASYNC pGrainAlloc)
4698	{
4699	int rc = VINF_SUCCESS;
4700	PVMDKGTCACHE pCache = pImage->pGTCache;
4701	uint32_t aGTDataTmp[VMDK_GT_CACHELINE_SIZE];
4702	uint32_t uGTHash, uGTBlockIndex;
4703	uint64_t uGTSector, uRGTSector, uGTBlock;
4704	uint64_t uSector = pGrainAlloc->uSector;
4705	PVMDKGTCACHEENTRY pGTCacheEntry;
4706
4707	LogFlowFunc(("pImage=%#p pExtent=%#p pCache=%#p pIoCtx=%#p pGrainAlloc=%#p\n",
4708	pImage, pExtent, pCache, pIoCtx, pGrainAlloc));
4709
4710	uGTSector = pGrainAlloc->uGTSector;
4711	uRGTSector = pGrainAlloc->uRGTSector;
4712	LogFlow(("uGTSector=%llu uRGTSector=%llu\n", uGTSector, uRGTSector));
4713
4714	/* Update the grain table (and the cache). */
4715	uGTBlock = uSector / (pExtent->cSectorsPerGrain * VMDK_GT_CACHELINE_SIZE);
4716	uGTHash = vmdkGTCacheHash(pCache, uGTBlock, pExtent->uExtent);
4717	pGTCacheEntry = &pCache->aGTCache[uGTHash];
4718	if ( pGTCacheEntry->uExtent != pExtent->uExtent
4719	\|\| pGTCacheEntry->uGTBlock != uGTBlock)
4720	{
4721	/* Cache miss, fetch data from disk. */
4722	LogFlow(("Cache miss, fetch data from disk\n"));
4723	PVDMETAXFER pMetaXfer = NULL;
4724	rc = vdIfIoIntFileReadMeta(pImage->pIfIo, pExtent->pFile->pStorage,
4725	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
4726	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
4727	&pMetaXfer, vmdkAllocGrainComplete, pGrainAlloc);
4728	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
4729	{
4730	pGrainAlloc->cIoXfersPending++;
4731	pGrainAlloc->fGTUpdateNeeded = true;
4732	/* Leave early, we will be called again after the read completed. */
4733	LogFlowFunc(("Metadata read in progress, leaving\n"));
4734	return rc;
4735	}
4736	else if (RT_FAILURE(rc))
4737	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot read allocated grain table entry in '%s'"), pExtent->pszFullname);
4738	vdIfIoIntMetaXferRelease(pImage->pIfIo, pMetaXfer);
4739	pGTCacheEntry->uExtent = pExtent->uExtent;
4740	pGTCacheEntry->uGTBlock = uGTBlock;
4741	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
4742	pGTCacheEntry->aGTData[i] = RT_LE2H_U32(aGTDataTmp[i]);
4743	}
4744	else
4745	{
4746	/* Cache hit. Convert grain table block back to disk format, otherwise
4747	* the code below will write garbage for all but the updated entry. */
4748	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
4749	aGTDataTmp[i] = RT_H2LE_U32(pGTCacheEntry->aGTData[i]);
4750	}
4751	pGrainAlloc->fGTUpdateNeeded = false;
4752	uGTBlockIndex = (uSector / pExtent->cSectorsPerGrain) % VMDK_GT_CACHELINE_SIZE;
4753	aGTDataTmp[uGTBlockIndex] = RT_H2LE_U32(VMDK_BYTE2SECTOR(pGrainAlloc->uGrainOffset));
4754	pGTCacheEntry->aGTData[uGTBlockIndex] = VMDK_BYTE2SECTOR(pGrainAlloc->uGrainOffset);
4755	/* Update grain table on disk. */
4756	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
4757	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
4758	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
4759	vmdkAllocGrainComplete, pGrainAlloc);
4760	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
4761	pGrainAlloc->cIoXfersPending++;
4762	else if (RT_FAILURE(rc))
4763	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write updated grain table in '%s'"), pExtent->pszFullname);
4764	if (pExtent->pRGD)
4765	{
4766	/* Update backup grain table on disk. */
4767	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
4768	VMDK_SECTOR2BYTE(uRGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
4769	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
4770	vmdkAllocGrainComplete, pGrainAlloc);
4771	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
4772	pGrainAlloc->cIoXfersPending++;
4773	else if (RT_FAILURE(rc))
4774	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write updated backup grain table in '%s'"), pExtent->pszFullname);
4775	}
4776	#ifdef VBOX_WITH_VMDK_ESX
4777	if (RT_SUCCESS(rc) && pExtent->enmType == VMDKETYPE_ESX_SPARSE)
4778	{
4779	pExtent->uFreeSector = uGTSector + VMDK_BYTE2SECTOR(cbWrite);
4780	pExtent->fMetaDirty = true;
4781	}
4782	#endif /* VBOX_WITH_VMDK_ESX */
4783
4784	LogFlowFunc(("leaving rc=%Rrc\n", rc));
4785
4786	return rc;
4787	}
4788
4789	/**
4790	* Internal - complete the grain allocation by updating disk grain table if required.
4791	*/
4792	static int vmdkAllocGrainComplete(void pBackendData, PVDIOCTX pIoCtx, void pvUser, int rcReq)
4793	{
4794	int rc = VINF_SUCCESS;
4795	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4796	PVMDKGRAINALLOCASYNC pGrainAlloc = (PVMDKGRAINALLOCASYNC)pvUser;
4797	PVMDKEXTENT pExtent = pGrainAlloc->pExtent;
4798
4799	LogFlowFunc(("pBackendData=%#p pIoCtx=%#p pvUser=%#p rcReq=%Rrc\n",
4800	pBackendData, pIoCtx, pvUser, rcReq));
4801
4802	pGrainAlloc->cIoXfersPending--;
4803	if (!pGrainAlloc->cIoXfersPending && pGrainAlloc->fGTUpdateNeeded)
4804	rc = vmdkAllocGrainGTUpdate(pImage, pGrainAlloc->pExtent, pIoCtx, pGrainAlloc);
4805
4806	if (!pGrainAlloc->cIoXfersPending)
4807	{
4808	/* Grain allocation completed. */
4809	RTMemFree(pGrainAlloc);
4810	}
4811
4812	LogFlowFunc(("Leaving rc=%Rrc\n", rc));
4813	return rc;
4814	}
4815
4816	/**
4817	* Internal. Allocates a new grain table (if necessary).
4818	*/
4819	static int vmdkAllocGrain(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, PVDIOCTX pIoCtx,
4820	uint64_t uSector, uint64_t cbWrite)
4821	{
4822	PVMDKGTCACHE pCache = pImage->pGTCache;
4823	uint64_t uGDIndex, uGTSector, uRGTSector;
4824	uint64_t uFileOffset;
4825	PVMDKGRAINALLOCASYNC pGrainAlloc = NULL;
4826	int rc;
4827
4828	LogFlowFunc(("pCache=%#p pExtent=%#p pIoCtx=%#p uSector=%llu cbWrite=%llu\n",
4829	pCache, pExtent, pIoCtx, uSector, cbWrite));
4830
4831	pGrainAlloc = (PVMDKGRAINALLOCASYNC)RTMemAllocZ(sizeof(VMDKGRAINALLOCASYNC));
4832	if (!pGrainAlloc)
4833	return VERR_NO_MEMORY;
4834
4835	pGrainAlloc->pExtent = pExtent;
4836	pGrainAlloc->uSector = uSector;
4837
4838	uGDIndex = uSector / pExtent->cSectorsPerGDE;
4839	if (uGDIndex >= pExtent->cGDEntries)
4840	{
4841	RTMemFree(pGrainAlloc);
4842	return VERR_OUT_OF_RANGE;
4843	}
4844	uGTSector = pExtent->pGD[uGDIndex];
4845	if (pExtent->pRGD)
4846	uRGTSector = pExtent->pRGD[uGDIndex];
4847	else
4848	uRGTSector = 0; /*< avoid compiler warning /
4849	if (!uGTSector)
4850	{
4851	LogFlow(("Allocating new grain table\n"));
4852
4853	/* There is no grain table referenced by this grain directory
4854	* entry. So there is absolutely no data in this area. Allocate
4855	* a new grain table and put the reference to it in the GDs. */
4856	uFileOffset = pExtent->uAppendPosition;
4857	if (!uFileOffset)
4858	{
4859	RTMemFree(pGrainAlloc);
4860	return VERR_INTERNAL_ERROR;
4861	}
4862	Assert(!(uFileOffset % 512));
4863
4864	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
4865	uGTSector = VMDK_BYTE2SECTOR(uFileOffset);
4866
4867	/* Normally the grain table is preallocated for hosted sparse extents
4868	* that support more than 32 bit sector numbers. So this shouldn't
4869	* ever happen on a valid extent. */
4870	if (uGTSector > UINT32_MAX)
4871	{
4872	RTMemFree(pGrainAlloc);
4873	return VERR_VD_VMDK_INVALID_HEADER;
4874	}
4875
4876	/* Write grain table by writing the required number of grain table
4877	* cache chunks. Allocate memory dynamically here or we flood the
4878	* metadata cache with very small entries. */
4879	size_t cbGTDataTmp = pExtent->cGTEntries * sizeof(uint32_t);
4880	uint32_t paGTDataTmp = (uint32_t )RTMemTmpAllocZ(cbGTDataTmp);
4881
4882	if (!paGTDataTmp)
4883	{
4884	RTMemFree(pGrainAlloc);
4885	return VERR_NO_MEMORY;
4886	}
4887
4888	memset(paGTDataTmp, '\0', cbGTDataTmp);
4889	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
4890	VMDK_SECTOR2BYTE(uGTSector),
4891	paGTDataTmp, cbGTDataTmp, pIoCtx,
4892	vmdkAllocGrainComplete, pGrainAlloc);
4893	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
4894	pGrainAlloc->cIoXfersPending++;
4895	else if (RT_FAILURE(rc))
4896	{
4897	RTMemTmpFree(paGTDataTmp);
4898	RTMemFree(pGrainAlloc);
4899	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write grain table allocation in '%s'"), pExtent->pszFullname);
4900	}
4901	pExtent->uAppendPosition = RT_ALIGN_64( pExtent->uAppendPosition
4902	+ cbGTDataTmp, 512);
4903
4904	if (pExtent->pRGD)
4905	{
4906	AssertReturn(!uRGTSector, VERR_VD_VMDK_INVALID_HEADER);
4907	uFileOffset = pExtent->uAppendPosition;
4908	if (!uFileOffset)
4909	return VERR_INTERNAL_ERROR;
4910	Assert(!(uFileOffset % 512));
4911	uRGTSector = VMDK_BYTE2SECTOR(uFileOffset);
4912
4913	/* Normally the redundant grain table is preallocated for hosted
4914	* sparse extents that support more than 32 bit sector numbers. So
4915	* this shouldn't ever happen on a valid extent. */
4916	if (uRGTSector > UINT32_MAX)
4917	{
4918	RTMemTmpFree(paGTDataTmp);
4919	return VERR_VD_VMDK_INVALID_HEADER;
4920	}
4921
4922	/* Write grain table by writing the required number of grain table
4923	* cache chunks. Allocate memory dynamically here or we flood the
4924	* metadata cache with very small entries. */
4925	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
4926	VMDK_SECTOR2BYTE(uRGTSector),
4927	paGTDataTmp, cbGTDataTmp, pIoCtx,
4928	vmdkAllocGrainComplete, pGrainAlloc);
4929	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
4930	pGrainAlloc->cIoXfersPending++;
4931	else if (RT_FAILURE(rc))
4932	{
4933	RTMemTmpFree(paGTDataTmp);
4934	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write backup grain table allocation in '%s'"), pExtent->pszFullname);
4935	}
4936
4937	pExtent->uAppendPosition = pExtent->uAppendPosition + cbGTDataTmp;
4938	}
4939
4940	RTMemTmpFree(paGTDataTmp);
4941
4942	/* Update the grain directory on disk (doing it before writing the
4943	* grain table will result in a garbled extent if the operation is
4944	* aborted for some reason. Otherwise the worst that can happen is
4945	* some unused sectors in the extent. */
4946	uint32_t uGTSectorLE = RT_H2LE_U64(uGTSector);
4947	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
4948	VMDK_SECTOR2BYTE(pExtent->uSectorGD) + uGDIndex * sizeof(uGTSectorLE),
4949	&uGTSectorLE, sizeof(uGTSectorLE), pIoCtx,
4950	vmdkAllocGrainComplete, pGrainAlloc);
4951	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
4952	pGrainAlloc->cIoXfersPending++;
4953	else if (RT_FAILURE(rc))
4954	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write grain directory entry in '%s'"), pExtent->pszFullname);
4955	if (pExtent->pRGD)
4956	{
4957	uint32_t uRGTSectorLE = RT_H2LE_U64(uRGTSector);
4958	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
4959	VMDK_SECTOR2BYTE(pExtent->uSectorRGD) + uGDIndex * sizeof(uGTSectorLE),
4960	&uRGTSectorLE, sizeof(uRGTSectorLE), pIoCtx,
4961	vmdkAllocGrainComplete, pGrainAlloc);
4962	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
4963	pGrainAlloc->cIoXfersPending++;
4964	else if (RT_FAILURE(rc))
4965	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write backup grain directory entry in '%s'"), pExtent->pszFullname);
4966	}
4967
4968	/* As the final step update the in-memory copy of the GDs. */
4969	pExtent->pGD[uGDIndex] = uGTSector;
4970	if (pExtent->pRGD)
4971	pExtent->pRGD[uGDIndex] = uRGTSector;
4972	}
4973
4974	LogFlow(("uGTSector=%llu uRGTSector=%llu\n", uGTSector, uRGTSector));
4975	pGrainAlloc->uGTSector = uGTSector;
4976	pGrainAlloc->uRGTSector = uRGTSector;
4977
4978	uFileOffset = pExtent->uAppendPosition;
4979	if (!uFileOffset)
4980	return VERR_INTERNAL_ERROR;
4981	Assert(!(uFileOffset % 512));
4982
4983	pGrainAlloc->uGrainOffset = uFileOffset;
4984
4985	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
4986	{
4987	AssertMsgReturn(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
4988	("Accesses to stream optimized images must be synchronous\n"),
4989	VERR_INVALID_STATE);
4990
4991	if (cbWrite != VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
4992	return vdIfError(pImage->pIfError, VERR_INTERNAL_ERROR, RT_SRC_POS, N_("VMDK: not enough data for a compressed data block in '%s'"), pExtent->pszFullname);
4993
4994	/* Invalidate cache, just in case some code incorrectly allows mixing
4995	* of reads and writes. Normally shouldn't be needed. */
4996	pExtent->uGrainSectorAbs = 0;
4997
4998	/* Write compressed data block and the markers. */
4999	uint32_t cbGrain = 0;
5000	size_t cbSeg = 0;
5001	RTSGSEG Segment;
5002	unsigned cSegments = 1;
5003
5004	cbSeg = vdIfIoIntIoCtxSegArrayCreate(pImage->pIfIo, pIoCtx, &Segment,
5005	&cSegments, cbWrite);
5006	Assert(cbSeg == cbWrite);
5007
5008	rc = vmdkFileDeflateSync(pImage, pExtent, uFileOffset,
5009	Segment.pvSeg, cbWrite, uSector, &cbGrain);
5010	if (RT_FAILURE(rc))
5011	{
5012	AssertRC(rc);
5013	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write allocated compressed data block in '%s'"), pExtent->pszFullname);
5014	}
5015	pExtent->uLastGrainAccess = uSector / pExtent->cSectorsPerGrain;
5016	pExtent->uAppendPosition += cbGrain;
5017	}
5018	else
5019	{
5020	/* Write the data. Always a full grain, or we're in big trouble. */
5021	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
5022	uFileOffset, pIoCtx, cbWrite,
5023	vmdkAllocGrainComplete, pGrainAlloc);
5024	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
5025	pGrainAlloc->cIoXfersPending++;
5026	else if (RT_FAILURE(rc))
5027	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write allocated data block in '%s'"), pExtent->pszFullname);
5028
5029	pExtent->uAppendPosition += cbWrite;
5030	}
5031
5032	rc = vmdkAllocGrainGTUpdate(pImage, pExtent, pIoCtx, pGrainAlloc);
5033
5034	if (!pGrainAlloc->cIoXfersPending)
5035	{
5036	/* Grain allocation completed. */
5037	RTMemFree(pGrainAlloc);
5038	}
5039
5040	LogFlowFunc(("leaving rc=%Rrc\n", rc));
5041
5042	return rc;
5043	}
5044
5045	/**
5046	* Internal. Reads the contents by sequentially going over the compressed
5047	* grains (hoping that they are in sequence).
5048	*/
5049	static int vmdkStreamReadSequential(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
5050	uint64_t uSector, PVDIOCTX pIoCtx,
5051	uint64_t cbRead)
5052	{
5053	int rc;
5054
5055	LogFlowFunc(("pImage=%#p pExtent=%#p uSector=%llu pIoCtx=%#p cbRead=%llu\n",
5056	pImage, pExtent, uSector, pIoCtx, cbRead));
5057
5058	AssertMsgReturn(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
5059	("Async I/O not supported for sequential stream optimized images\n"),
5060	VERR_INVALID_STATE);
5061
5062	/* Do not allow to go back. */
5063	uint32_t uGrain = uSector / pExtent->cSectorsPerGrain;
5064	if (uGrain < pExtent->uLastGrainAccess)
5065	return VERR_VD_VMDK_INVALID_STATE;
5066	pExtent->uLastGrainAccess = uGrain;
5067
5068	/* After a previous error do not attempt to recover, as it would need
5069	* seeking (in the general case backwards which is forbidden). */
5070	if (!pExtent->uGrainSectorAbs)
5071	return VERR_VD_VMDK_INVALID_STATE;
5072
5073	/* Check if we need to read something from the image or if what we have
5074	* in the buffer is good to fulfill the request. */
5075	if (!pExtent->cbGrainStreamRead \|\| uGrain > pExtent->uGrain)
5076	{
5077	uint32_t uGrainSectorAbs = pExtent->uGrainSectorAbs
5078	+ VMDK_BYTE2SECTOR(pExtent->cbGrainStreamRead);
5079
5080	/* Get the marker from the next data block - and skip everything which
5081	* is not a compressed grain. If it's a compressed grain which is for
5082	* the requested sector (or after), read it. */
5083	VMDKMARKER Marker;
5084	do
5085	{
5086	RT_ZERO(Marker);
5087	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
5088	VMDK_SECTOR2BYTE(uGrainSectorAbs),
5089	&Marker, RT_OFFSETOF(VMDKMARKER, uType));
5090	if (RT_FAILURE(rc))
5091	return rc;
5092	Marker.uSector = RT_LE2H_U64(Marker.uSector);
5093	Marker.cbSize = RT_LE2H_U32(Marker.cbSize);
5094
5095	if (Marker.cbSize == 0)
5096	{
5097	/* A marker for something else than a compressed grain. */
5098	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
5099	VMDK_SECTOR2BYTE(uGrainSectorAbs)
5100	+ RT_OFFSETOF(VMDKMARKER, uType),
5101	&Marker.uType, sizeof(Marker.uType));
5102	if (RT_FAILURE(rc))
5103	return rc;
5104	Marker.uType = RT_LE2H_U32(Marker.uType);
5105	switch (Marker.uType)
5106	{
5107	case VMDK_MARKER_EOS:
5108	uGrainSectorAbs++;
5109	/* Read (or mostly skip) to the end of file. Uses the
5110	* Marker (LBA sector) as it is unused anyway. This
5111	* makes sure that really everything is read in the
5112	* success case. If this read fails it means the image
5113	* is truncated, but this is harmless so ignore. */
5114	vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
5115	VMDK_SECTOR2BYTE(uGrainSectorAbs)
5116	+ 511,
5117	&Marker.uSector, 1);
5118	break;
5119	case VMDK_MARKER_GT:
5120	uGrainSectorAbs += 1 + VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
5121	break;
5122	case VMDK_MARKER_GD:
5123	uGrainSectorAbs += 1 + VMDK_BYTE2SECTOR(RT_ALIGN(pExtent->cGDEntries * sizeof(uint32_t), 512));
5124	break;
5125	case VMDK_MARKER_FOOTER:
5126	uGrainSectorAbs += 2;
5127	break;
5128	case VMDK_MARKER_UNSPECIFIED:
5129	/* Skip over the contents of the unspecified marker
5130	* type 4 which exists in some vSphere created files. */
5131	/** @todo figure out what the payload means. */
5132	uGrainSectorAbs += 1;
5133	break;
5134	default:
5135	AssertMsgFailed(("VMDK: corrupted marker, type=%#x\n", Marker.uType));
5136	pExtent->uGrainSectorAbs = 0;
5137	return VERR_VD_VMDK_INVALID_STATE;
5138	}
5139	pExtent->cbGrainStreamRead = 0;
5140	}
5141	else
5142	{
5143	/* A compressed grain marker. If it is at/after what we're
5144	* interested in read and decompress data. */
5145	if (uSector > Marker.uSector + pExtent->cSectorsPerGrain)
5146	{
5147	uGrainSectorAbs += VMDK_BYTE2SECTOR(RT_ALIGN(Marker.cbSize + RT_OFFSETOF(VMDKMARKER, uType), 512));
5148	continue;
5149	}
5150	uint64_t uLBA = 0;
5151	uint32_t cbGrainStreamRead = 0;
5152	rc = vmdkFileInflateSync(pImage, pExtent,
5153	VMDK_SECTOR2BYTE(uGrainSectorAbs),
5154	pExtent->pvGrain,
5155	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
5156	&Marker, &uLBA, &cbGrainStreamRead);
5157	if (RT_FAILURE(rc))
5158	{
5159	pExtent->uGrainSectorAbs = 0;
5160	return rc;
5161	}
5162	if ( pExtent->uGrain
5163	&& uLBA / pExtent->cSectorsPerGrain <= pExtent->uGrain)
5164	{
5165	pExtent->uGrainSectorAbs = 0;
5166	return VERR_VD_VMDK_INVALID_STATE;
5167	}
5168	pExtent->uGrain = uLBA / pExtent->cSectorsPerGrain;
5169	pExtent->cbGrainStreamRead = cbGrainStreamRead;
5170	break;
5171	}
5172	} while (Marker.uType != VMDK_MARKER_EOS);
5173
5174	pExtent->uGrainSectorAbs = uGrainSectorAbs;
5175
5176	if (!pExtent->cbGrainStreamRead && Marker.uType == VMDK_MARKER_EOS)
5177	{
5178	pExtent->uGrain = UINT32_MAX;
5179	/* Must set a non-zero value for pExtent->cbGrainStreamRead or
5180	* the next read would try to get more data, and we're at EOF. */
5181	pExtent->cbGrainStreamRead = 1;
5182	}
5183	}
5184
5185	if (pExtent->uGrain > uSector / pExtent->cSectorsPerGrain)
5186	{
5187	/* The next data block we have is not for this area, so just return
5188	* that there is no data. */
5189	LogFlowFunc(("returns VERR_VD_BLOCK_FREE\n"));
5190	return VERR_VD_BLOCK_FREE;
5191	}
5192
5193	uint32_t uSectorInGrain = uSector % pExtent->cSectorsPerGrain;
5194	vdIfIoIntIoCtxCopyTo(pImage->pIfIo, pIoCtx,
5195	(uint8_t *)pExtent->pvGrain + VMDK_SECTOR2BYTE(uSectorInGrain),
5196	cbRead);
5197	LogFlowFunc(("returns VINF_SUCCESS\n"));
5198	return VINF_SUCCESS;
5199	}
5200
5201	/**
5202	* Replaces a fragment of a string with the specified string.
5203	*
5204	* @returns Pointer to the allocated UTF-8 string.
5205	* @param pszWhere UTF-8 string to search in.
5206	* @param pszWhat UTF-8 string to search for.
5207	* @param pszByWhat UTF-8 string to replace the found string with.
5208	*/
5209	static char vmdkStrReplace(const char pszWhere, const char *pszWhat,
5210	const char *pszByWhat)
5211	{
5212	AssertPtr(pszWhere);
5213	AssertPtr(pszWhat);
5214	AssertPtr(pszByWhat);
5215	const char *pszFoundStr = strstr(pszWhere, pszWhat);
5216	if (!pszFoundStr)
5217	return NULL;
5218	size_t cFinal = strlen(pszWhere) + 1 + strlen(pszByWhat) - strlen(pszWhat);
5219	char pszNewStr = (char )RTMemAlloc(cFinal);
5220	if (pszNewStr)
5221	{
5222	char *pszTmp = pszNewStr;
5223	memcpy(pszTmp, pszWhere, pszFoundStr - pszWhere);
5224	pszTmp += pszFoundStr - pszWhere;
5225	memcpy(pszTmp, pszByWhat, strlen(pszByWhat));
5226	pszTmp += strlen(pszByWhat);
5227	strcpy(pszTmp, pszFoundStr + strlen(pszWhat));
5228	}
5229	return pszNewStr;
5230	}
5231
5232
5233	/** @copydoc VBOXHDDBACKEND::pfnCheckIfValid */
5234	static DECLCALLBACK(int) vmdkCheckIfValid(const char *pszFilename, PVDINTERFACE pVDIfsDisk,
5235	PVDINTERFACE pVDIfsImage, VDTYPE *penmType)
5236	{
5237	LogFlowFunc(("pszFilename=\"%s\" pVDIfsDisk=%#p pVDIfsImage=%#p penmType=%#p\n",
5238	pszFilename, pVDIfsDisk, pVDIfsImage, penmType));
5239	int rc = VINF_SUCCESS;
5240	PVMDKIMAGE pImage;
5241
5242	if ( !pszFilename
5243	\|\| !*pszFilename
5244	\|\| strchr(pszFilename, '"'))
5245	{
5246	rc = VERR_INVALID_PARAMETER;
5247	goto out;
5248	}
5249
5250	pImage = (PVMDKIMAGE)RTMemAllocZ(sizeof(VMDKIMAGE));
5251	if (!pImage)
5252	{
5253	rc = VERR_NO_MEMORY;
5254	goto out;
5255	}
5256	pImage->pszFilename = pszFilename;
5257	pImage->pFile = NULL;
5258	pImage->pExtents = NULL;
5259	pImage->pFiles = NULL;
5260	pImage->pGTCache = NULL;
5261	pImage->pDescData = NULL;
5262	pImage->pVDIfsDisk = pVDIfsDisk;
5263	pImage->pVDIfsImage = pVDIfsImage;
5264	/** @todo speed up this test open (VD_OPEN_FLAGS_INFO) by skipping as
5265	* much as possible in vmdkOpenImage. */
5266	rc = vmdkOpenImage(pImage, VD_OPEN_FLAGS_INFO \| VD_OPEN_FLAGS_READONLY);
5267	vmdkFreeImage(pImage, false);
5268	RTMemFree(pImage);
5269
5270	if (RT_SUCCESS(rc))
5271	*penmType = VDTYPE_HDD;
5272
5273	out:
5274	LogFlowFunc(("returns %Rrc\n", rc));
5275	return rc;
5276	}
5277
5278	/** @copydoc VBOXHDDBACKEND::pfnOpen */
5279	static DECLCALLBACK(int) vmdkOpen(const char *pszFilename, unsigned uOpenFlags,
5280	PVDINTERFACE pVDIfsDisk, PVDINTERFACE pVDIfsImage,
5281	VDTYPE enmType, void **ppBackendData)
5282	{
5283	LogFlowFunc(("pszFilename=\"%s\" uOpenFlags=%#x pVDIfsDisk=%#p pVDIfsImage=%#p enmType=%u ppBackendData=%#p\n", pszFilename, uOpenFlags, pVDIfsDisk, pVDIfsImage, enmType, ppBackendData));
5284	int rc;
5285	PVMDKIMAGE pImage;
5286
5287	NOREF(enmType); /*< @todo r=klaus make use of the type info. /
5288
5289	/* Check open flags. All valid flags are supported. */
5290	if (uOpenFlags & ~VD_OPEN_FLAGS_MASK)
5291	{
5292	rc = VERR_INVALID_PARAMETER;
5293	goto out;
5294	}
5295
5296	/* Check remaining arguments. */
5297	if ( !VALID_PTR(pszFilename)
5298	\|\| !*pszFilename
5299	\|\| strchr(pszFilename, '"'))
5300	{
5301	rc = VERR_INVALID_PARAMETER;
5302	goto out;
5303	}
5304
5305	pImage = (PVMDKIMAGE)RTMemAllocZ(sizeof(VMDKIMAGE));
5306	if (!pImage)
5307	{
5308	rc = VERR_NO_MEMORY;
5309	goto out;
5310	}
5311	pImage->pszFilename = pszFilename;
5312	pImage->pFile = NULL;
5313	pImage->pExtents = NULL;
5314	pImage->pFiles = NULL;
5315	pImage->pGTCache = NULL;
5316	pImage->pDescData = NULL;
5317	pImage->pVDIfsDisk = pVDIfsDisk;
5318	pImage->pVDIfsImage = pVDIfsImage;
5319
5320	rc = vmdkOpenImage(pImage, uOpenFlags);
5321	if (RT_SUCCESS(rc))
5322	*ppBackendData = pImage;
5323	else
5324	RTMemFree(pImage);
5325
5326	out:
5327	LogFlowFunc(("returns %Rrc (pBackendData=%#p)\n", rc, *ppBackendData));
5328	return rc;
5329	}
5330
5331	/** @copydoc VBOXHDDBACKEND::pfnCreate */
5332	static DECLCALLBACK(int) vmdkCreate(const char *pszFilename, uint64_t cbSize,
5333	unsigned uImageFlags, const char *pszComment,
5334	PCVDGEOMETRY pPCHSGeometry, PCVDGEOMETRY pLCHSGeometry,
5335	PCRTUUID pUuid, unsigned uOpenFlags,
5336	unsigned uPercentStart, unsigned uPercentSpan,
5337	PVDINTERFACE pVDIfsDisk, PVDINTERFACE pVDIfsImage,
5338	PVDINTERFACE pVDIfsOperation, VDTYPE enmType,
5339	void **ppBackendData)
5340	{
5341	LogFlowFunc(("pszFilename=\"%s\" cbSize=%llu uImageFlags=%#x pszComment=\"%s\" pPCHSGeometry=%#p pLCHSGeometry=%#p Uuid=%RTuuid uOpenFlags=%#x uPercentStart=%u uPercentSpan=%u pVDIfsDisk=%#p pVDIfsImage=%#p pVDIfsOperation=%#p enmType=%u ppBackendData=%#p\n",
5342	pszFilename, cbSize, uImageFlags, pszComment, pPCHSGeometry, pLCHSGeometry, pUuid, uOpenFlags, uPercentStart, uPercentSpan, pVDIfsDisk, pVDIfsImage, pVDIfsOperation, enmType, ppBackendData));
5343	int rc;
5344	PVMDKIMAGE pImage;
5345
5346	PFNVDPROGRESS pfnProgress = NULL;
5347	void *pvUser = NULL;
5348	PVDINTERFACEPROGRESS pIfProgress = VDIfProgressGet(pVDIfsOperation);
5349	if (pIfProgress)
5350	{
5351	pfnProgress = pIfProgress->pfnProgress;
5352	pvUser = pIfProgress->Core.pvUser;
5353	}
5354
5355	/* Check the image flags. */
5356	if ((uImageFlags & ~VD_VMDK_IMAGE_FLAGS_MASK) != 0)
5357	{
5358	rc = VERR_VD_INVALID_TYPE;
5359	goto out;
5360	}
5361
5362	/* Check the VD container type. */
5363	if (enmType != VDTYPE_HDD)
5364	{
5365	rc = VERR_VD_INVALID_TYPE;
5366	goto out;
5367	}
5368
5369	/* Check open flags. All valid flags are supported. */
5370	if (uOpenFlags & ~VD_OPEN_FLAGS_MASK)
5371	{
5372	rc = VERR_INVALID_PARAMETER;
5373	goto out;
5374	}
5375
5376	/* Check size. Maximum 2TB-64K for sparse images, otherwise unlimited. */
5377	if ( !cbSize
5378	\|\| (!(uImageFlags & VD_IMAGE_FLAGS_FIXED) && cbSize >= _1T * 2 - _64K))
5379	{
5380	rc = VERR_VD_INVALID_SIZE;
5381	goto out;
5382	}
5383
5384	/* Check remaining arguments. */
5385	if ( !VALID_PTR(pszFilename)
5386	\|\| !*pszFilename
5387	\|\| strchr(pszFilename, '"')
5388	\|\| !VALID_PTR(pPCHSGeometry)
5389	\|\| !VALID_PTR(pLCHSGeometry)
5390	#ifndef VBOX_WITH_VMDK_ESX
5391	\|\| ( uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX
5392	&& !(uImageFlags & VD_IMAGE_FLAGS_FIXED))
5393	#endif
5394	\|\| ( (uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5395	&& (uImageFlags & ~(VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED \| VD_IMAGE_FLAGS_DIFF))))
5396	{
5397	rc = VERR_INVALID_PARAMETER;
5398	goto out;
5399	}
5400
5401	pImage = (PVMDKIMAGE)RTMemAllocZ(sizeof(VMDKIMAGE));
5402	if (!pImage)
5403	{
5404	rc = VERR_NO_MEMORY;
5405	goto out;
5406	}
5407	pImage->pszFilename = pszFilename;
5408	pImage->pFile = NULL;
5409	pImage->pExtents = NULL;
5410	pImage->pFiles = NULL;
5411	pImage->pGTCache = NULL;
5412	pImage->pDescData = NULL;
5413	pImage->pVDIfsDisk = pVDIfsDisk;
5414	pImage->pVDIfsImage = pVDIfsImage;
5415	/* Descriptors for split images can be pretty large, especially if the
5416	* filename is long. So prepare for the worst, and allocate quite some
5417	* memory for the descriptor in this case. */
5418	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
5419	pImage->cbDescAlloc = VMDK_SECTOR2BYTE(200);
5420	else
5421	pImage->cbDescAlloc = VMDK_SECTOR2BYTE(20);
5422	pImage->pDescData = (char *)RTMemAllocZ(pImage->cbDescAlloc);
5423	if (!pImage->pDescData)
5424	{
5425	RTMemFree(pImage);
5426	rc = VERR_NO_MEMORY;
5427	goto out;
5428	}
5429
5430	rc = vmdkCreateImage(pImage, cbSize, uImageFlags, pszComment,
5431	pPCHSGeometry, pLCHSGeometry, pUuid,
5432	pfnProgress, pvUser, uPercentStart, uPercentSpan);
5433	if (RT_SUCCESS(rc))
5434	{
5435	/* So far the image is opened in read/write mode. Make sure the
5436	* image is opened in read-only mode if the caller requested that. */
5437	if (uOpenFlags & VD_OPEN_FLAGS_READONLY)
5438	{
5439	vmdkFreeImage(pImage, false);
5440	rc = vmdkOpenImage(pImage, uOpenFlags);
5441	if (RT_FAILURE(rc))
5442	goto out;
5443	}
5444	*ppBackendData = pImage;
5445	}
5446	else
5447	{
5448	RTMemFree(pImage->pDescData);
5449	RTMemFree(pImage);
5450	}
5451
5452	out:
5453	LogFlowFunc(("returns %Rrc (pBackendData=%#p)\n", rc, *ppBackendData));
5454	return rc;
5455	}
5456
5457	/** @copydoc VBOXHDDBACKEND::pfnRename */
5458	static DECLCALLBACK(int) vmdkRename(void pBackendData, const char pszFilename)
5459	{
5460	LogFlowFunc(("pBackendData=%#p pszFilename=%#p\n", pBackendData, pszFilename));
5461
5462	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
5463	int rc = VINF_SUCCESS;
5464	char **apszOldName = NULL;
5465	char **apszNewName = NULL;
5466	char **apszNewLines = NULL;
5467	char *pszOldDescName = NULL;
5468	bool fImageFreed = false;
5469	bool fEmbeddedDesc = false;
5470	unsigned cExtents = 0;
5471	char *pszNewBaseName = NULL;
5472	char *pszOldBaseName = NULL;
5473	char *pszNewFullName = NULL;
5474	char *pszOldFullName = NULL;
5475	const char *pszOldImageName;
5476	unsigned i, line;
5477	VMDKDESCRIPTOR DescriptorCopy;
5478	VMDKEXTENT ExtentCopy;
5479
5480	memset(&DescriptorCopy, 0, sizeof(DescriptorCopy));
5481
5482	/* Check arguments. */
5483	if ( !pImage
5484	\|\| (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
5485	\|\| !VALID_PTR(pszFilename)
5486	\|\| !*pszFilename)
5487	{
5488	rc = VERR_INVALID_PARAMETER;
5489	goto out;
5490	}
5491
5492	cExtents = pImage->cExtents;
5493
5494	/*
5495	* Allocate an array to store both old and new names of renamed files
5496	* in case we have to roll back the changes. Arrays are initialized
5497	* with zeros. We actually save stuff when and if we change it.
5498	*/
5499	apszOldName = (char *)RTMemTmpAllocZ((cExtents + 1) sizeof(char*));
5500	apszNewName = (char *)RTMemTmpAllocZ((cExtents + 1) sizeof(char*));
5501	apszNewLines = (char *)RTMemTmpAllocZ((cExtents) sizeof(char*));
5502	if (!apszOldName \|\| !apszNewName \|\| !apszNewLines)
5503	{
5504	rc = VERR_NO_MEMORY;
5505	goto out;
5506	}
5507
5508	/* Save the descriptor size and position. */
5509	if (pImage->pDescData)
5510	{
5511	/* Separate descriptor file. */
5512	fEmbeddedDesc = false;
5513	}
5514	else
5515	{
5516	/* Embedded descriptor file. */
5517	ExtentCopy = pImage->pExtents[0];
5518	fEmbeddedDesc = true;
5519	}
5520	/* Save the descriptor content. */
5521	DescriptorCopy.cLines = pImage->Descriptor.cLines;
5522	for (i = 0; i < DescriptorCopy.cLines; i++)
5523	{
5524	DescriptorCopy.aLines[i] = RTStrDup(pImage->Descriptor.aLines[i]);
5525	if (!DescriptorCopy.aLines[i])
5526	{
5527	rc = VERR_NO_MEMORY;
5528	goto out;
5529	}
5530	}
5531
5532	/* Prepare both old and new base names used for string replacement. */
5533	pszNewBaseName = RTStrDup(RTPathFilename(pszFilename));
5534	RTPathStripSuffix(pszNewBaseName);
5535	pszOldBaseName = RTStrDup(RTPathFilename(pImage->pszFilename));
5536	RTPathStripSuffix(pszOldBaseName);
5537	/* Prepare both old and new full names used for string replacement. */
5538	pszNewFullName = RTStrDup(pszFilename);
5539	RTPathStripSuffix(pszNewFullName);
5540	pszOldFullName = RTStrDup(pImage->pszFilename);
5541	RTPathStripSuffix(pszOldFullName);
5542
5543	/* --- Up to this point we have not done any damage yet. --- */
5544
5545	/* Save the old name for easy access to the old descriptor file. */
5546	pszOldDescName = RTStrDup(pImage->pszFilename);
5547	/* Save old image name. */
5548	pszOldImageName = pImage->pszFilename;
5549
5550	/* Update the descriptor with modified extent names. */
5551	for (i = 0, line = pImage->Descriptor.uFirstExtent;
5552	i < cExtents;
5553	i++, line = pImage->Descriptor.aNextLines[line])
5554	{
5555	/* Assume that vmdkStrReplace will fail. */
5556	rc = VERR_NO_MEMORY;
5557	/* Update the descriptor. */
5558	apszNewLines[i] = vmdkStrReplace(pImage->Descriptor.aLines[line],
5559	pszOldBaseName, pszNewBaseName);
5560	if (!apszNewLines[i])
5561	goto rollback;
5562	pImage->Descriptor.aLines[line] = apszNewLines[i];
5563	}
5564	/* Make sure the descriptor gets written back. */
5565	pImage->Descriptor.fDirty = true;
5566	/* Flush the descriptor now, in case it is embedded. */
5567	vmdkFlushImage(pImage, NULL);
5568
5569	/* Close and rename/move extents. */
5570	for (i = 0; i < cExtents; i++)
5571	{
5572	PVMDKEXTENT pExtent = &pImage->pExtents[i];
5573	/* Compose new name for the extent. */
5574	apszNewName[i] = vmdkStrReplace(pExtent->pszFullname,
5575	pszOldFullName, pszNewFullName);
5576	if (!apszNewName[i])
5577	goto rollback;
5578	/* Close the extent file. */
5579	rc = vmdkFileClose(pImage, &pExtent->pFile, false);
5580	if (RT_FAILURE(rc))
5581	goto rollback;
5582
5583	/* Rename the extent file. */
5584	rc = vdIfIoIntFileMove(pImage->pIfIo, pExtent->pszFullname, apszNewName[i], 0);
5585	if (RT_FAILURE(rc))
5586	goto rollback;
5587	/* Remember the old name. */
5588	apszOldName[i] = RTStrDup(pExtent->pszFullname);
5589	}
5590	/* Release all old stuff. */
5591	rc = vmdkFreeImage(pImage, false);
5592	if (RT_FAILURE(rc))
5593	goto rollback;
5594
5595	fImageFreed = true;
5596
5597	/* Last elements of new/old name arrays are intended for
5598	* storing descriptor's names.
5599	*/
5600	apszNewName[cExtents] = RTStrDup(pszFilename);
5601	/* Rename the descriptor file if it's separate. */
5602	if (!fEmbeddedDesc)
5603	{
5604	rc = vdIfIoIntFileMove(pImage->pIfIo, pImage->pszFilename, apszNewName[cExtents], 0);
5605	if (RT_FAILURE(rc))
5606	goto rollback;
5607	/* Save old name only if we may need to change it back. */
5608	apszOldName[cExtents] = RTStrDup(pszFilename);
5609	}
5610
5611	/* Update pImage with the new information. */
5612	pImage->pszFilename = pszFilename;
5613
5614	/* Open the new image. */
5615	rc = vmdkOpenImage(pImage, pImage->uOpenFlags);
5616	if (RT_SUCCESS(rc))
5617	goto out;
5618
5619	rollback:
5620	/* Roll back all changes in case of failure. */
5621	if (RT_FAILURE(rc))
5622	{
5623	int rrc;
5624	if (!fImageFreed)
5625	{
5626	/*
5627	* Some extents may have been closed, close the rest. We will
5628	* re-open the whole thing later.
5629	*/
5630	vmdkFreeImage(pImage, false);
5631	}
5632	/* Rename files back. */
5633	for (i = 0; i <= cExtents; i++)
5634	{
5635	if (apszOldName[i])
5636	{
5637	rrc = vdIfIoIntFileMove(pImage->pIfIo, apszNewName[i], apszOldName[i], 0);
5638	AssertRC(rrc);
5639	}
5640	}
5641	/* Restore the old descriptor. */
5642	PVMDKFILE pFile;
5643	rrc = vmdkFileOpen(pImage, &pFile, pszOldDescName,
5644	VDOpenFlagsToFileOpenFlags(VD_OPEN_FLAGS_NORMAL,
5645	false /* fCreate */));
5646	AssertRC(rrc);
5647	if (fEmbeddedDesc)
5648	{
5649	ExtentCopy.pFile = pFile;
5650	pImage->pExtents = &ExtentCopy;
5651	}
5652	else
5653	{
5654	/* Shouldn't be null for separate descriptor.
5655	* There will be no access to the actual content.
5656	*/
5657	pImage->pDescData = pszOldDescName;
5658	pImage->pFile = pFile;
5659	}
5660	pImage->Descriptor = DescriptorCopy;
5661	vmdkWriteDescriptor(pImage, NULL);
5662	vmdkFileClose(pImage, &pFile, false);
5663	/* Get rid of the stuff we implanted. */
5664	pImage->pExtents = NULL;
5665	pImage->pFile = NULL;
5666	pImage->pDescData = NULL;
5667	/* Re-open the image back. */
5668	pImage->pszFilename = pszOldImageName;
5669	rrc = vmdkOpenImage(pImage, pImage->uOpenFlags);
5670	AssertRC(rrc);
5671	}
5672
5673	out:
5674	for (i = 0; i < DescriptorCopy.cLines; i++)
5675	if (DescriptorCopy.aLines[i])
5676	RTStrFree(DescriptorCopy.aLines[i]);
5677	if (apszOldName)
5678	{
5679	for (i = 0; i <= cExtents; i++)
5680	if (apszOldName[i])
5681	RTStrFree(apszOldName[i]);
5682	RTMemTmpFree(apszOldName);
5683	}
5684	if (apszNewName)
5685	{
5686	for (i = 0; i <= cExtents; i++)
5687	if (apszNewName[i])
5688	RTStrFree(apszNewName[i]);
5689	RTMemTmpFree(apszNewName);
5690	}
5691	if (apszNewLines)
5692	{
5693	for (i = 0; i < cExtents; i++)
5694	if (apszNewLines[i])
5695	RTStrFree(apszNewLines[i]);
5696	RTMemTmpFree(apszNewLines);
5697	}
5698	if (pszOldDescName)
5699	RTStrFree(pszOldDescName);
5700	if (pszOldBaseName)
5701	RTStrFree(pszOldBaseName);
5702	if (pszNewBaseName)
5703	RTStrFree(pszNewBaseName);
5704	if (pszOldFullName)
5705	RTStrFree(pszOldFullName);
5706	if (pszNewFullName)
5707	RTStrFree(pszNewFullName);
5708	LogFlowFunc(("returns %Rrc\n", rc));
5709	return rc;
5710	}
5711
5712	/** @copydoc VBOXHDDBACKEND::pfnClose */
5713	static DECLCALLBACK(int) vmdkClose(void *pBackendData, bool fDelete)
5714	{
5715	LogFlowFunc(("pBackendData=%#p fDelete=%d\n", pBackendData, fDelete));
5716	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
5717	int rc;
5718
5719	rc = vmdkFreeImage(pImage, fDelete);
5720	RTMemFree(pImage);
5721
5722	LogFlowFunc(("returns %Rrc\n", rc));
5723	return rc;
5724	}
5725
5726	/** @copydoc VBOXHDDBACKEND::pfnRead */
5727	static DECLCALLBACK(int) vmdkRead(void *pBackendData, uint64_t uOffset, size_t cbToRead,
5728	PVDIOCTX pIoCtx, size_t *pcbActuallyRead)
5729	{
5730	LogFlowFunc(("pBackendData=%#p uOffset=%llu pIoCtx=%#p cbToRead=%zu pcbActuallyRead=%#p\n",
5731	pBackendData, uOffset, pIoCtx, cbToRead, pcbActuallyRead));
5732	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
5733	PVMDKEXTENT pExtent;
5734	uint64_t uSectorExtentRel;
5735	uint64_t uSectorExtentAbs;
5736	int rc;
5737
5738	AssertPtr(pImage);
5739	Assert(uOffset % 512 == 0);
5740	Assert(cbToRead % 512 == 0);
5741
5742	if ( uOffset + cbToRead > pImage->cbSize
5743	\|\| cbToRead == 0)
5744	{
5745	rc = VERR_INVALID_PARAMETER;
5746	goto out;
5747	}
5748
5749	rc = vmdkFindExtent(pImage, VMDK_BYTE2SECTOR(uOffset),
5750	&pExtent, &uSectorExtentRel);
5751	if (RT_FAILURE(rc))
5752	goto out;
5753
5754	/* Check access permissions as defined in the extent descriptor. */
5755	if (pExtent->enmAccess == VMDKACCESS_NOACCESS)
5756	{
5757	rc = VERR_VD_VMDK_INVALID_STATE;
5758	goto out;
5759	}
5760
5761	/* Clip read range to remain in this extent. */
5762	cbToRead = RT_MIN(cbToRead, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
5763
5764	/* Handle the read according to the current extent type. */
5765	switch (pExtent->enmType)
5766	{
5767	case VMDKETYPE_HOSTED_SPARSE:
5768	#ifdef VBOX_WITH_VMDK_ESX
5769	case VMDKETYPE_ESX_SPARSE:
5770	#endif /* VBOX_WITH_VMDK_ESX */
5771	rc = vmdkGetSector(pImage, pIoCtx, pExtent, uSectorExtentRel, &uSectorExtentAbs);
5772	if (RT_FAILURE(rc))
5773	goto out;
5774	/* Clip read range to at most the rest of the grain. */
5775	cbToRead = RT_MIN(cbToRead, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain - uSectorExtentRel % pExtent->cSectorsPerGrain));
5776	Assert(!(cbToRead % 512));
5777	if (uSectorExtentAbs == 0)
5778	{
5779	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5780	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
5781	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
5782	rc = VERR_VD_BLOCK_FREE;
5783	else
5784	rc = vmdkStreamReadSequential(pImage, pExtent,
5785	uSectorExtentRel,
5786	pIoCtx, cbToRead);
5787	}
5788	else
5789	{
5790	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5791	{
5792	AssertMsg(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
5793	("Async I/O is not supported for stream optimized VMDK's\n"));
5794
5795	uint32_t uSectorInGrain = uSectorExtentRel % pExtent->cSectorsPerGrain;
5796	uSectorExtentAbs -= uSectorInGrain;
5797	if (pExtent->uGrainSectorAbs != uSectorExtentAbs)
5798	{
5799	uint64_t uLBA = 0; /* gcc maybe uninitialized */
5800	rc = vmdkFileInflateSync(pImage, pExtent,
5801	VMDK_SECTOR2BYTE(uSectorExtentAbs),
5802	pExtent->pvGrain,
5803	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
5804	NULL, &uLBA, NULL);
5805	if (RT_FAILURE(rc))
5806	{
5807	pExtent->uGrainSectorAbs = 0;
5808	AssertRC(rc);
5809	goto out;
5810	}
5811	pExtent->uGrainSectorAbs = uSectorExtentAbs;
5812	pExtent->uGrain = uSectorExtentRel / pExtent->cSectorsPerGrain;
5813	Assert(uLBA == uSectorExtentRel);
5814	}
5815	vdIfIoIntIoCtxCopyTo(pImage->pIfIo, pIoCtx,
5816	(uint8_t *)pExtent->pvGrain
5817	+ VMDK_SECTOR2BYTE(uSectorInGrain),
5818	cbToRead);
5819	}
5820	else
5821	rc = vdIfIoIntFileReadUser(pImage->pIfIo, pExtent->pFile->pStorage,
5822	VMDK_SECTOR2BYTE(uSectorExtentAbs),
5823	pIoCtx, cbToRead);
5824	}
5825	break;
5826	case VMDKETYPE_VMFS:
5827	case VMDKETYPE_FLAT:
5828	rc = vdIfIoIntFileReadUser(pImage->pIfIo, pExtent->pFile->pStorage,
5829	VMDK_SECTOR2BYTE(uSectorExtentRel),
5830	pIoCtx, cbToRead);
5831	break;
5832	case VMDKETYPE_ZERO:
5833	size_t cbSet;
5834
5835	cbSet = vdIfIoIntIoCtxSet(pImage->pIfIo, pIoCtx, 0, cbToRead);
5836	Assert(cbSet == cbToRead);
5837
5838	rc = VINF_SUCCESS;
5839	break;
5840	}
5841	if (pcbActuallyRead)
5842	*pcbActuallyRead = cbToRead;
5843
5844	out:
5845	LogFlowFunc(("returns %Rrc\n", rc));
5846	return rc;
5847	}
5848
5849	/** @copydoc VBOXHDDBACKEND::pfnWrite */
5850	static DECLCALLBACK(int) vmdkWrite(void *pBackendData, uint64_t uOffset, size_t cbToWrite,
5851	PVDIOCTX pIoCtx, size_t pcbWriteProcess, size_t pcbPreRead,
5852	size_t *pcbPostRead, unsigned fWrite)
5853	{
5854	LogFlowFunc(("pBackendData=%#p uOffset=%llu pIoCtx=%#p cbToWrite=%zu pcbWriteProcess=%#p pcbPreRead=%#p pcbPostRead=%#p\n",
5855	pBackendData, uOffset, pIoCtx, cbToWrite, pcbWriteProcess, pcbPreRead, pcbPostRead));
5856	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
5857	PVMDKEXTENT pExtent;
5858	uint64_t uSectorExtentRel;
5859	uint64_t uSectorExtentAbs;
5860	int rc;
5861
5862	AssertPtr(pImage);
5863	Assert(uOffset % 512 == 0);
5864	Assert(cbToWrite % 512 == 0);
5865
5866	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
5867	{
5868	rc = VERR_VD_IMAGE_READ_ONLY;
5869	goto out;
5870	}
5871
5872	if (cbToWrite == 0)
5873	{
5874	rc = VERR_INVALID_PARAMETER;
5875	goto out;
5876	}
5877
5878	/* No size check here, will do that later when the extent is located.
5879	* There are sparse images out there which according to the spec are
5880	* invalid, because the total size is not a multiple of the grain size.
5881	* Also for sparse images which are stitched together in odd ways (not at
5882	* grain boundaries, and with the nominal size not being a multiple of the
5883	* grain size), this would prevent writing to the last grain. */
5884
5885	rc = vmdkFindExtent(pImage, VMDK_BYTE2SECTOR(uOffset),
5886	&pExtent, &uSectorExtentRel);
5887	if (RT_FAILURE(rc))
5888	goto out;
5889
5890	/* Check access permissions as defined in the extent descriptor. */
5891	if ( pExtent->enmAccess != VMDKACCESS_READWRITE
5892	&& ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5893	&& !pImage->pExtents[0].uAppendPosition
5894	&& pExtent->enmAccess != VMDKACCESS_READONLY))
5895	{
5896	rc = VERR_VD_VMDK_INVALID_STATE;
5897	goto out;
5898	}
5899
5900	/* Handle the write according to the current extent type. */
5901	switch (pExtent->enmType)
5902	{
5903	case VMDKETYPE_HOSTED_SPARSE:
5904	#ifdef VBOX_WITH_VMDK_ESX
5905	case VMDKETYPE_ESX_SPARSE:
5906	#endif /* VBOX_WITH_VMDK_ESX */
5907	rc = vmdkGetSector(pImage, pIoCtx, pExtent, uSectorExtentRel, &uSectorExtentAbs);
5908	if (RT_FAILURE(rc))
5909	goto out;
5910	/* Clip write range to at most the rest of the grain. */
5911	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain - uSectorExtentRel % pExtent->cSectorsPerGrain));
5912	if ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
5913	&& uSectorExtentRel < (uint64_t)pExtent->uLastGrainAccess * pExtent->cSectorsPerGrain)
5914	{
5915	rc = VERR_VD_VMDK_INVALID_WRITE;
5916	goto out;
5917	}
5918	if (uSectorExtentAbs == 0)
5919	{
5920	if (!(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
5921	{
5922	if (cbToWrite == VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
5923	{
5924	/* Full block write to a previously unallocated block.
5925	* Check if the caller wants to avoid the automatic alloc. */
5926	if (!(fWrite & VD_WRITE_NO_ALLOC))
5927	{
5928	/* Allocate GT and find out where to store the grain. */
5929	rc = vmdkAllocGrain(pImage, pExtent, pIoCtx,
5930	uSectorExtentRel, cbToWrite);
5931	}
5932	else
5933	rc = VERR_VD_BLOCK_FREE;
5934	*pcbPreRead = 0;
5935	*pcbPostRead = 0;
5936	}
5937	else
5938	{
5939	/* Clip write range to remain in this extent. */
5940	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
5941	*pcbPreRead = VMDK_SECTOR2BYTE(uSectorExtentRel % pExtent->cSectorsPerGrain);
5942	pcbPostRead = VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain) - cbToWrite - pcbPreRead;
5943	rc = VERR_VD_BLOCK_FREE;
5944	}
5945	}
5946	else
5947	{
5948	rc = vmdkStreamAllocGrain(pImage, pExtent,
5949	uSectorExtentRel,
5950	pIoCtx, cbToWrite);
5951	}
5952	}
5953	else
5954	{
5955	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5956	{
5957	/* A partial write to a streamOptimized image is simply
5958	* invalid. It requires rewriting already compressed data
5959	* which is somewhere between expensive and impossible. */
5960	rc = VERR_VD_VMDK_INVALID_STATE;
5961	pExtent->uGrainSectorAbs = 0;
5962	AssertRC(rc);
5963	}
5964	else
5965	{
5966	Assert(!(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED));
5967	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
5968	VMDK_SECTOR2BYTE(uSectorExtentAbs),
5969	pIoCtx, cbToWrite, NULL, NULL);
5970	}
5971	}
5972	break;
5973	case VMDKETYPE_VMFS:
5974	case VMDKETYPE_FLAT:
5975	/* Clip write range to remain in this extent. */
5976	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
5977	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
5978	VMDK_SECTOR2BYTE(uSectorExtentRel),
5979	pIoCtx, cbToWrite, NULL, NULL);
5980	break;
5981	case VMDKETYPE_ZERO:
5982	/* Clip write range to remain in this extent. */
5983	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
5984	break;
5985	}
5986
5987	if (pcbWriteProcess)
5988	*pcbWriteProcess = cbToWrite;
5989
5990	out:
5991	LogFlowFunc(("returns %Rrc\n", rc));
5992	return rc;
5993	}
5994
5995	/** @copydoc VBOXHDDBACKEND::pfnFlush */
5996	static DECLCALLBACK(int) vmdkFlush(void *pBackendData, PVDIOCTX pIoCtx)
5997	{
5998	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
5999
6000	return vmdkFlushImage(pImage, pIoCtx);
6001	}
6002
6003	/** @copydoc VBOXHDDBACKEND::pfnGetVersion */
6004	static DECLCALLBACK(unsigned) vmdkGetVersion(void *pBackendData)
6005	{
6006	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
6007	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6008
6009	AssertPtr(pImage);
6010
6011	if (pImage)
6012	return VMDK_IMAGE_VERSION;
6013	else
6014	return 0;
6015	}
6016
6017	/** @copydoc VBOXHDDBACKEND::pfnGetSectorSize */
6018	static DECLCALLBACK(uint32_t) vmdkGetSectorSize(void *pBackendData)
6019	{
6020	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
6021	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6022
6023	AssertPtr(pImage);
6024
6025	if (pImage)
6026	return 512;
6027	else
6028	return 0;
6029	}
6030
6031	/** @copydoc VBOXHDDBACKEND::pfnGetSize */
6032	static DECLCALLBACK(uint64_t) vmdkGetSize(void *pBackendData)
6033	{
6034	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
6035	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6036
6037	AssertPtr(pImage);
6038
6039	if (pImage)
6040	return pImage->cbSize;
6041	else
6042	return 0;
6043	}
6044
6045	/** @copydoc VBOXHDDBACKEND::pfnGetFileSize */
6046	static DECLCALLBACK(uint64_t) vmdkGetFileSize(void *pBackendData)
6047	{
6048	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
6049	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6050	uint64_t cb = 0;
6051
6052	AssertPtr(pImage);
6053
6054	if (pImage)
6055	{
6056	uint64_t cbFile;
6057	if (pImage->pFile != NULL)
6058	{
6059	int rc = vdIfIoIntFileGetSize(pImage->pIfIo, pImage->pFile->pStorage, &cbFile);
6060	if (RT_SUCCESS(rc))
6061	cb += cbFile;
6062	}
6063	for (unsigned i = 0; i < pImage->cExtents; i++)
6064	{
6065	if (pImage->pExtents[i].pFile != NULL)
6066	{
6067	int rc = vdIfIoIntFileGetSize(pImage->pIfIo, pImage->pExtents[i].pFile->pStorage, &cbFile);
6068	if (RT_SUCCESS(rc))
6069	cb += cbFile;
6070	}
6071	}
6072	}
6073
6074	LogFlowFunc(("returns %lld\n", cb));
6075	return cb;
6076	}
6077
6078	/** @copydoc VBOXHDDBACKEND::pfnGetPCHSGeometry */
6079	static DECLCALLBACK(int) vmdkGetPCHSGeometry(void *pBackendData, PVDGEOMETRY pPCHSGeometry)
6080	{
6081	LogFlowFunc(("pBackendData=%#p pPCHSGeometry=%#p\n", pBackendData, pPCHSGeometry));
6082	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6083	int rc;
6084
6085	AssertPtr(pImage);
6086
6087	if (pImage)
6088	{
6089	if (pImage->PCHSGeometry.cCylinders)
6090	{
6091	*pPCHSGeometry = pImage->PCHSGeometry;
6092	rc = VINF_SUCCESS;
6093	}
6094	else
6095	rc = VERR_VD_GEOMETRY_NOT_SET;
6096	}
6097	else
6098	rc = VERR_VD_NOT_OPENED;
6099
6100	LogFlowFunc(("returns %Rrc (PCHS=%u/%u/%u)\n", rc, pPCHSGeometry->cCylinders, pPCHSGeometry->cHeads, pPCHSGeometry->cSectors));
6101	return rc;
6102	}
6103
6104	/** @copydoc VBOXHDDBACKEND::pfnSetPCHSGeometry */
6105	static DECLCALLBACK(int) vmdkSetPCHSGeometry(void *pBackendData, PCVDGEOMETRY pPCHSGeometry)
6106	{
6107	LogFlowFunc(("pBackendData=%#p pPCHSGeometry=%#p PCHS=%u/%u/%u\n", pBackendData, pPCHSGeometry, pPCHSGeometry->cCylinders, pPCHSGeometry->cHeads, pPCHSGeometry->cSectors));
6108	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6109	int rc;
6110
6111	AssertPtr(pImage);
6112
6113	if (pImage)
6114	{
6115	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
6116	{
6117	rc = VERR_VD_IMAGE_READ_ONLY;
6118	goto out;
6119	}
6120	if (pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6121	{
6122	rc = VERR_NOT_SUPPORTED;
6123	goto out;
6124	}
6125	rc = vmdkDescSetPCHSGeometry(pImage, pPCHSGeometry);
6126	if (RT_FAILURE(rc))
6127	goto out;
6128
6129	pImage->PCHSGeometry = *pPCHSGeometry;
6130	rc = VINF_SUCCESS;
6131	}
6132	else
6133	rc = VERR_VD_NOT_OPENED;
6134
6135	out:
6136	LogFlowFunc(("returns %Rrc\n", rc));
6137	return rc;
6138	}
6139
6140	/** @copydoc VBOXHDDBACKEND::pfnGetLCHSGeometry */
6141	static DECLCALLBACK(int) vmdkGetLCHSGeometry(void *pBackendData, PVDGEOMETRY pLCHSGeometry)
6142	{
6143	LogFlowFunc(("pBackendData=%#p pLCHSGeometry=%#p\n", pBackendData, pLCHSGeometry));
6144	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6145	int rc;
6146
6147	AssertPtr(pImage);
6148
6149	if (pImage)
6150	{
6151	if (pImage->LCHSGeometry.cCylinders)
6152	{
6153	*pLCHSGeometry = pImage->LCHSGeometry;
6154	rc = VINF_SUCCESS;
6155	}
6156	else
6157	rc = VERR_VD_GEOMETRY_NOT_SET;
6158	}
6159	else
6160	rc = VERR_VD_NOT_OPENED;
6161
6162	LogFlowFunc(("returns %Rrc (LCHS=%u/%u/%u)\n", rc, pLCHSGeometry->cCylinders, pLCHSGeometry->cHeads, pLCHSGeometry->cSectors));
6163	return rc;
6164	}
6165
6166	/** @copydoc VBOXHDDBACKEND::pfnSetLCHSGeometry */
6167	static DECLCALLBACK(int) vmdkSetLCHSGeometry(void *pBackendData, PCVDGEOMETRY pLCHSGeometry)
6168	{
6169	LogFlowFunc(("pBackendData=%#p pLCHSGeometry=%#p LCHS=%u/%u/%u\n", pBackendData, pLCHSGeometry, pLCHSGeometry->cCylinders, pLCHSGeometry->cHeads, pLCHSGeometry->cSectors));
6170	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6171	int rc;
6172
6173	AssertPtr(pImage);
6174
6175	if (pImage)
6176	{
6177	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
6178	{
6179	rc = VERR_VD_IMAGE_READ_ONLY;
6180	goto out;
6181	}
6182	if (pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6183	{
6184	rc = VERR_NOT_SUPPORTED;
6185	goto out;
6186	}
6187	rc = vmdkDescSetLCHSGeometry(pImage, pLCHSGeometry);
6188	if (RT_FAILURE(rc))
6189	goto out;
6190
6191	pImage->LCHSGeometry = *pLCHSGeometry;
6192	rc = VINF_SUCCESS;
6193	}
6194	else
6195	rc = VERR_VD_NOT_OPENED;
6196
6197	out:
6198	LogFlowFunc(("returns %Rrc\n", rc));
6199	return rc;
6200	}
6201
6202	/** @copydoc VBOXHDDBACKEND::pfnGetImageFlags */
6203	static DECLCALLBACK(unsigned) vmdkGetImageFlags(void *pBackendData)
6204	{
6205	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
6206	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6207	unsigned uImageFlags;
6208
6209	AssertPtr(pImage);
6210
6211	if (pImage)
6212	uImageFlags = pImage->uImageFlags;
6213	else
6214	uImageFlags = 0;
6215
6216	LogFlowFunc(("returns %#x\n", uImageFlags));
6217	return uImageFlags;
6218	}
6219
6220	/** @copydoc VBOXHDDBACKEND::pfnGetOpenFlags */
6221	static DECLCALLBACK(unsigned) vmdkGetOpenFlags(void *pBackendData)
6222	{
6223	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
6224	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6225	unsigned uOpenFlags;
6226
6227	AssertPtr(pImage);
6228
6229	if (pImage)
6230	uOpenFlags = pImage->uOpenFlags;
6231	else
6232	uOpenFlags = 0;
6233
6234	LogFlowFunc(("returns %#x\n", uOpenFlags));
6235	return uOpenFlags;
6236	}
6237
6238	/** @copydoc VBOXHDDBACKEND::pfnSetOpenFlags */
6239	static DECLCALLBACK(int) vmdkSetOpenFlags(void *pBackendData, unsigned uOpenFlags)
6240	{
6241	LogFlowFunc(("pBackendData=%#p uOpenFlags=%#x\n", pBackendData, uOpenFlags));
6242	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6243	int rc;
6244
6245	/* Image must be opened and the new flags must be valid. */
6246	if (!pImage \|\| (uOpenFlags & ~( VD_OPEN_FLAGS_READONLY \| VD_OPEN_FLAGS_INFO
6247	\| VD_OPEN_FLAGS_ASYNC_IO \| VD_OPEN_FLAGS_SHAREABLE
6248	\| VD_OPEN_FLAGS_SEQUENTIAL \| VD_OPEN_FLAGS_SKIP_CONSISTENCY_CHECKS)))
6249	{
6250	rc = VERR_INVALID_PARAMETER;
6251	goto out;
6252	}
6253
6254	/* StreamOptimized images need special treatment: reopen is prohibited. */
6255	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6256	{
6257	if (pImage->uOpenFlags == uOpenFlags)
6258	rc = VINF_SUCCESS;
6259	else
6260	rc = VERR_INVALID_PARAMETER;
6261	}
6262	else
6263	{
6264	/* Implement this operation via reopening the image. */
6265	vmdkFreeImage(pImage, false);
6266	rc = vmdkOpenImage(pImage, uOpenFlags);
6267	}
6268
6269	out:
6270	LogFlowFunc(("returns %Rrc\n", rc));
6271	return rc;
6272	}
6273
6274	/** @copydoc VBOXHDDBACKEND::pfnGetComment */
6275	static DECLCALLBACK(int) vmdkGetComment(void pBackendData, char pszComment,
6276	size_t cbComment)
6277	{
6278	LogFlowFunc(("pBackendData=%#p pszComment=%#p cbComment=%zu\n", pBackendData, pszComment, cbComment));
6279	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6280	int rc;
6281
6282	AssertPtr(pImage);
6283
6284	if (pImage)
6285	{
6286	const char *pszCommentEncoded = NULL;
6287	rc = vmdkDescDDBGetStr(pImage, &pImage->Descriptor,
6288	"ddb.comment", &pszCommentEncoded);
6289	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
6290	pszCommentEncoded = NULL;
6291	else if (RT_FAILURE(rc))
6292	goto out;
6293
6294	if (pszComment && pszCommentEncoded)
6295	rc = vmdkDecodeString(pszCommentEncoded, pszComment, cbComment);
6296	else
6297	{
6298	if (pszComment)
6299	*pszComment = '\0';
6300	rc = VINF_SUCCESS;
6301	}
6302	if (pszCommentEncoded)
6303	RTStrFree((char )(void )pszCommentEncoded);
6304	}
6305	else
6306	rc = VERR_VD_NOT_OPENED;
6307
6308	out:
6309	LogFlowFunc(("returns %Rrc comment='%s'\n", rc, pszComment));
6310	return rc;
6311	}
6312
6313	/** @copydoc VBOXHDDBACKEND::pfnSetComment */
6314	static DECLCALLBACK(int) vmdkSetComment(void pBackendData, const char pszComment)
6315	{
6316	LogFlowFunc(("pBackendData=%#p pszComment=\"%s\"\n", pBackendData, pszComment));
6317	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6318	int rc;
6319
6320	AssertPtr(pImage);
6321
6322	if (pImage)
6323	{
6324	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
6325	{
6326	rc = VERR_VD_IMAGE_READ_ONLY;
6327	goto out;
6328	}
6329	if (pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6330	{
6331	rc = VERR_NOT_SUPPORTED;
6332	goto out;
6333	}
6334
6335	rc = vmdkSetImageComment(pImage, pszComment);
6336	}
6337	else
6338	rc = VERR_VD_NOT_OPENED;
6339
6340	out:
6341	LogFlowFunc(("returns %Rrc\n", rc));
6342	return rc;
6343	}
6344
6345	/** @copydoc VBOXHDDBACKEND::pfnGetUuid */
6346	static DECLCALLBACK(int) vmdkGetUuid(void *pBackendData, PRTUUID pUuid)
6347	{
6348	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
6349	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6350	int rc;
6351
6352	AssertPtr(pImage);
6353
6354	if (pImage)
6355	{
6356	*pUuid = pImage->ImageUuid;
6357	rc = VINF_SUCCESS;
6358	}
6359	else
6360	rc = VERR_VD_NOT_OPENED;
6361
6362	LogFlowFunc(("returns %Rrc (%RTuuid)\n", rc, pUuid));
6363	return rc;
6364	}
6365
6366	/** @copydoc VBOXHDDBACKEND::pfnSetUuid */
6367	static DECLCALLBACK(int) vmdkSetUuid(void *pBackendData, PCRTUUID pUuid)
6368	{
6369	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
6370	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6371	int rc;
6372
6373	LogFlowFunc(("%RTuuid\n", pUuid));
6374	AssertPtr(pImage);
6375
6376	if (pImage)
6377	{
6378	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
6379	{
6380	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
6381	{
6382	pImage->ImageUuid = *pUuid;
6383	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
6384	VMDK_DDB_IMAGE_UUID, pUuid);
6385	if (RT_FAILURE(rc))
6386	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image UUID in descriptor in '%s'"), pImage->pszFilename);
6387	rc = VINF_SUCCESS;
6388	}
6389	else
6390	rc = VERR_NOT_SUPPORTED;
6391	}
6392	else
6393	rc = VERR_VD_IMAGE_READ_ONLY;
6394	}
6395	else
6396	rc = VERR_VD_NOT_OPENED;
6397
6398	LogFlowFunc(("returns %Rrc\n", rc));
6399	return rc;
6400	}
6401
6402	/** @copydoc VBOXHDDBACKEND::pfnGetModificationUuid */
6403	static DECLCALLBACK(int) vmdkGetModificationUuid(void *pBackendData, PRTUUID pUuid)
6404	{
6405	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
6406	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6407	int rc;
6408
6409	AssertPtr(pImage);
6410
6411	if (pImage)
6412	{
6413	*pUuid = pImage->ModificationUuid;
6414	rc = VINF_SUCCESS;
6415	}
6416	else
6417	rc = VERR_VD_NOT_OPENED;
6418
6419	LogFlowFunc(("returns %Rrc (%RTuuid)\n", rc, pUuid));
6420	return rc;
6421	}
6422
6423	/** @copydoc VBOXHDDBACKEND::pfnSetModificationUuid */
6424	static DECLCALLBACK(int) vmdkSetModificationUuid(void *pBackendData, PCRTUUID pUuid)
6425	{
6426	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
6427	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6428	int rc;
6429
6430	AssertPtr(pImage);
6431
6432	if (pImage)
6433	{
6434	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
6435	{
6436	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
6437	{
6438	/* Only touch the modification uuid if it changed. */
6439	if (RTUuidCompare(&pImage->ModificationUuid, pUuid))
6440	{
6441	pImage->ModificationUuid = *pUuid;
6442	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
6443	VMDK_DDB_MODIFICATION_UUID, pUuid);
6444	if (RT_FAILURE(rc))
6445	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing modification UUID in descriptor in '%s'"), pImage->pszFilename);
6446	}
6447	rc = VINF_SUCCESS;
6448	}
6449	else
6450	rc = VERR_NOT_SUPPORTED;
6451	}
6452	else
6453	rc = VERR_VD_IMAGE_READ_ONLY;
6454	}
6455	else
6456	rc = VERR_VD_NOT_OPENED;
6457
6458	LogFlowFunc(("returns %Rrc\n", rc));
6459	return rc;
6460	}
6461
6462	/** @copydoc VBOXHDDBACKEND::pfnGetParentUuid */
6463	static DECLCALLBACK(int) vmdkGetParentUuid(void *pBackendData, PRTUUID pUuid)
6464	{
6465	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
6466	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6467	int rc;
6468
6469	AssertPtr(pImage);
6470
6471	if (pImage)
6472	{
6473	*pUuid = pImage->ParentUuid;
6474	rc = VINF_SUCCESS;
6475	}
6476	else
6477	rc = VERR_VD_NOT_OPENED;
6478
6479	LogFlowFunc(("returns %Rrc (%RTuuid)\n", rc, pUuid));
6480	return rc;
6481	}
6482
6483	/** @copydoc VBOXHDDBACKEND::pfnSetParentUuid */
6484	static DECLCALLBACK(int) vmdkSetParentUuid(void *pBackendData, PCRTUUID pUuid)
6485	{
6486	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
6487	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6488	int rc;
6489
6490	AssertPtr(pImage);
6491
6492	if (pImage)
6493	{
6494	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
6495	{
6496	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
6497	{
6498	pImage->ParentUuid = *pUuid;
6499	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
6500	VMDK_DDB_PARENT_UUID, pUuid);
6501	if (RT_FAILURE(rc))
6502	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent image UUID in descriptor in '%s'"), pImage->pszFilename);
6503	rc = VINF_SUCCESS;
6504	}
6505	else
6506	rc = VERR_NOT_SUPPORTED;
6507	}
6508	else
6509	rc = VERR_VD_IMAGE_READ_ONLY;
6510	}
6511	else
6512	rc = VERR_VD_NOT_OPENED;
6513
6514	LogFlowFunc(("returns %Rrc\n", rc));
6515	return rc;
6516	}
6517
6518	/** @copydoc VBOXHDDBACKEND::pfnGetParentModificationUuid */
6519	static DECLCALLBACK(int) vmdkGetParentModificationUuid(void *pBackendData, PRTUUID pUuid)
6520	{
6521	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
6522	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6523	int rc;
6524
6525	AssertPtr(pImage);
6526
6527	if (pImage)
6528	{
6529	*pUuid = pImage->ParentModificationUuid;
6530	rc = VINF_SUCCESS;
6531	}
6532	else
6533	rc = VERR_VD_NOT_OPENED;
6534
6535	LogFlowFunc(("returns %Rrc (%RTuuid)\n", rc, pUuid));
6536	return rc;
6537	}
6538
6539	/** @copydoc VBOXHDDBACKEND::pfnSetParentModificationUuid */
6540	static DECLCALLBACK(int) vmdkSetParentModificationUuid(void *pBackendData, PCRTUUID pUuid)
6541	{
6542	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
6543	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6544	int rc;
6545
6546	AssertPtr(pImage);
6547
6548	if (pImage)
6549	{
6550	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
6551	{
6552	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
6553	{
6554	pImage->ParentModificationUuid = *pUuid;
6555	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
6556	VMDK_DDB_PARENT_MODIFICATION_UUID, pUuid);
6557	if (RT_FAILURE(rc))
6558	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent image UUID in descriptor in '%s'"), pImage->pszFilename);
6559	rc = VINF_SUCCESS;
6560	}
6561	else
6562	rc = VERR_NOT_SUPPORTED;
6563	}
6564	else
6565	rc = VERR_VD_IMAGE_READ_ONLY;
6566	}
6567	else
6568	rc = VERR_VD_NOT_OPENED;
6569
6570	LogFlowFunc(("returns %Rrc\n", rc));
6571	return rc;
6572	}
6573
6574	/** @copydoc VBOXHDDBACKEND::pfnDump */
6575	static DECLCALLBACK(void) vmdkDump(void *pBackendData)
6576	{
6577	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6578
6579	AssertPtr(pImage);
6580	if (pImage)
6581	{
6582	vdIfErrorMessage(pImage->pIfError, "Header: Geometry PCHS=%u/%u/%u LCHS=%u/%u/%u cbSector=%llu\n",
6583	pImage->PCHSGeometry.cCylinders, pImage->PCHSGeometry.cHeads, pImage->PCHSGeometry.cSectors,
6584	pImage->LCHSGeometry.cCylinders, pImage->LCHSGeometry.cHeads, pImage->LCHSGeometry.cSectors,
6585	VMDK_BYTE2SECTOR(pImage->cbSize));
6586	vdIfErrorMessage(pImage->pIfError, "Header: uuidCreation={%RTuuid}\n", &pImage->ImageUuid);
6587	vdIfErrorMessage(pImage->pIfError, "Header: uuidModification={%RTuuid}\n", &pImage->ModificationUuid);
6588	vdIfErrorMessage(pImage->pIfError, "Header: uuidParent={%RTuuid}\n", &pImage->ParentUuid);
6589	vdIfErrorMessage(pImage->pIfError, "Header: uuidParentModification={%RTuuid}\n", &pImage->ParentModificationUuid);
6590	}
6591	}
6592
6593
6594
6595	const VBOXHDDBACKEND g_VmdkBackend =
6596	{
6597	/* pszBackendName */
6598	"VMDK",
6599	/* cbSize */
6600	sizeof(VBOXHDDBACKEND),
6601	/* uBackendCaps */
6602	VD_CAP_UUID \| VD_CAP_CREATE_FIXED \| VD_CAP_CREATE_DYNAMIC
6603	\| VD_CAP_CREATE_SPLIT_2G \| VD_CAP_DIFF \| VD_CAP_FILE \| VD_CAP_ASYNC
6604	\| VD_CAP_VFS,
6605	/* paFileExtensions */
6606	s_aVmdkFileExtensions,
6607	/* paConfigInfo */
6608	NULL,
6609	/* pfnCheckIfValid */
6610	vmdkCheckIfValid,
6611	/* pfnOpen */
6612	vmdkOpen,
6613	/* pfnCreate */
6614	vmdkCreate,
6615	/* pfnRename */
6616	vmdkRename,
6617	/* pfnClose */
6618	vmdkClose,
6619	/* pfnRead */
6620	vmdkRead,
6621	/* pfnWrite */
6622	vmdkWrite,
6623	/* pfnFlush */
6624	vmdkFlush,
6625	/* pfnDiscard */
6626	NULL,
6627	/* pfnGetVersion */
6628	vmdkGetVersion,
6629	/* pfnGetSectorSize */
6630	vmdkGetSectorSize,
6631	/* pfnGetSize */
6632	vmdkGetSize,
6633	/* pfnGetFileSize */
6634	vmdkGetFileSize,
6635	/* pfnGetPCHSGeometry */
6636	vmdkGetPCHSGeometry,
6637	/* pfnSetPCHSGeometry */
6638	vmdkSetPCHSGeometry,
6639	/* pfnGetLCHSGeometry */
6640	vmdkGetLCHSGeometry,
6641	/* pfnSetLCHSGeometry */
6642	vmdkSetLCHSGeometry,
6643	/* pfnGetImageFlags */
6644	vmdkGetImageFlags,
6645	/* pfnGetOpenFlags */
6646	vmdkGetOpenFlags,
6647	/* pfnSetOpenFlags */
6648	vmdkSetOpenFlags,
6649	/* pfnGetComment */
6650	vmdkGetComment,
6651	/* pfnSetComment */
6652	vmdkSetComment,
6653	/* pfnGetUuid */
6654	vmdkGetUuid,
6655	/* pfnSetUuid */
6656	vmdkSetUuid,
6657	/* pfnGetModificationUuid */
6658	vmdkGetModificationUuid,
6659	/* pfnSetModificationUuid */
6660	vmdkSetModificationUuid,
6661	/* pfnGetParentUuid */
6662	vmdkGetParentUuid,
6663	/* pfnSetParentUuid */
6664	vmdkSetParentUuid,
6665	/* pfnGetParentModificationUuid */
6666	vmdkGetParentModificationUuid,
6667	/* pfnSetParentModificationUuid */
6668	vmdkSetParentModificationUuid,
6669	/* pfnDump */
6670	vmdkDump,
6671	/* pfnGetTimestamp */
6672	NULL,
6673	/* pfnGetParentTimestamp */
6674	NULL,
6675	/* pfnSetParentTimestamp */
6676	NULL,
6677	/* pfnGetParentFilename */
6678	NULL,
6679	/* pfnSetParentFilename */
6680	NULL,
6681	/* pfnComposeLocation */
6682	genericFileComposeLocation,
6683	/* pfnComposeName */
6684	genericFileComposeName,
6685	/* pfnCompact */
6686	NULL,
6687	/* pfnResize */
6688	NULL,
6689	/* pfnRepair */
6690	NULL,
6691	/* pfnTraverseMetadata */
6692	NULL
6693	};

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/Storage/VMDK.cpp@ 58160

Download in other formats: