VMDK.cpp@ 89825

Last change on this file since 89825 was 87209, checked in by vboxsync, 4 years ago
Main: bugref:9224: Added device and partition paths checks for Darwin
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 339.4 KB

Line
1	/* $Id: VMDK.cpp 87209 2021-01-11 10:09:09Z vboxsync $ */
2	/** @file
3	* VMDK disk image, core code.
4	*/
5	/*
6	* Copyright (C) 2006-2020 Oracle Corporation
7	*
8	* This file is part of VirtualBox Open Source Edition (OSE), as
9	* available from http://www.virtualbox.org. This file is free software;
10	* you can redistribute it and/or modify it under the terms of the GNU
11	* General Public License (GPL) as published by the Free Software
12	* Foundation, in version 2 as it comes in the "COPYING" file of the
13	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
14	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
15	*/
16
17
18	/*********************************************************************************************************************************
19	* Header Files *
20	*********************************************************************************************************************************/
21	#define LOG_GROUP LOG_GROUP_VD_VMDK
22	#include <VBox/log.h> /* before VBox/vd-ifs.h */
23	#include <VBox/vd-plugin.h>
24	#include <VBox/err.h>
25	#include <iprt/assert.h>
26	#include <iprt/alloc.h>
27	#include <iprt/base64.h>
28	#include <iprt/ctype.h>
29	#include <iprt/crc.h>
30	#include <iprt/dvm.h>
31	#include <iprt/uuid.h>
32	#include <iprt/path.h>
33	#include <iprt/rand.h>
34	#include <iprt/string.h>
35	#include <iprt/sort.h>
36	#include <iprt/zip.h>
37	#include <iprt/asm.h>
38	#ifdef RT_OS_WINDOWS
39	# include <iprt/utf16.h>
40	# include <iprt/uni.h>
41	# include <iprt/uni.h>
42	# include <iprt/nt/nt-and-windows.h>
43	# include <winioctl.h>
44	#endif
45	#ifdef RT_OS_LINUX
46	# include <errno.h>
47	# include <sys/stat.h>
48	# include <iprt/dir.h>
49	# include <iprt/symlink.h>
50	# include <iprt/linux/sysfs.h>
51	#endif
52	#ifdef RT_OS_FREEBSD
53	#include <libgeom.h>
54	#include <sys/stat.h>
55	#include <stdlib.h>
56	#endif
57	#ifdef RT_OS_SOLARIS
58	#include <sys/dkio.h>
59	#include <sys/vtoc.h>
60	#include <sys/efi_partition.h>
61	#include <unistd.h>
62	#include <errno.h>
63	#endif
64	#ifdef RT_OS_DARWIN
65	# include <sys/stat.h>
66	# include <sys/disk.h>
67	# include <errno.h>
68	/* The following structure and IOCTLs are defined in znu bsd/sys/disk.h but
69	inside KERNEL ifdefs and thus stripped from the SDK edition of the header.
70	While we could try include the header from the Kernel.framework, it's a lot
71	easier to just add the structure and 4 defines here. */
72	typedef struct
73	{
74	uint64_t offset;
75	uint64_t length;
76	uint8_t reserved0128[12];
77	dev_t dev;
78	} dk_physical_extent_t;
79	# define DKIOCGETBASE _IOR( 'd', 73, uint64_t)
80	# define DKIOCLOCKPHYSICALEXTENTS _IO( 'd', 81)
81	# define DKIOCGETPHYSICALEXTENT _IOWR('d', 82, dk_physical_extent_t)
82	# define DKIOCUNLOCKPHYSICALEXTENTS _IO( 'd', 83)
83	#endif /* RT_OS_DARWIN */
84	#include "VDBackends.h"
85
86
87	/*********************************************************************************************************************************
88	* Constants And Macros, Structures and Typedefs *
89	*********************************************************************************************************************************/
90	/** Maximum encoded string size (including NUL) we allow for VMDK images.
91	* Deliberately not set high to avoid running out of descriptor space. */
92	#define VMDK_ENCODED_COMMENT_MAX 1024
93	/** VMDK descriptor DDB entry for PCHS cylinders. */
94	#define VMDK_DDB_GEO_PCHS_CYLINDERS "ddb.geometry.cylinders"
95	/** VMDK descriptor DDB entry for PCHS heads. */
96	#define VMDK_DDB_GEO_PCHS_HEADS "ddb.geometry.heads"
97	/** VMDK descriptor DDB entry for PCHS sectors. */
98	#define VMDK_DDB_GEO_PCHS_SECTORS "ddb.geometry.sectors"
99	/** VMDK descriptor DDB entry for LCHS cylinders. */
100	#define VMDK_DDB_GEO_LCHS_CYLINDERS "ddb.geometry.biosCylinders"
101	/** VMDK descriptor DDB entry for LCHS heads. */
102	#define VMDK_DDB_GEO_LCHS_HEADS "ddb.geometry.biosHeads"
103	/** VMDK descriptor DDB entry for LCHS sectors. */
104	#define VMDK_DDB_GEO_LCHS_SECTORS "ddb.geometry.biosSectors"
105	/** VMDK descriptor DDB entry for image UUID. */
106	#define VMDK_DDB_IMAGE_UUID "ddb.uuid.image"
107	/** VMDK descriptor DDB entry for image modification UUID. */
108	#define VMDK_DDB_MODIFICATION_UUID "ddb.uuid.modification"
109	/** VMDK descriptor DDB entry for parent image UUID. */
110	#define VMDK_DDB_PARENT_UUID "ddb.uuid.parent"
111	/** VMDK descriptor DDB entry for parent image modification UUID. */
112	#define VMDK_DDB_PARENT_MODIFICATION_UUID "ddb.uuid.parentmodification"
113	/** No compression for streamOptimized files. */
114	#define VMDK_COMPRESSION_NONE 0
115	/** Deflate compression for streamOptimized files. */
116	#define VMDK_COMPRESSION_DEFLATE 1
117	/** Marker that the actual GD value is stored in the footer. */
118	#define VMDK_GD_AT_END 0xffffffffffffffffULL
119	/** Marker for end-of-stream in streamOptimized images. */
120	#define VMDK_MARKER_EOS 0
121	/** Marker for grain table block in streamOptimized images. */
122	#define VMDK_MARKER_GT 1
123	/** Marker for grain directory block in streamOptimized images. */
124	#define VMDK_MARKER_GD 2
125	/** Marker for footer in streamOptimized images. */
126	#define VMDK_MARKER_FOOTER 3
127	/** Marker for unknown purpose in streamOptimized images.
128	* Shows up in very recent images created by vSphere, but only sporadically.
129	* They "forgot" to document that one in the VMDK specification. */
130	#define VMDK_MARKER_UNSPECIFIED 4
131	/** Dummy marker for "don't check the marker value". */
132	#define VMDK_MARKER_IGNORE 0xffffffffU
133	/**
134	* Magic number for hosted images created by VMware Workstation 4, VMware
135	* Workstation 5, VMware Server or VMware Player. Not necessarily sparse.
136	*/
137	#define VMDK_SPARSE_MAGICNUMBER 0x564d444b /* 'V' 'M' 'D' 'K' */
138	/**
139	* VMDK hosted binary extent header. The "Sparse" is a total misnomer, as
140	* this header is also used for monolithic flat images.
141	*/
142	#pragma pack(1)
143	typedef struct SparseExtentHeader
144	{
145	uint32_t magicNumber;
146	uint32_t version;
147	uint32_t flags;
148	uint64_t capacity;
149	uint64_t grainSize;
150	uint64_t descriptorOffset;
151	uint64_t descriptorSize;
152	uint32_t numGTEsPerGT;
153	uint64_t rgdOffset;
154	uint64_t gdOffset;
155	uint64_t overHead;
156	bool uncleanShutdown;
157	char singleEndLineChar;
158	char nonEndLineChar;
159	char doubleEndLineChar1;
160	char doubleEndLineChar2;
161	uint16_t compressAlgorithm;
162	uint8_t pad[433];
163	} SparseExtentHeader;
164	#pragma pack()
165	/** The maximum allowed descriptor size in the extent header in sectors. */
166	#define VMDK_SPARSE_DESCRIPTOR_SIZE_MAX UINT64_C(20480) /* 10MB */
167	/** VMDK capacity for a single chunk when 2G splitting is turned on. Should be
168	* divisible by the default grain size (64K) */
169	#define VMDK_2G_SPLIT_SIZE (2047 * 1024 * 1024)
170	/** VMDK streamOptimized file format marker. The type field may or may not
171	* be actually valid, but there's always data to read there. */
172	#pragma pack(1)
173	typedef struct VMDKMARKER
174	{
175	uint64_t uSector;
176	uint32_t cbSize;
177	uint32_t uType;
178	} VMDKMARKER, *PVMDKMARKER;
179	#pragma pack()
180	/** Convert sector number/size to byte offset/size. */
181	#define VMDK_SECTOR2BYTE(u) ((uint64_t)(u) << 9)
182	/** Convert byte offset/size to sector number/size. */
183	#define VMDK_BYTE2SECTOR(u) ((u) >> 9)
184	/**
185	* VMDK extent type.
186	*/
187	typedef enum VMDKETYPE
188	{
189	/** Hosted sparse extent. */
190	VMDKETYPE_HOSTED_SPARSE = 1,
191	/** Flat extent. */
192	VMDKETYPE_FLAT,
193	/** Zero extent. */
194	VMDKETYPE_ZERO,
195	/** VMFS extent, used by ESX. */
196	VMDKETYPE_VMFS
197	} VMDKETYPE, *PVMDKETYPE;
198	/**
199	* VMDK access type for a extent.
200	*/
201	typedef enum VMDKACCESS
202	{
203	/** No access allowed. */
204	VMDKACCESS_NOACCESS = 0,
205	/** Read-only access. */
206	VMDKACCESS_READONLY,
207	/** Read-write access. */
208	VMDKACCESS_READWRITE
209	} VMDKACCESS, *PVMDKACCESS;
210	/** Forward declaration for PVMDKIMAGE. */
211	typedef struct VMDKIMAGE *PVMDKIMAGE;
212	/**
213	* Extents files entry. Used for opening a particular file only once.
214	*/
215	typedef struct VMDKFILE
216	{
217	/** Pointer to file path. Local copy. */
218	const char *pszFilename;
219	/** Pointer to base name. Local copy. */
220	const char *pszBasename;
221	/** File open flags for consistency checking. */
222	unsigned fOpen;
223	/** Handle for sync/async file abstraction.*/
224	PVDIOSTORAGE pStorage;
225	/** Reference counter. */
226	unsigned uReferences;
227	/** Flag whether the file should be deleted on last close. */
228	bool fDelete;
229	/** Pointer to the image we belong to (for debugging purposes). */
230	PVMDKIMAGE pImage;
231	/** Pointer to next file descriptor. */
232	struct VMDKFILE *pNext;
233	/** Pointer to the previous file descriptor. */
234	struct VMDKFILE *pPrev;
235	} VMDKFILE, *PVMDKFILE;
236	/**
237	* VMDK extent data structure.
238	*/
239	typedef struct VMDKEXTENT
240	{
241	/** File handle. */
242	PVMDKFILE pFile;
243	/** Base name of the image extent. */
244	const char *pszBasename;
245	/** Full name of the image extent. */
246	const char *pszFullname;
247	/** Number of sectors in this extent. */
248	uint64_t cSectors;
249	/** Number of sectors per block (grain in VMDK speak). */
250	uint64_t cSectorsPerGrain;
251	/** Starting sector number of descriptor. */
252	uint64_t uDescriptorSector;
253	/** Size of descriptor in sectors. */
254	uint64_t cDescriptorSectors;
255	/** Starting sector number of grain directory. */
256	uint64_t uSectorGD;
257	/** Starting sector number of redundant grain directory. */
258	uint64_t uSectorRGD;
259	/** Total number of metadata sectors. */
260	uint64_t cOverheadSectors;
261	/** Nominal size (i.e. as described by the descriptor) of this extent. */
262	uint64_t cNominalSectors;
263	/** Sector offset (i.e. as described by the descriptor) of this extent. */
264	uint64_t uSectorOffset;
265	/** Number of entries in a grain table. */
266	uint32_t cGTEntries;
267	/** Number of sectors reachable via a grain directory entry. */
268	uint32_t cSectorsPerGDE;
269	/** Number of entries in the grain directory. */
270	uint32_t cGDEntries;
271	/** Pointer to the next free sector. Legacy information. Do not use. */
272	uint32_t uFreeSector;
273	/** Number of this extent in the list of images. */
274	uint32_t uExtent;
275	/** Pointer to the descriptor (NULL if no descriptor in this extent). */
276	char *pDescData;
277	/** Pointer to the grain directory. */
278	uint32_t *pGD;
279	/** Pointer to the redundant grain directory. */
280	uint32_t *pRGD;
281	/** VMDK version of this extent. 1=1.0/1.1 */
282	uint32_t uVersion;
283	/** Type of this extent. */
284	VMDKETYPE enmType;
285	/** Access to this extent. */
286	VMDKACCESS enmAccess;
287	/** Flag whether this extent is marked as unclean. */
288	bool fUncleanShutdown;
289	/** Flag whether the metadata in the extent header needs to be updated. */
290	bool fMetaDirty;
291	/** Flag whether there is a footer in this extent. */
292	bool fFooter;
293	/** Compression type for this extent. */
294	uint16_t uCompression;
295	/** Append position for writing new grain. Only for sparse extents. */
296	uint64_t uAppendPosition;
297	/** Last grain which was accessed. Only for streamOptimized extents. */
298	uint32_t uLastGrainAccess;
299	/** Starting sector corresponding to the grain buffer. */
300	uint32_t uGrainSectorAbs;
301	/** Grain number corresponding to the grain buffer. */
302	uint32_t uGrain;
303	/** Actual size of the compressed data, only valid for reading. */
304	uint32_t cbGrainStreamRead;
305	/** Size of compressed grain buffer for streamOptimized extents. */
306	size_t cbCompGrain;
307	/** Compressed grain buffer for streamOptimized extents, with marker. */
308	void *pvCompGrain;
309	/** Decompressed grain buffer for streamOptimized extents. */
310	void *pvGrain;
311	/** Reference to the image in which this extent is used. Do not use this
312	* on a regular basis to avoid passing pImage references to functions
313	* explicitly. */
314	struct VMDKIMAGE *pImage;
315	} VMDKEXTENT, *PVMDKEXTENT;
316	/**
317	* Grain table cache size. Allocated per image.
318	*/
319	#define VMDK_GT_CACHE_SIZE 256
320	/**
321	* Grain table block size. Smaller than an actual grain table block to allow
322	* more grain table blocks to be cached without having to allocate excessive
323	* amounts of memory for the cache.
324	*/
325	#define VMDK_GT_CACHELINE_SIZE 128
326	/**
327	* Maximum number of lines in a descriptor file. Not worth the effort of
328	* making it variable. Descriptor files are generally very short (~20 lines),
329	* with the exception of sparse files split in 2G chunks, which need for the
330	* maximum size (almost 2T) exactly 1025 lines for the disk database.
331	*/
332	#define VMDK_DESCRIPTOR_LINES_MAX 1100U
333	/**
334	* Parsed descriptor information. Allows easy access and update of the
335	* descriptor (whether separate file or not). Free form text files suck.
336	*/
337	typedef struct VMDKDESCRIPTOR
338	{
339	/** Line number of first entry of the disk descriptor. */
340	unsigned uFirstDesc;
341	/** Line number of first entry in the extent description. */
342	unsigned uFirstExtent;
343	/** Line number of first disk database entry. */
344	unsigned uFirstDDB;
345	/** Total number of lines. */
346	unsigned cLines;
347	/** Total amount of memory available for the descriptor. */
348	size_t cbDescAlloc;
349	/** Set if descriptor has been changed and not yet written to disk. */
350	bool fDirty;
351	/** Array of pointers to the data in the descriptor. */
352	char *aLines[VMDK_DESCRIPTOR_LINES_MAX];
353	/** Array of line indices pointing to the next non-comment line. */
354	unsigned aNextLines[VMDK_DESCRIPTOR_LINES_MAX];
355	} VMDKDESCRIPTOR, *PVMDKDESCRIPTOR;
356	/**
357	* Cache entry for translating extent/sector to a sector number in that
358	* extent.
359	*/
360	typedef struct VMDKGTCACHEENTRY
361	{
362	/** Extent number for which this entry is valid. */
363	uint32_t uExtent;
364	/** GT data block number. */
365	uint64_t uGTBlock;
366	/** Data part of the cache entry. */
367	uint32_t aGTData[VMDK_GT_CACHELINE_SIZE];
368	} VMDKGTCACHEENTRY, *PVMDKGTCACHEENTRY;
369	/**
370	* Cache data structure for blocks of grain table entries. For now this is a
371	* fixed size direct mapping cache, but this should be adapted to the size of
372	* the sparse image and maybe converted to a set-associative cache. The
373	* implementation below implements a write-through cache with write allocate.
374	*/
375	typedef struct VMDKGTCACHE
376	{
377	/** Cache entries. */
378	VMDKGTCACHEENTRY aGTCache[VMDK_GT_CACHE_SIZE];
379	/** Number of cache entries (currently unused). */
380	unsigned cEntries;
381	} VMDKGTCACHE, *PVMDKGTCACHE;
382	/**
383	* Complete VMDK image data structure. Mainly a collection of extents and a few
384	* extra global data fields.
385	*/
386	typedef struct VMDKIMAGE
387	{
388	/** Image name. */
389	const char *pszFilename;
390	/** Descriptor file if applicable. */
391	PVMDKFILE pFile;
392	/** Pointer to the per-disk VD interface list. */
393	PVDINTERFACE pVDIfsDisk;
394	/** Pointer to the per-image VD interface list. */
395	PVDINTERFACE pVDIfsImage;
396	/** Error interface. */
397	PVDINTERFACEERROR pIfError;
398	/** I/O interface. */
399	PVDINTERFACEIOINT pIfIo;
400	/** Pointer to the image extents. */
401	PVMDKEXTENT pExtents;
402	/** Number of image extents. */
403	unsigned cExtents;
404	/** Pointer to the files list, for opening a file referenced multiple
405	* times only once (happens mainly with raw partition access). */
406	PVMDKFILE pFiles;
407	/**
408	* Pointer to an array of segment entries for async I/O.
409	* This is an optimization because the task number to submit is not known
410	* and allocating/freeing an array in the read/write functions every time
411	* is too expensive.
412	*/
413	PPDMDATASEG paSegments;
414	/** Entries available in the segments array. */
415	unsigned cSegments;
416	/** Open flags passed by VBoxHD layer. */
417	unsigned uOpenFlags;
418	/** Image flags defined during creation or determined during open. */
419	unsigned uImageFlags;
420	/** Total size of the image. */
421	uint64_t cbSize;
422	/** Physical geometry of this image. */
423	VDGEOMETRY PCHSGeometry;
424	/** Logical geometry of this image. */
425	VDGEOMETRY LCHSGeometry;
426	/** Image UUID. */
427	RTUUID ImageUuid;
428	/** Image modification UUID. */
429	RTUUID ModificationUuid;
430	/** Parent image UUID. */
431	RTUUID ParentUuid;
432	/** Parent image modification UUID. */
433	RTUUID ParentModificationUuid;
434	/** Pointer to grain table cache, if this image contains sparse extents. */
435	PVMDKGTCACHE pGTCache;
436	/** Pointer to the descriptor (NULL if no separate descriptor file). */
437	char *pDescData;
438	/** Allocation size of the descriptor file. */
439	size_t cbDescAlloc;
440	/** Parsed descriptor file content. */
441	VMDKDESCRIPTOR Descriptor;
442	/** The static region list. */
443	VDREGIONLIST RegionList;
444	} VMDKIMAGE;
445	/** State for the input/output callout of the inflate reader/deflate writer. */
446	typedef struct VMDKCOMPRESSIO
447	{
448	/* Image this operation relates to. */
449	PVMDKIMAGE pImage;
450	/* Current read position. */
451	ssize_t iOffset;
452	/* Size of the compressed grain buffer (available data). */
453	size_t cbCompGrain;
454	/* Pointer to the compressed grain buffer. */
455	void *pvCompGrain;
456	} VMDKCOMPRESSIO;
457	/** Tracks async grain allocation. */
458	typedef struct VMDKGRAINALLOCASYNC
459	{
460	/** Flag whether the allocation failed. */
461	bool fIoErr;
462	/** Current number of transfers pending.
463	* If reached 0 and there is an error the old state is restored. */
464	unsigned cIoXfersPending;
465	/** Sector number */
466	uint64_t uSector;
467	/** Flag whether the grain table needs to be updated. */
468	bool fGTUpdateNeeded;
469	/** Extent the allocation happens. */
470	PVMDKEXTENT pExtent;
471	/** Position of the new grain, required for the grain table update. */
472	uint64_t uGrainOffset;
473	/** Grain table sector. */
474	uint64_t uGTSector;
475	/** Backup grain table sector. */
476	uint64_t uRGTSector;
477	} VMDKGRAINALLOCASYNC, *PVMDKGRAINALLOCASYNC;
478	/**
479	* State information for vmdkRename() and helpers.
480	*/
481	typedef struct VMDKRENAMESTATE
482	{
483	/** Array of old filenames. */
484	char **apszOldName;
485	/** Array of new filenames. */
486	char **apszNewName;
487	/** Array of new lines in the extent descriptor. */
488	char **apszNewLines;
489	/** Name of the old descriptor file if not a sparse image. */
490	char *pszOldDescName;
491	/** Flag whether we called vmdkFreeImage(). */
492	bool fImageFreed;
493	/** Flag whther the descriptor is embedded in the image (sparse) or
494	* in a separate file. */
495	bool fEmbeddedDesc;
496	/** Number of extents in the image. */
497	unsigned cExtents;
498	/** New base filename. */
499	char *pszNewBaseName;
500	/** The old base filename. */
501	char *pszOldBaseName;
502	/** New full filename. */
503	char *pszNewFullName;
504	/** Old full filename. */
505	char *pszOldFullName;
506	/** The old image name. */
507	const char *pszOldImageName;
508	/** Copy of the original VMDK descriptor. */
509	VMDKDESCRIPTOR DescriptorCopy;
510	/** Copy of the extent state for sparse images. */
511	VMDKEXTENT ExtentCopy;
512	} VMDKRENAMESTATE;
513	/** Pointer to a VMDK rename state. */
514	typedef VMDKRENAMESTATE *PVMDKRENAMESTATE;
515
516
517	/*********************************************************************************************************************************
518	* Static Variables *
519	*********************************************************************************************************************************/
520	/** NULL-terminated array of supported file extensions. */
521	static const VDFILEEXTENSION s_aVmdkFileExtensions[] =
522	{
523	{"vmdk", VDTYPE_HDD},
524	{NULL, VDTYPE_INVALID}
525	};
526	/** NULL-terminated array of configuration option. */
527	static const VDCONFIGINFO s_aVmdkConfigInfo[] =
528	{
529	/* Options for VMDK raw disks */
530	{ "RawDrive", NULL, VDCFGVALUETYPE_STRING, 0 },
531	{ "Partitions", NULL, VDCFGVALUETYPE_STRING, 0 },
532	{ "BootSector", NULL, VDCFGVALUETYPE_BYTES, 0 },
533	{ "Relative", NULL, VDCFGVALUETYPE_INTEGER, 0 },
534	/* End of options list */
535	{ NULL, NULL, VDCFGVALUETYPE_INTEGER, 0 }
536	};
537
538
539	/*********************************************************************************************************************************
540	* Internal Functions *
541	*********************************************************************************************************************************/
542	static void vmdkFreeStreamBuffers(PVMDKEXTENT pExtent);
543	static int vmdkFreeExtentData(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
544	bool fDelete);
545	static int vmdkCreateExtents(PVMDKIMAGE pImage, unsigned cExtents);
546	static int vmdkFlushImage(PVMDKIMAGE pImage, PVDIOCTX pIoCtx);
547	static int vmdkSetImageComment(PVMDKIMAGE pImage, const char *pszComment);
548	static int vmdkFreeImage(PVMDKIMAGE pImage, bool fDelete, bool fFlush);
549	static DECLCALLBACK(int) vmdkAllocGrainComplete(void *pBackendData, PVDIOCTX pIoCtx,
550	void *pvUser, int rcReq);
551	/**
552	* Internal: open a file (using a file descriptor cache to ensure each file
553	* is only opened once - anything else can cause locking problems).
554	*/
555	static int vmdkFileOpen(PVMDKIMAGE pImage, PVMDKFILE *ppVmdkFile,
556	const char pszBasename, const char pszFilename, uint32_t fOpen)
557	{
558	int rc = VINF_SUCCESS;
559	PVMDKFILE pVmdkFile;
560	for (pVmdkFile = pImage->pFiles;
561	pVmdkFile != NULL;
562	pVmdkFile = pVmdkFile->pNext)
563	{
564	if (!strcmp(pszFilename, pVmdkFile->pszFilename))
565	{
566	Assert(fOpen == pVmdkFile->fOpen);
567	pVmdkFile->uReferences++;
568	*ppVmdkFile = pVmdkFile;
569	return rc;
570	}
571	}
572	/* If we get here, there's no matching entry in the cache. */
573	pVmdkFile = (PVMDKFILE)RTMemAllocZ(sizeof(VMDKFILE));
574	if (!pVmdkFile)
575	{
576	*ppVmdkFile = NULL;
577	return VERR_NO_MEMORY;
578	}
579	pVmdkFile->pszFilename = RTStrDup(pszFilename);
580	if (!pVmdkFile->pszFilename)
581	{
582	RTMemFree(pVmdkFile);
583	*ppVmdkFile = NULL;
584	return VERR_NO_MEMORY;
585	}
586	if (pszBasename)
587	{
588	pVmdkFile->pszBasename = RTStrDup(pszBasename);
589	if (!pVmdkFile->pszBasename)
590	{
591	RTStrFree((char )(void )pVmdkFile->pszFilename);
592	RTMemFree(pVmdkFile);
593	*ppVmdkFile = NULL;
594	return VERR_NO_MEMORY;
595	}
596	}
597	pVmdkFile->fOpen = fOpen;
598	rc = vdIfIoIntFileOpen(pImage->pIfIo, pszFilename, fOpen,
599	&pVmdkFile->pStorage);
600	if (RT_SUCCESS(rc))
601	{
602	pVmdkFile->uReferences = 1;
603	pVmdkFile->pImage = pImage;
604	pVmdkFile->pNext = pImage->pFiles;
605	if (pImage->pFiles)
606	pImage->pFiles->pPrev = pVmdkFile;
607	pImage->pFiles = pVmdkFile;
608	*ppVmdkFile = pVmdkFile;
609	}
610	else
611	{
612	RTStrFree((char )(void )pVmdkFile->pszFilename);
613	RTMemFree(pVmdkFile);
614	*ppVmdkFile = NULL;
615	}
616	return rc;
617	}
618	/**
619	* Internal: close a file, updating the file descriptor cache.
620	*/
621	static int vmdkFileClose(PVMDKIMAGE pImage, PVMDKFILE *ppVmdkFile, bool fDelete)
622	{
623	int rc = VINF_SUCCESS;
624	PVMDKFILE pVmdkFile = *ppVmdkFile;
625	AssertPtr(pVmdkFile);
626	pVmdkFile->fDelete \|= fDelete;
627	Assert(pVmdkFile->uReferences);
628	pVmdkFile->uReferences--;
629	if (pVmdkFile->uReferences == 0)
630	{
631	PVMDKFILE pPrev;
632	PVMDKFILE pNext;
633	/* Unchain the element from the list. */
634	pPrev = pVmdkFile->pPrev;
635	pNext = pVmdkFile->pNext;
636	if (pNext)
637	pNext->pPrev = pPrev;
638	if (pPrev)
639	pPrev->pNext = pNext;
640	else
641	pImage->pFiles = pNext;
642	rc = vdIfIoIntFileClose(pImage->pIfIo, pVmdkFile->pStorage);
643	bool fFileDel = pVmdkFile->fDelete;
644	if ( pVmdkFile->pszBasename
645	&& fFileDel)
646	{
647	const char *pszSuffix = RTPathSuffix(pVmdkFile->pszBasename);
648	if ( RTPathHasPath(pVmdkFile->pszBasename)
649	\|\| !pszSuffix
650	\|\| ( strcmp(pszSuffix, ".vmdk")
651	&& strcmp(pszSuffix, ".bin")
652	&& strcmp(pszSuffix, ".img")))
653	fFileDel = false;
654	}
655	if (fFileDel)
656	{
657	int rc2 = vdIfIoIntFileDelete(pImage->pIfIo, pVmdkFile->pszFilename);
658	if (RT_SUCCESS(rc))
659	rc = rc2;
660	}
661	else if (pVmdkFile->fDelete)
662	LogRel(("VMDK: Denying deletion of %s\n", pVmdkFile->pszBasename));
663	RTStrFree((char )(void )pVmdkFile->pszFilename);
664	if (pVmdkFile->pszBasename)
665	RTStrFree((char )(void )pVmdkFile->pszBasename);
666	RTMemFree(pVmdkFile);
667	}
668	*ppVmdkFile = NULL;
669	return rc;
670	}
671	/#define VMDK_USE_BLOCK_DECOMP_API - test and enable /
672	#ifndef VMDK_USE_BLOCK_DECOMP_API
673	static DECLCALLBACK(int) vmdkFileInflateHelper(void pvUser, void pvBuf, size_t cbBuf, size_t *pcbBuf)
674	{
675	VMDKCOMPRESSIO pInflateState = (VMDKCOMPRESSIO )pvUser;
676	size_t cbInjected = 0;
677	Assert(cbBuf);
678	if (pInflateState->iOffset < 0)
679	{
680	(uint8_t )pvBuf = RTZIPTYPE_ZLIB;
681	pvBuf = (uint8_t *)pvBuf + 1;
682	cbBuf--;
683	cbInjected = 1;
684	pInflateState->iOffset = RT_UOFFSETOF(VMDKMARKER, uType);
685	}
686	if (!cbBuf)
687	{
688	if (pcbBuf)
689	*pcbBuf = cbInjected;
690	return VINF_SUCCESS;
691	}
692	cbBuf = RT_MIN(cbBuf, pInflateState->cbCompGrain - pInflateState->iOffset);
693	memcpy(pvBuf,
694	(uint8_t *)pInflateState->pvCompGrain + pInflateState->iOffset,
695	cbBuf);
696	pInflateState->iOffset += cbBuf;
697	Assert(pcbBuf);
698	*pcbBuf = cbBuf + cbInjected;
699	return VINF_SUCCESS;
700	}
701	#endif
702	/**
703	* Internal: read from a file and inflate the compressed data,
704	* distinguishing between async and normal operation
705	*/
706	DECLINLINE(int) vmdkFileInflateSync(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
707	uint64_t uOffset, void *pvBuf,
708	size_t cbToRead, const void *pcvMarker,
709	uint64_t puLBA, uint32_t pcbMarkerData)
710	{
711	int rc;
712	#ifndef VMDK_USE_BLOCK_DECOMP_API
713	PRTZIPDECOMP pZip = NULL;
714	#endif
715	VMDKMARKER pMarker = (VMDKMARKER )pExtent->pvCompGrain;
716	size_t cbCompSize, cbActuallyRead;
717	if (!pcvMarker)
718	{
719	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
720	uOffset, pMarker, RT_UOFFSETOF(VMDKMARKER, uType));
721	if (RT_FAILURE(rc))
722	return rc;
723	}
724	else
725	{
726	memcpy(pMarker, pcvMarker, RT_UOFFSETOF(VMDKMARKER, uType));
727	/* pcvMarker endianness has already been partially transformed, fix it */
728	pMarker->uSector = RT_H2LE_U64(pMarker->uSector);
729	pMarker->cbSize = RT_H2LE_U32(pMarker->cbSize);
730	}
731	cbCompSize = RT_LE2H_U32(pMarker->cbSize);
732	if (cbCompSize == 0)
733	{
734	AssertMsgFailed(("VMDK: corrupted marker\n"));
735	return VERR_VD_VMDK_INVALID_FORMAT;
736	}
737	/* Sanity check - the expansion ratio should be much less than 2. */
738	Assert(cbCompSize < 2 * cbToRead);
739	if (cbCompSize >= 2 * cbToRead)
740	return VERR_VD_VMDK_INVALID_FORMAT;
741	/* Compressed grain marker. Data follows immediately. */
742	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
743	uOffset + RT_UOFFSETOF(VMDKMARKER, uType),
744	(uint8_t *)pExtent->pvCompGrain
745	+ RT_UOFFSETOF(VMDKMARKER, uType),
746	RT_ALIGN_Z( cbCompSize
747	+ RT_UOFFSETOF(VMDKMARKER, uType),
748	512)
749	- RT_UOFFSETOF(VMDKMARKER, uType));
750	if (puLBA)
751	*puLBA = RT_LE2H_U64(pMarker->uSector);
752	if (pcbMarkerData)
753	*pcbMarkerData = RT_ALIGN( cbCompSize
754	+ RT_UOFFSETOF(VMDKMARKER, uType),
755	512);
756	#ifdef VMDK_USE_BLOCK_DECOMP_API
757	rc = RTZipBlockDecompress(RTZIPTYPE_ZLIB, 0 /fFlags/,
758	pExtent->pvCompGrain, cbCompSize + RT_UOFFSETOF(VMDKMARKER, uType), NULL,
759	pvBuf, cbToRead, &cbActuallyRead);
760	#else
761	VMDKCOMPRESSIO InflateState;
762	InflateState.pImage = pImage;
763	InflateState.iOffset = -1;
764	InflateState.cbCompGrain = cbCompSize + RT_UOFFSETOF(VMDKMARKER, uType);
765	InflateState.pvCompGrain = pExtent->pvCompGrain;
766	rc = RTZipDecompCreate(&pZip, &InflateState, vmdkFileInflateHelper);
767	if (RT_FAILURE(rc))
768	return rc;
769	rc = RTZipDecompress(pZip, pvBuf, cbToRead, &cbActuallyRead);
770	RTZipDecompDestroy(pZip);
771	#endif /* !VMDK_USE_BLOCK_DECOMP_API */
772	if (RT_FAILURE(rc))
773	{
774	if (rc == VERR_ZIP_CORRUPTED)
775	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: Compressed image is corrupted '%s'"), pExtent->pszFullname);
776	return rc;
777	}
778	if (cbActuallyRead != cbToRead)
779	rc = VERR_VD_VMDK_INVALID_FORMAT;
780	return rc;
781	}
782	static DECLCALLBACK(int) vmdkFileDeflateHelper(void pvUser, const void pvBuf, size_t cbBuf)
783	{
784	VMDKCOMPRESSIO pDeflateState = (VMDKCOMPRESSIO )pvUser;
785	Assert(cbBuf);
786	if (pDeflateState->iOffset < 0)
787	{
788	pvBuf = (const uint8_t *)pvBuf + 1;
789	cbBuf--;
790	pDeflateState->iOffset = RT_UOFFSETOF(VMDKMARKER, uType);
791	}
792	if (!cbBuf)
793	return VINF_SUCCESS;
794	if (pDeflateState->iOffset + cbBuf > pDeflateState->cbCompGrain)
795	return VERR_BUFFER_OVERFLOW;
796	memcpy((uint8_t *)pDeflateState->pvCompGrain + pDeflateState->iOffset,
797	pvBuf, cbBuf);
798	pDeflateState->iOffset += cbBuf;
799	return VINF_SUCCESS;
800	}
801	/**
802	* Internal: deflate the uncompressed data and write to a file,
803	* distinguishing between async and normal operation
804	*/
805	DECLINLINE(int) vmdkFileDeflateSync(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
806	uint64_t uOffset, const void *pvBuf,
807	size_t cbToWrite, uint64_t uLBA,
808	uint32_t *pcbMarkerData)
809	{
810	int rc;
811	PRTZIPCOMP pZip = NULL;
812	VMDKCOMPRESSIO DeflateState;
813	DeflateState.pImage = pImage;
814	DeflateState.iOffset = -1;
815	DeflateState.cbCompGrain = pExtent->cbCompGrain;
816	DeflateState.pvCompGrain = pExtent->pvCompGrain;
817	rc = RTZipCompCreate(&pZip, &DeflateState, vmdkFileDeflateHelper,
818	RTZIPTYPE_ZLIB, RTZIPLEVEL_DEFAULT);
819	if (RT_FAILURE(rc))
820	return rc;
821	rc = RTZipCompress(pZip, pvBuf, cbToWrite);
822	if (RT_SUCCESS(rc))
823	rc = RTZipCompFinish(pZip);
824	RTZipCompDestroy(pZip);
825	if (RT_SUCCESS(rc))
826	{
827	Assert( DeflateState.iOffset > 0
828	&& (size_t)DeflateState.iOffset <= DeflateState.cbCompGrain);
829	/* pad with zeroes to get to a full sector size */
830	uint32_t uSize = DeflateState.iOffset;
831	if (uSize % 512)
832	{
833	uint32_t uSizeAlign = RT_ALIGN(uSize, 512);
834	memset((uint8_t *)pExtent->pvCompGrain + uSize, '\0',
835	uSizeAlign - uSize);
836	uSize = uSizeAlign;
837	}
838	if (pcbMarkerData)
839	*pcbMarkerData = uSize;
840	/* Compressed grain marker. Data follows immediately. */
841	VMDKMARKER pMarker = (VMDKMARKER )pExtent->pvCompGrain;
842	pMarker->uSector = RT_H2LE_U64(uLBA);
843	pMarker->cbSize = RT_H2LE_U32( DeflateState.iOffset
844	- RT_UOFFSETOF(VMDKMARKER, uType));
845	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
846	uOffset, pMarker, uSize);
847	if (RT_FAILURE(rc))
848	return rc;
849	}
850	return rc;
851	}
852	/**
853	* Internal: check if all files are closed, prevent leaking resources.
854	*/
855	static int vmdkFileCheckAllClose(PVMDKIMAGE pImage)
856	{
857	int rc = VINF_SUCCESS, rc2;
858	PVMDKFILE pVmdkFile;
859	Assert(pImage->pFiles == NULL);
860	for (pVmdkFile = pImage->pFiles;
861	pVmdkFile != NULL;
862	pVmdkFile = pVmdkFile->pNext)
863	{
864	LogRel(("VMDK: leaking reference to file \"%s\"\n",
865	pVmdkFile->pszFilename));
866	pImage->pFiles = pVmdkFile->pNext;
867	rc2 = vmdkFileClose(pImage, &pVmdkFile, pVmdkFile->fDelete);
868	if (RT_SUCCESS(rc))
869	rc = rc2;
870	}
871	return rc;
872	}
873	/**
874	* Internal: truncate a string (at a UTF8 code point boundary) and encode the
875	* critical non-ASCII characters.
876	*/
877	static char vmdkEncodeString(const char psz)
878	{
879	char szEnc[VMDK_ENCODED_COMMENT_MAX + 3];
880	char *pszDst = szEnc;
881	AssertPtr(psz);
882	for (; *psz; psz = RTStrNextCp(psz))
883	{
884	char *pszDstPrev = pszDst;
885	RTUNICP Cp = RTStrGetCp(psz);
886	if (Cp == '\\')
887	{
888	pszDst = RTStrPutCp(pszDst, Cp);
889	pszDst = RTStrPutCp(pszDst, Cp);
890	}
891	else if (Cp == '\n')
892	{
893	pszDst = RTStrPutCp(pszDst, '\\');
894	pszDst = RTStrPutCp(pszDst, 'n');
895	}
896	else if (Cp == '\r')
897	{
898	pszDst = RTStrPutCp(pszDst, '\\');
899	pszDst = RTStrPutCp(pszDst, 'r');
900	}
901	else
902	pszDst = RTStrPutCp(pszDst, Cp);
903	if (pszDst - szEnc >= VMDK_ENCODED_COMMENT_MAX - 1)
904	{
905	pszDst = pszDstPrev;
906	break;
907	}
908	}
909	*pszDst = '\0';
910	return RTStrDup(szEnc);
911	}
912	/**
913	* Internal: decode a string and store it into the specified string.
914	*/
915	static int vmdkDecodeString(const char pszEncoded, char psz, size_t cb)
916	{
917	int rc = VINF_SUCCESS;
918	char szBuf[4];
919	if (!cb)
920	return VERR_BUFFER_OVERFLOW;
921	AssertPtr(psz);
922	for (; *pszEncoded; pszEncoded = RTStrNextCp(pszEncoded))
923	{
924	char *pszDst = szBuf;
925	RTUNICP Cp = RTStrGetCp(pszEncoded);
926	if (Cp == '\\')
927	{
928	pszEncoded = RTStrNextCp(pszEncoded);
929	RTUNICP CpQ = RTStrGetCp(pszEncoded);
930	if (CpQ == 'n')
931	RTStrPutCp(pszDst, '\n');
932	else if (CpQ == 'r')
933	RTStrPutCp(pszDst, '\r');
934	else if (CpQ == '\0')
935	{
936	rc = VERR_VD_VMDK_INVALID_HEADER;
937	break;
938	}
939	else
940	RTStrPutCp(pszDst, CpQ);
941	}
942	else
943	pszDst = RTStrPutCp(pszDst, Cp);
944	/* Need to leave space for terminating NUL. */
945	if ((size_t)(pszDst - szBuf) + 1 >= cb)
946	{
947	rc = VERR_BUFFER_OVERFLOW;
948	break;
949	}
950	memcpy(psz, szBuf, pszDst - szBuf);
951	psz += pszDst - szBuf;
952	}
953	*psz = '\0';
954	return rc;
955	}
956	/**
957	* Internal: free all buffers associated with grain directories.
958	*/
959	static void vmdkFreeGrainDirectory(PVMDKEXTENT pExtent)
960	{
961	if (pExtent->pGD)
962	{
963	RTMemFree(pExtent->pGD);
964	pExtent->pGD = NULL;
965	}
966	if (pExtent->pRGD)
967	{
968	RTMemFree(pExtent->pRGD);
969	pExtent->pRGD = NULL;
970	}
971	}
972	/**
973	* Internal: allocate the compressed/uncompressed buffers for streamOptimized
974	* images.
975	*/
976	static int vmdkAllocStreamBuffers(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
977	{
978	int rc = VINF_SUCCESS;
979	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
980	{
981	/* streamOptimized extents need a compressed grain buffer, which must
982	* be big enough to hold uncompressible data (which needs ~8 bytes
983	* more than the uncompressed data), the marker and padding. */
984	pExtent->cbCompGrain = RT_ALIGN_Z( VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain)
985	+ 8 + sizeof(VMDKMARKER), 512);
986	pExtent->pvCompGrain = RTMemAlloc(pExtent->cbCompGrain);
987	if (RT_LIKELY(pExtent->pvCompGrain))
988	{
989	/* streamOptimized extents need a decompressed grain buffer. */
990	pExtent->pvGrain = RTMemAlloc(VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
991	if (!pExtent->pvGrain)
992	rc = VERR_NO_MEMORY;
993	}
994	else
995	rc = VERR_NO_MEMORY;
996	}
997	if (RT_FAILURE(rc))
998	vmdkFreeStreamBuffers(pExtent);
999	return rc;
1000	}
1001	/**
1002	* Internal: allocate all buffers associated with grain directories.
1003	*/
1004	static int vmdkAllocGrainDirectory(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
1005	{
1006	RT_NOREF1(pImage);
1007	int rc = VINF_SUCCESS;
1008	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
1009	pExtent->pGD = (uint32_t *)RTMemAllocZ(cbGD);
1010	if (RT_LIKELY(pExtent->pGD))
1011	{
1012	if (pExtent->uSectorRGD)
1013	{
1014	pExtent->pRGD = (uint32_t *)RTMemAllocZ(cbGD);
1015	if (RT_UNLIKELY(!pExtent->pRGD))
1016	rc = VERR_NO_MEMORY;
1017	}
1018	}
1019	else
1020	rc = VERR_NO_MEMORY;
1021	if (RT_FAILURE(rc))
1022	vmdkFreeGrainDirectory(pExtent);
1023	return rc;
1024	}
1025	/**
1026	* Converts the grain directory from little to host endianess.
1027	*
1028	* @returns nothing.
1029	* @param pGD The grain directory.
1030	* @param cGDEntries Number of entries in the grain directory to convert.
1031	*/
1032	DECLINLINE(void) vmdkGrainDirectoryConvToHost(uint32_t *pGD, uint32_t cGDEntries)
1033	{
1034	uint32_t *pGDTmp = pGD;
1035	for (uint32_t i = 0; i < cGDEntries; i++, pGDTmp++)
1036	pGDTmp = RT_LE2H_U32(pGDTmp);
1037	}
1038	/**
1039	* Read the grain directory and allocated grain tables verifying them against
1040	* their back up copies if available.
1041	*
1042	* @returns VBox status code.
1043	* @param pImage Image instance data.
1044	* @param pExtent The VMDK extent.
1045	*/
1046	static int vmdkReadGrainDirectory(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
1047	{
1048	int rc = VINF_SUCCESS;
1049	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
1050	AssertReturn(( pExtent->enmType == VMDKETYPE_HOSTED_SPARSE
1051	&& pExtent->uSectorGD != VMDK_GD_AT_END
1052	&& pExtent->uSectorRGD != VMDK_GD_AT_END), VERR_INTERNAL_ERROR);
1053	rc = vmdkAllocGrainDirectory(pImage, pExtent);
1054	if (RT_SUCCESS(rc))
1055	{
1056	/* The VMDK 1.1 spec seems to talk about compressed grain directories,
1057	* but in reality they are not compressed. */
1058	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1059	VMDK_SECTOR2BYTE(pExtent->uSectorGD),
1060	pExtent->pGD, cbGD);
1061	if (RT_SUCCESS(rc))
1062	{
1063	vmdkGrainDirectoryConvToHost(pExtent->pGD, pExtent->cGDEntries);
1064	if ( pExtent->uSectorRGD
1065	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_SKIP_CONSISTENCY_CHECKS))
1066	{
1067	/* The VMDK 1.1 spec seems to talk about compressed grain directories,
1068	* but in reality they are not compressed. */
1069	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1070	VMDK_SECTOR2BYTE(pExtent->uSectorRGD),
1071	pExtent->pRGD, cbGD);
1072	if (RT_SUCCESS(rc))
1073	{
1074	vmdkGrainDirectoryConvToHost(pExtent->pRGD, pExtent->cGDEntries);
1075	/* Check grain table and redundant grain table for consistency. */
1076	size_t cbGT = pExtent->cGTEntries * sizeof(uint32_t);
1077	size_t cbGTBuffers = cbGT; /* Start with space for one GT. */
1078	size_t cbGTBuffersMax = _1M;
1079	uint32_t pTmpGT1 = (uint32_t )RTMemAlloc(cbGTBuffers);
1080	uint32_t pTmpGT2 = (uint32_t )RTMemAlloc(cbGTBuffers);
1081	if ( !pTmpGT1
1082	\|\| !pTmpGT2)
1083	rc = VERR_NO_MEMORY;
1084	size_t i = 0;
1085	uint32_t *pGDTmp = pExtent->pGD;
1086	uint32_t *pRGDTmp = pExtent->pRGD;
1087	/* Loop through all entries. */
1088	while (i < pExtent->cGDEntries)
1089	{
1090	uint32_t uGTStart = *pGDTmp;
1091	uint32_t uRGTStart = *pRGDTmp;
1092	size_t cbGTRead = cbGT;
1093	/* If no grain table is allocated skip the entry. */
1094	if (pGDTmp == 0 && pRGDTmp == 0)
1095	{
1096	i++;
1097	continue;
1098	}
1099	if (pGDTmp == 0 \|\| pRGDTmp == 0 \|\| pGDTmp == pRGDTmp)
1100	{
1101	/* Just one grain directory entry refers to a not yet allocated
1102	* grain table or both grain directory copies refer to the same
1103	* grain table. Not allowed. */
1104	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1105	N_("VMDK: inconsistent references to grain directory in '%s'"), pExtent->pszFullname);
1106	break;
1107	}
1108	i++;
1109	pGDTmp++;
1110	pRGDTmp++;
1111	/*
1112	* Read a few tables at once if adjacent to decrease the number
1113	* of I/O requests. Read at maximum 1MB at once.
1114	*/
1115	while ( i < pExtent->cGDEntries
1116	&& cbGTRead < cbGTBuffersMax)
1117	{
1118	/* If no grain table is allocated skip the entry. */
1119	if (pGDTmp == 0 && pRGDTmp == 0)
1120	{
1121	i++;
1122	continue;
1123	}
1124	if (pGDTmp == 0 \|\| pRGDTmp == 0 \|\| pGDTmp == pRGDTmp)
1125	{
1126	/* Just one grain directory entry refers to a not yet allocated
1127	* grain table or both grain directory copies refer to the same
1128	* grain table. Not allowed. */
1129	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1130	N_("VMDK: inconsistent references to grain directory in '%s'"), pExtent->pszFullname);
1131	break;
1132	}
1133	/* Check that the start offsets are adjacent.*/
1134	if ( VMDK_SECTOR2BYTE(uGTStart) + cbGTRead != VMDK_SECTOR2BYTE(*pGDTmp)
1135	\|\| VMDK_SECTOR2BYTE(uRGTStart) + cbGTRead != VMDK_SECTOR2BYTE(*pRGDTmp))
1136	break;
1137	i++;
1138	pGDTmp++;
1139	pRGDTmp++;
1140	cbGTRead += cbGT;
1141	}
1142	/* Increase buffers if required. */
1143	if ( RT_SUCCESS(rc)
1144	&& cbGTBuffers < cbGTRead)
1145	{
1146	uint32_t *pTmp;
1147	pTmp = (uint32_t *)RTMemRealloc(pTmpGT1, cbGTRead);
1148	if (pTmp)
1149	{
1150	pTmpGT1 = pTmp;
1151	pTmp = (uint32_t *)RTMemRealloc(pTmpGT2, cbGTRead);
1152	if (pTmp)
1153	pTmpGT2 = pTmp;
1154	else
1155	rc = VERR_NO_MEMORY;
1156	}
1157	else
1158	rc = VERR_NO_MEMORY;
1159	if (rc == VERR_NO_MEMORY)
1160	{
1161	/* Reset to the old values. */
1162	rc = VINF_SUCCESS;
1163	i -= cbGTRead / cbGT;
1164	cbGTRead = cbGT;
1165	/* Don't try to increase the buffer again in the next run. */
1166	cbGTBuffersMax = cbGTBuffers;
1167	}
1168	}
1169	if (RT_SUCCESS(rc))
1170	{
1171	/* The VMDK 1.1 spec seems to talk about compressed grain tables,
1172	* but in reality they are not compressed. */
1173	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1174	VMDK_SECTOR2BYTE(uGTStart),
1175	pTmpGT1, cbGTRead);
1176	if (RT_FAILURE(rc))
1177	{
1178	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1179	N_("VMDK: error reading grain table in '%s'"), pExtent->pszFullname);
1180	break;
1181	}
1182	/* The VMDK 1.1 spec seems to talk about compressed grain tables,
1183	* but in reality they are not compressed. */
1184	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1185	VMDK_SECTOR2BYTE(uRGTStart),
1186	pTmpGT2, cbGTRead);
1187	if (RT_FAILURE(rc))
1188	{
1189	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1190	N_("VMDK: error reading backup grain table in '%s'"), pExtent->pszFullname);
1191	break;
1192	}
1193	if (memcmp(pTmpGT1, pTmpGT2, cbGTRead))
1194	{
1195	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1196	N_("VMDK: inconsistency between grain table and backup grain table in '%s'"), pExtent->pszFullname);
1197	break;
1198	}
1199	}
1200	} /* while (i < pExtent->cGDEntries) */
1201	/** @todo figure out what to do for unclean VMDKs. */
1202	if (pTmpGT1)
1203	RTMemFree(pTmpGT1);
1204	if (pTmpGT2)
1205	RTMemFree(pTmpGT2);
1206	}
1207	else
1208	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1209	N_("VMDK: could not read redundant grain directory in '%s'"), pExtent->pszFullname);
1210	}
1211	}
1212	else
1213	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1214	N_("VMDK: could not read grain directory in '%s': %Rrc"), pExtent->pszFullname, rc);
1215	}
1216	if (RT_FAILURE(rc))
1217	vmdkFreeGrainDirectory(pExtent);
1218	return rc;
1219	}
1220	/**
1221	* Creates a new grain directory for the given extent at the given start sector.
1222	*
1223	* @returns VBox status code.
1224	* @param pImage Image instance data.
1225	* @param pExtent The VMDK extent.
1226	* @param uStartSector Where the grain directory should be stored in the image.
1227	* @param fPreAlloc Flag whether to pre allocate the grain tables at this point.
1228	*/
1229	static int vmdkCreateGrainDirectory(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
1230	uint64_t uStartSector, bool fPreAlloc)
1231	{
1232	int rc = VINF_SUCCESS;
1233	unsigned i;
1234	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
1235	size_t cbGDRounded = RT_ALIGN_64(cbGD, 512);
1236	size_t cbGTRounded;
1237	uint64_t cbOverhead;
1238	if (fPreAlloc)
1239	{
1240	cbGTRounded = RT_ALIGN_64(pExtent->cGDEntries * pExtent->cGTEntries * sizeof(uint32_t), 512);
1241	cbOverhead = VMDK_SECTOR2BYTE(uStartSector) + cbGDRounded + cbGTRounded;
1242	}
1243	else
1244	{
1245	/* Use a dummy start sector for layout computation. */
1246	if (uStartSector == VMDK_GD_AT_END)
1247	uStartSector = 1;
1248	cbGTRounded = 0;
1249	cbOverhead = VMDK_SECTOR2BYTE(uStartSector) + cbGDRounded;
1250	}
1251	/* For streamOptimized extents there is only one grain directory,
1252	* and for all others take redundant grain directory into account. */
1253	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
1254	{
1255	cbOverhead = RT_ALIGN_64(cbOverhead,
1256	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
1257	}
1258	else
1259	{
1260	cbOverhead += cbGDRounded + cbGTRounded;
1261	cbOverhead = RT_ALIGN_64(cbOverhead,
1262	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
1263	rc = vdIfIoIntFileSetSize(pImage->pIfIo, pExtent->pFile->pStorage, cbOverhead);
1264	}
1265	if (RT_SUCCESS(rc))
1266	{
1267	pExtent->uAppendPosition = cbOverhead;
1268	pExtent->cOverheadSectors = VMDK_BYTE2SECTOR(cbOverhead);
1269	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
1270	{
1271	pExtent->uSectorRGD = 0;
1272	pExtent->uSectorGD = uStartSector;
1273	}
1274	else
1275	{
1276	pExtent->uSectorRGD = uStartSector;
1277	pExtent->uSectorGD = uStartSector + VMDK_BYTE2SECTOR(cbGDRounded + cbGTRounded);
1278	}
1279	rc = vmdkAllocStreamBuffers(pImage, pExtent);
1280	if (RT_SUCCESS(rc))
1281	{
1282	rc = vmdkAllocGrainDirectory(pImage, pExtent);
1283	if ( RT_SUCCESS(rc)
1284	&& fPreAlloc)
1285	{
1286	uint32_t uGTSectorLE;
1287	uint64_t uOffsetSectors;
1288	if (pExtent->pRGD)
1289	{
1290	uOffsetSectors = pExtent->uSectorRGD + VMDK_BYTE2SECTOR(cbGDRounded);
1291	for (i = 0; i < pExtent->cGDEntries; i++)
1292	{
1293	pExtent->pRGD[i] = uOffsetSectors;
1294	uGTSectorLE = RT_H2LE_U64(uOffsetSectors);
1295	/* Write the redundant grain directory entry to disk. */
1296	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
1297	VMDK_SECTOR2BYTE(pExtent->uSectorRGD) + i * sizeof(uGTSectorLE),
1298	&uGTSectorLE, sizeof(uGTSectorLE));
1299	if (RT_FAILURE(rc))
1300	{
1301	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write new redundant grain directory entry in '%s'"), pExtent->pszFullname);
1302	break;
1303	}
1304	uOffsetSectors += VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
1305	}
1306	}
1307	if (RT_SUCCESS(rc))
1308	{
1309	uOffsetSectors = pExtent->uSectorGD + VMDK_BYTE2SECTOR(cbGDRounded);
1310	for (i = 0; i < pExtent->cGDEntries; i++)
1311	{
1312	pExtent->pGD[i] = uOffsetSectors;
1313	uGTSectorLE = RT_H2LE_U64(uOffsetSectors);
1314	/* Write the grain directory entry to disk. */
1315	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
1316	VMDK_SECTOR2BYTE(pExtent->uSectorGD) + i * sizeof(uGTSectorLE),
1317	&uGTSectorLE, sizeof(uGTSectorLE));
1318	if (RT_FAILURE(rc))
1319	{
1320	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write new grain directory entry in '%s'"), pExtent->pszFullname);
1321	break;
1322	}
1323	uOffsetSectors += VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
1324	}
1325	}
1326	}
1327	}
1328	}
1329	if (RT_FAILURE(rc))
1330	vmdkFreeGrainDirectory(pExtent);
1331	return rc;
1332	}
1333	/**
1334	* Unquotes the given string returning the result in a separate buffer.
1335	*
1336	* @returns VBox status code.
1337	* @param pImage The VMDK image state.
1338	* @param pszStr The string to unquote.
1339	* @param ppszUnquoted Where to store the return value, use RTMemTmpFree to
1340	* free.
1341	* @param ppszNext Where to store the pointer to any character following
1342	* the quoted value, optional.
1343	*/
1344	static int vmdkStringUnquote(PVMDKIMAGE pImage, const char *pszStr,
1345	char ppszUnquoted, char ppszNext)
1346	{
1347	const char *pszStart = pszStr;
1348	char *pszQ;
1349	char *pszUnquoted;
1350	/* Skip over whitespace. */
1351	while (pszStr == ' ' \|\| pszStr == '\t')
1352	pszStr++;
1353	if (*pszStr != '"')
1354	{
1355	pszQ = (char *)pszStr;
1356	while (pszQ && pszQ != ' ' && *pszQ != '\t')
1357	pszQ++;
1358	}
1359	else
1360	{
1361	pszStr++;
1362	pszQ = (char *)strchr(pszStr, '"');
1363	if (pszQ == NULL)
1364	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrectly quoted value in descriptor in '%s' (raw value %s)"),
1365	pImage->pszFilename, pszStart);
1366	}
1367	pszUnquoted = (char *)RTMemTmpAlloc(pszQ - pszStr + 1);
1368	if (!pszUnquoted)
1369	return VERR_NO_MEMORY;
1370	memcpy(pszUnquoted, pszStr, pszQ - pszStr);
1371	pszUnquoted[pszQ - pszStr] = '\0';
1372	*ppszUnquoted = pszUnquoted;
1373	if (ppszNext)
1374	*ppszNext = pszQ + 1;
1375	return VINF_SUCCESS;
1376	}
1377	static int vmdkDescInitStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1378	const char *pszLine)
1379	{
1380	char *pEnd = pDescriptor->aLines[pDescriptor->cLines];
1381	ssize_t cbDiff = strlen(pszLine) + 1;
1382	if ( pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1
1383	&& pEnd - pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff)
1384	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1385	memcpy(pEnd, pszLine, cbDiff);
1386	pDescriptor->cLines++;
1387	pDescriptor->aLines[pDescriptor->cLines] = pEnd + cbDiff;
1388	pDescriptor->fDirty = true;
1389	return VINF_SUCCESS;
1390	}
1391	static bool vmdkDescGetStr(PVMDKDESCRIPTOR pDescriptor, unsigned uStart,
1392	const char pszKey, const char *ppszValue)
1393	{
1394	size_t cbKey = strlen(pszKey);
1395	const char *pszValue;
1396	while (uStart != 0)
1397	{
1398	if (!strncmp(pDescriptor->aLines[uStart], pszKey, cbKey))
1399	{
1400	/* Key matches, check for a '=' (preceded by whitespace). */
1401	pszValue = pDescriptor->aLines[uStart] + cbKey;
1402	while (pszValue == ' ' \|\| pszValue == '\t')
1403	pszValue++;
1404	if (*pszValue == '=')
1405	{
1406	*ppszValue = pszValue + 1;
1407	break;
1408	}
1409	}
1410	uStart = pDescriptor->aNextLines[uStart];
1411	}
1412	return !!uStart;
1413	}
1414	static int vmdkDescSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1415	unsigned uStart,
1416	const char pszKey, const char pszValue)
1417	{
1418	char pszTmp = NULL; / (MSC naturally cannot figure this isn't used uninitialized) */
1419	size_t cbKey = strlen(pszKey);
1420	unsigned uLast = 0;
1421	while (uStart != 0)
1422	{
1423	if (!strncmp(pDescriptor->aLines[uStart], pszKey, cbKey))
1424	{
1425	/* Key matches, check for a '=' (preceded by whitespace). */
1426	pszTmp = pDescriptor->aLines[uStart] + cbKey;
1427	while (pszTmp == ' ' \|\| pszTmp == '\t')
1428	pszTmp++;
1429	if (*pszTmp == '=')
1430	{
1431	pszTmp++;
1432	/** @todo r=bird: Doesn't skipping trailing blanks here just cause unecessary
1433	* bloat and potentially out of space error? */
1434	while (pszTmp == ' ' \|\| pszTmp == '\t')
1435	pszTmp++;
1436	break;
1437	}
1438	}
1439	if (!pDescriptor->aNextLines[uStart])
1440	uLast = uStart;
1441	uStart = pDescriptor->aNextLines[uStart];
1442	}
1443	if (uStart)
1444	{
1445	if (pszValue)
1446	{
1447	/* Key already exists, replace existing value. */
1448	size_t cbOldVal = strlen(pszTmp);
1449	size_t cbNewVal = strlen(pszValue);
1450	ssize_t cbDiff = cbNewVal - cbOldVal;
1451	/* Check for buffer overflow. */
1452	if ( pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[0]
1453	> (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff)
1454	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1455	memmove(pszTmp + cbNewVal, pszTmp + cbOldVal,
1456	pDescriptor->aLines[pDescriptor->cLines] - pszTmp - cbOldVal);
1457	memcpy(pszTmp, pszValue, cbNewVal + 1);
1458	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1459	pDescriptor->aLines[i] += cbDiff;
1460	}
1461	else
1462	{
1463	memmove(pDescriptor->aLines[uStart], pDescriptor->aLines[uStart+1],
1464	pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[uStart+1] + 1);
1465	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1466	{
1467	pDescriptor->aLines[i-1] = pDescriptor->aLines[i];
1468	if (pDescriptor->aNextLines[i])
1469	pDescriptor->aNextLines[i-1] = pDescriptor->aNextLines[i] - 1;
1470	else
1471	pDescriptor->aNextLines[i-1] = 0;
1472	}
1473	pDescriptor->cLines--;
1474	/* Adjust starting line numbers of following descriptor sections. */
1475	if (uStart < pDescriptor->uFirstExtent)
1476	pDescriptor->uFirstExtent--;
1477	if (uStart < pDescriptor->uFirstDDB)
1478	pDescriptor->uFirstDDB--;
1479	}
1480	}
1481	else
1482	{
1483	/* Key doesn't exist, append after the last entry in this category. */
1484	if (!pszValue)
1485	{
1486	/* Key doesn't exist, and it should be removed. Simply a no-op. */
1487	return VINF_SUCCESS;
1488	}
1489	cbKey = strlen(pszKey);
1490	size_t cbValue = strlen(pszValue);
1491	ssize_t cbDiff = cbKey + 1 + cbValue + 1;
1492	/* Check for buffer overflow. */
1493	if ( (pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1)
1494	\|\| ( pDescriptor->aLines[pDescriptor->cLines]
1495	- pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff))
1496	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1497	for (unsigned i = pDescriptor->cLines + 1; i > uLast + 1; i--)
1498	{
1499	pDescriptor->aLines[i] = pDescriptor->aLines[i - 1];
1500	if (pDescriptor->aNextLines[i - 1])
1501	pDescriptor->aNextLines[i] = pDescriptor->aNextLines[i - 1] + 1;
1502	else
1503	pDescriptor->aNextLines[i] = 0;
1504	}
1505	uStart = uLast + 1;
1506	pDescriptor->aNextLines[uLast] = uStart;
1507	pDescriptor->aNextLines[uStart] = 0;
1508	pDescriptor->cLines++;
1509	pszTmp = pDescriptor->aLines[uStart];
1510	memmove(pszTmp + cbDiff, pszTmp,
1511	pDescriptor->aLines[pDescriptor->cLines] - pszTmp);
1512	memcpy(pDescriptor->aLines[uStart], pszKey, cbKey);
1513	pDescriptor->aLines[uStart][cbKey] = '=';
1514	memcpy(pDescriptor->aLines[uStart] + cbKey + 1, pszValue, cbValue + 1);
1515	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1516	pDescriptor->aLines[i] += cbDiff;
1517	/* Adjust starting line numbers of following descriptor sections. */
1518	if (uStart <= pDescriptor->uFirstExtent)
1519	pDescriptor->uFirstExtent++;
1520	if (uStart <= pDescriptor->uFirstDDB)
1521	pDescriptor->uFirstDDB++;
1522	}
1523	pDescriptor->fDirty = true;
1524	return VINF_SUCCESS;
1525	}
1526	static int vmdkDescBaseGetU32(PVMDKDESCRIPTOR pDescriptor, const char *pszKey,
1527	uint32_t *puValue)
1528	{
1529	const char *pszValue;
1530	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDesc, pszKey,
1531	&pszValue))
1532	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1533	return RTStrToUInt32Ex(pszValue, NULL, 10, puValue);
1534	}
1535	/**
1536	* Returns the value of the given key as a string allocating the necessary memory.
1537	*
1538	* @returns VBox status code.
1539	* @retval VERR_VD_VMDK_VALUE_NOT_FOUND if the value could not be found.
1540	* @param pImage The VMDK image state.
1541	* @param pDescriptor The descriptor to fetch the value from.
1542	* @param pszKey The key to get the value from.
1543	* @param ppszValue Where to store the return value, use RTMemTmpFree to
1544	* free.
1545	*/
1546	static int vmdkDescBaseGetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1547	const char pszKey, char *ppszValue)
1548	{
1549	const char *pszValue;
1550	char *pszValueUnquoted;
1551	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDesc, pszKey,
1552	&pszValue))
1553	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1554	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1555	if (RT_FAILURE(rc))
1556	return rc;
1557	*ppszValue = pszValueUnquoted;
1558	return rc;
1559	}
1560	static int vmdkDescBaseSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1561	const char pszKey, const char pszValue)
1562	{
1563	char *pszValueQuoted;
1564	RTStrAPrintf(&pszValueQuoted, "\"%s\"", pszValue);
1565	if (!pszValueQuoted)
1566	return VERR_NO_STR_MEMORY;
1567	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc, pszKey,
1568	pszValueQuoted);
1569	RTStrFree(pszValueQuoted);
1570	return rc;
1571	}
1572	static void vmdkDescExtRemoveDummy(PVMDKIMAGE pImage,
1573	PVMDKDESCRIPTOR pDescriptor)
1574	{
1575	RT_NOREF1(pImage);
1576	unsigned uEntry = pDescriptor->uFirstExtent;
1577	ssize_t cbDiff;
1578	if (!uEntry)
1579	return;
1580	cbDiff = strlen(pDescriptor->aLines[uEntry]) + 1;
1581	/* Move everything including \0 in the entry marking the end of buffer. */
1582	memmove(pDescriptor->aLines[uEntry], pDescriptor->aLines[uEntry + 1],
1583	pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[uEntry + 1] + 1);
1584	for (unsigned i = uEntry + 1; i <= pDescriptor->cLines; i++)
1585	{
1586	pDescriptor->aLines[i - 1] = pDescriptor->aLines[i] - cbDiff;
1587	if (pDescriptor->aNextLines[i])
1588	pDescriptor->aNextLines[i - 1] = pDescriptor->aNextLines[i] - 1;
1589	else
1590	pDescriptor->aNextLines[i - 1] = 0;
1591	}
1592	pDescriptor->cLines--;
1593	if (pDescriptor->uFirstDDB)
1594	pDescriptor->uFirstDDB--;
1595	return;
1596	}
1597	static int vmdkDescExtInsert(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1598	VMDKACCESS enmAccess, uint64_t cNominalSectors,
1599	VMDKETYPE enmType, const char *pszBasename,
1600	uint64_t uSectorOffset)
1601	{
1602	static const char *apszAccess[] = { "NOACCESS", "RDONLY", "RW" };
1603	static const char *apszType[] = { "", "SPARSE", "FLAT", "ZERO", "VMFS" };
1604	char *pszTmp;
1605	unsigned uStart = pDescriptor->uFirstExtent, uLast = 0;
1606	char szExt[1024];
1607	ssize_t cbDiff;
1608	Assert((unsigned)enmAccess < RT_ELEMENTS(apszAccess));
1609	Assert((unsigned)enmType < RT_ELEMENTS(apszType));
1610	/* Find last entry in extent description. */
1611	while (uStart)
1612	{
1613	if (!pDescriptor->aNextLines[uStart])
1614	uLast = uStart;
1615	uStart = pDescriptor->aNextLines[uStart];
1616	}
1617	if (enmType == VMDKETYPE_ZERO)
1618	{
1619	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s ", apszAccess[enmAccess],
1620	cNominalSectors, apszType[enmType]);
1621	}
1622	else if (enmType == VMDKETYPE_FLAT)
1623	{
1624	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s \"%s\" %llu",
1625	apszAccess[enmAccess], cNominalSectors,
1626	apszType[enmType], pszBasename, uSectorOffset);
1627	}
1628	else
1629	{
1630	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s \"%s\"",
1631	apszAccess[enmAccess], cNominalSectors,
1632	apszType[enmType], pszBasename);
1633	}
1634	cbDiff = strlen(szExt) + 1;
1635	/* Check for buffer overflow. */
1636	if ( (pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1)
1637	\|\| ( pDescriptor->aLines[pDescriptor->cLines]
1638	- pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff))
1639	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1640	for (unsigned i = pDescriptor->cLines + 1; i > uLast + 1; i--)
1641	{
1642	pDescriptor->aLines[i] = pDescriptor->aLines[i - 1];
1643	if (pDescriptor->aNextLines[i - 1])
1644	pDescriptor->aNextLines[i] = pDescriptor->aNextLines[i - 1] + 1;
1645	else
1646	pDescriptor->aNextLines[i] = 0;
1647	}
1648	uStart = uLast + 1;
1649	pDescriptor->aNextLines[uLast] = uStart;
1650	pDescriptor->aNextLines[uStart] = 0;
1651	pDescriptor->cLines++;
1652	pszTmp = pDescriptor->aLines[uStart];
1653	memmove(pszTmp + cbDiff, pszTmp,
1654	pDescriptor->aLines[pDescriptor->cLines] - pszTmp);
1655	memcpy(pDescriptor->aLines[uStart], szExt, cbDiff);
1656	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1657	pDescriptor->aLines[i] += cbDiff;
1658	/* Adjust starting line numbers of following descriptor sections. */
1659	if (uStart <= pDescriptor->uFirstDDB)
1660	pDescriptor->uFirstDDB++;
1661	pDescriptor->fDirty = true;
1662	return VINF_SUCCESS;
1663	}
1664	/**
1665	* Returns the value of the given key from the DDB as a string allocating
1666	* the necessary memory.
1667	*
1668	* @returns VBox status code.
1669	* @retval VERR_VD_VMDK_VALUE_NOT_FOUND if the value could not be found.
1670	* @param pImage The VMDK image state.
1671	* @param pDescriptor The descriptor to fetch the value from.
1672	* @param pszKey The key to get the value from.
1673	* @param ppszValue Where to store the return value, use RTMemTmpFree to
1674	* free.
1675	*/
1676	static int vmdkDescDDBGetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1677	const char pszKey, char *ppszValue)
1678	{
1679	const char *pszValue;
1680	char *pszValueUnquoted;
1681	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1682	&pszValue))
1683	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1684	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1685	if (RT_FAILURE(rc))
1686	return rc;
1687	*ppszValue = pszValueUnquoted;
1688	return rc;
1689	}
1690	static int vmdkDescDDBGetU32(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1691	const char pszKey, uint32_t puValue)
1692	{
1693	const char *pszValue;
1694	char *pszValueUnquoted;
1695	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1696	&pszValue))
1697	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1698	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1699	if (RT_FAILURE(rc))
1700	return rc;
1701	rc = RTStrToUInt32Ex(pszValueUnquoted, NULL, 10, puValue);
1702	RTMemTmpFree(pszValueUnquoted);
1703	return rc;
1704	}
1705	static int vmdkDescDDBGetUuid(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1706	const char *pszKey, PRTUUID pUuid)
1707	{
1708	const char *pszValue;
1709	char *pszValueUnquoted;
1710	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1711	&pszValue))
1712	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1713	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1714	if (RT_FAILURE(rc))
1715	return rc;
1716	rc = RTUuidFromStr(pUuid, pszValueUnquoted);
1717	RTMemTmpFree(pszValueUnquoted);
1718	return rc;
1719	}
1720	static int vmdkDescDDBSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1721	const char pszKey, const char pszVal)
1722	{
1723	int rc;
1724	char *pszValQuoted;
1725	if (pszVal)
1726	{
1727	RTStrAPrintf(&pszValQuoted, "\"%s\"", pszVal);
1728	if (!pszValQuoted)
1729	return VERR_NO_STR_MEMORY;
1730	}
1731	else
1732	pszValQuoted = NULL;
1733	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1734	pszValQuoted);
1735	if (pszValQuoted)
1736	RTStrFree(pszValQuoted);
1737	return rc;
1738	}
1739	static int vmdkDescDDBSetUuid(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1740	const char *pszKey, PCRTUUID pUuid)
1741	{
1742	char *pszUuid;
1743	RTStrAPrintf(&pszUuid, "\"%RTuuid\"", pUuid);
1744	if (!pszUuid)
1745	return VERR_NO_STR_MEMORY;
1746	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1747	pszUuid);
1748	RTStrFree(pszUuid);
1749	return rc;
1750	}
1751	static int vmdkDescDDBSetU32(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1752	const char *pszKey, uint32_t uValue)
1753	{
1754	char *pszValue;
1755	RTStrAPrintf(&pszValue, "\"%d\"", uValue);
1756	if (!pszValue)
1757	return VERR_NO_STR_MEMORY;
1758	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1759	pszValue);
1760	RTStrFree(pszValue);
1761	return rc;
1762	}
1763	/**
1764	* Splits the descriptor data into individual lines checking for correct line
1765	* endings and descriptor size.
1766	*
1767	* @returns VBox status code.
1768	* @param pImage The image instance.
1769	* @param pDesc The descriptor.
1770	* @param pszTmp The raw descriptor data from the image.
1771	*/
1772	static int vmdkDescSplitLines(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDesc, char *pszTmp)
1773	{
1774	unsigned cLine = 0;
1775	int rc = VINF_SUCCESS;
1776	while ( RT_SUCCESS(rc)
1777	&& *pszTmp != '\0')
1778	{
1779	pDesc->aLines[cLine++] = pszTmp;
1780	if (cLine >= VMDK_DESCRIPTOR_LINES_MAX)
1781	{
1782	vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1783	rc = VERR_VD_VMDK_INVALID_HEADER;
1784	break;
1785	}
1786	while (pszTmp != '\0' && pszTmp != '\n')
1787	{
1788	if (*pszTmp == '\r')
1789	{
1790	if (*(pszTmp + 1) != '\n')
1791	{
1792	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: unsupported end of line in descriptor in '%s'"), pImage->pszFilename);
1793	break;
1794	}
1795	else
1796	{
1797	/* Get rid of CR character. */
1798	*pszTmp = '\0';
1799	}
1800	}
1801	pszTmp++;
1802	}
1803	if (RT_FAILURE(rc))
1804	break;
1805	/* Get rid of LF character. */
1806	if (*pszTmp == '\n')
1807	{
1808	*pszTmp = '\0';
1809	pszTmp++;
1810	}
1811	}
1812	if (RT_SUCCESS(rc))
1813	{
1814	pDesc->cLines = cLine;
1815	/* Pointer right after the end of the used part of the buffer. */
1816	pDesc->aLines[cLine] = pszTmp;
1817	}
1818	return rc;
1819	}
1820	static int vmdkPreprocessDescriptor(PVMDKIMAGE pImage, char *pDescData,
1821	size_t cbDescData, PVMDKDESCRIPTOR pDescriptor)
1822	{
1823	pDescriptor->cbDescAlloc = cbDescData;
1824	int rc = vmdkDescSplitLines(pImage, pDescriptor, pDescData);
1825	if (RT_SUCCESS(rc))
1826	{
1827	if ( strcmp(pDescriptor->aLines[0], "# Disk DescriptorFile")
1828	&& strcmp(pDescriptor->aLines[0], "# Disk Descriptor File")
1829	&& strcmp(pDescriptor->aLines[0], "#Disk Descriptor File")
1830	&& strcmp(pDescriptor->aLines[0], "#Disk DescriptorFile"))
1831	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1832	N_("VMDK: descriptor does not start as expected in '%s'"), pImage->pszFilename);
1833	else
1834	{
1835	unsigned uLastNonEmptyLine = 0;
1836	/* Initialize those, because we need to be able to reopen an image. */
1837	pDescriptor->uFirstDesc = 0;
1838	pDescriptor->uFirstExtent = 0;
1839	pDescriptor->uFirstDDB = 0;
1840	for (unsigned i = 0; i < pDescriptor->cLines; i++)
1841	{
1842	if (pDescriptor->aLines[i] != '#' && pDescriptor->aLines[i] != '\0')
1843	{
1844	if ( !strncmp(pDescriptor->aLines[i], "RW", 2)
1845	\|\| !strncmp(pDescriptor->aLines[i], "RDONLY", 6)
1846	\|\| !strncmp(pDescriptor->aLines[i], "NOACCESS", 8) )
1847	{
1848	/* An extent descriptor. */
1849	if (!pDescriptor->uFirstDesc \|\| pDescriptor->uFirstDDB)
1850	{
1851	/* Incorrect ordering of entries. */
1852	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1853	N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
1854	break;
1855	}
1856	if (!pDescriptor->uFirstExtent)
1857	{
1858	pDescriptor->uFirstExtent = i;
1859	uLastNonEmptyLine = 0;
1860	}
1861	}
1862	else if (!strncmp(pDescriptor->aLines[i], "ddb.", 4))
1863	{
1864	/* A disk database entry. */
1865	if (!pDescriptor->uFirstDesc \|\| !pDescriptor->uFirstExtent)
1866	{
1867	/* Incorrect ordering of entries. */
1868	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1869	N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
1870	break;
1871	}
1872	if (!pDescriptor->uFirstDDB)
1873	{
1874	pDescriptor->uFirstDDB = i;
1875	uLastNonEmptyLine = 0;
1876	}
1877	}
1878	else
1879	{
1880	/* A normal entry. */
1881	if (pDescriptor->uFirstExtent \|\| pDescriptor->uFirstDDB)
1882	{
1883	/* Incorrect ordering of entries. */
1884	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1885	N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
1886	break;
1887	}
1888	if (!pDescriptor->uFirstDesc)
1889	{
1890	pDescriptor->uFirstDesc = i;
1891	uLastNonEmptyLine = 0;
1892	}
1893	}
1894	if (uLastNonEmptyLine)
1895	pDescriptor->aNextLines[uLastNonEmptyLine] = i;
1896	uLastNonEmptyLine = i;
1897	}
1898	}
1899	}
1900	}
1901	return rc;
1902	}
1903	static int vmdkDescSetPCHSGeometry(PVMDKIMAGE pImage,
1904	PCVDGEOMETRY pPCHSGeometry)
1905	{
1906	int rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1907	VMDK_DDB_GEO_PCHS_CYLINDERS,
1908	pPCHSGeometry->cCylinders);
1909	if (RT_FAILURE(rc))
1910	return rc;
1911	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1912	VMDK_DDB_GEO_PCHS_HEADS,
1913	pPCHSGeometry->cHeads);
1914	if (RT_FAILURE(rc))
1915	return rc;
1916	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1917	VMDK_DDB_GEO_PCHS_SECTORS,
1918	pPCHSGeometry->cSectors);
1919	return rc;
1920	}
1921	static int vmdkDescSetLCHSGeometry(PVMDKIMAGE pImage,
1922	PCVDGEOMETRY pLCHSGeometry)
1923	{
1924	int rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1925	VMDK_DDB_GEO_LCHS_CYLINDERS,
1926	pLCHSGeometry->cCylinders);
1927	if (RT_FAILURE(rc))
1928	return rc;
1929	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1930	VMDK_DDB_GEO_LCHS_HEADS,
1931	pLCHSGeometry->cHeads);
1932	if (RT_FAILURE(rc))
1933	return rc;
1934	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1935	VMDK_DDB_GEO_LCHS_SECTORS,
1936	pLCHSGeometry->cSectors);
1937	return rc;
1938	}
1939	static int vmdkCreateDescriptor(PVMDKIMAGE pImage, char *pDescData,
1940	size_t cbDescData, PVMDKDESCRIPTOR pDescriptor)
1941	{
1942	pDescriptor->uFirstDesc = 0;
1943	pDescriptor->uFirstExtent = 0;
1944	pDescriptor->uFirstDDB = 0;
1945	pDescriptor->cLines = 0;
1946	pDescriptor->cbDescAlloc = cbDescData;
1947	pDescriptor->fDirty = false;
1948	pDescriptor->aLines[pDescriptor->cLines] = pDescData;
1949	memset(pDescriptor->aNextLines, '\0', sizeof(pDescriptor->aNextLines));
1950	int rc = vmdkDescInitStr(pImage, pDescriptor, "# Disk DescriptorFile");
1951	if (RT_SUCCESS(rc))
1952	rc = vmdkDescInitStr(pImage, pDescriptor, "version=1");
1953	if (RT_SUCCESS(rc))
1954	{
1955	pDescriptor->uFirstDesc = pDescriptor->cLines - 1;
1956	rc = vmdkDescInitStr(pImage, pDescriptor, "");
1957	}
1958	if (RT_SUCCESS(rc))
1959	rc = vmdkDescInitStr(pImage, pDescriptor, "# Extent description");
1960	if (RT_SUCCESS(rc))
1961	rc = vmdkDescInitStr(pImage, pDescriptor, "NOACCESS 0 ZERO ");
1962	if (RT_SUCCESS(rc))
1963	{
1964	pDescriptor->uFirstExtent = pDescriptor->cLines - 1;
1965	rc = vmdkDescInitStr(pImage, pDescriptor, "");
1966	}
1967	if (RT_SUCCESS(rc))
1968	{
1969	/* The trailing space is created by VMware, too. */
1970	rc = vmdkDescInitStr(pImage, pDescriptor, "# The disk Data Base ");
1971	}
1972	if (RT_SUCCESS(rc))
1973	rc = vmdkDescInitStr(pImage, pDescriptor, "#DDB");
1974	if (RT_SUCCESS(rc))
1975	rc = vmdkDescInitStr(pImage, pDescriptor, "");
1976	if (RT_SUCCESS(rc))
1977	rc = vmdkDescInitStr(pImage, pDescriptor, "ddb.virtualHWVersion = \"4\"");
1978	if (RT_SUCCESS(rc))
1979	{
1980	pDescriptor->uFirstDDB = pDescriptor->cLines - 1;
1981	/* Now that the framework is in place, use the normal functions to insert
1982	* the remaining keys. */
1983	char szBuf[9];
1984	RTStrPrintf(szBuf, sizeof(szBuf), "%08x", RTRandU32());
1985	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc,
1986	"CID", szBuf);
1987	}
1988	if (RT_SUCCESS(rc))
1989	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc,
1990	"parentCID", "ffffffff");
1991	if (RT_SUCCESS(rc))
1992	rc = vmdkDescDDBSetStr(pImage, pDescriptor, "ddb.adapterType", "ide");
1993	return rc;
1994	}
1995	static int vmdkParseDescriptor(PVMDKIMAGE pImage, char *pDescData, size_t cbDescData)
1996	{
1997	int rc;
1998	unsigned cExtents;
1999	unsigned uLine;
2000	unsigned i;
2001	rc = vmdkPreprocessDescriptor(pImage, pDescData, cbDescData,
2002	&pImage->Descriptor);
2003	if (RT_FAILURE(rc))
2004	return rc;
2005	/* Check version, must be 1. */
2006	uint32_t uVersion;
2007	rc = vmdkDescBaseGetU32(&pImage->Descriptor, "version", &uVersion);
2008	if (RT_FAILURE(rc))
2009	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error finding key 'version' in descriptor in '%s'"), pImage->pszFilename);
2010	if (uVersion != 1)
2011	return vdIfError(pImage->pIfError, VERR_VD_VMDK_UNSUPPORTED_VERSION, RT_SRC_POS, N_("VMDK: unsupported format version in descriptor in '%s'"), pImage->pszFilename);
2012	/* Get image creation type and determine image flags. */
2013	char pszCreateType = NULL; / initialized to make gcc shut up */
2014	rc = vmdkDescBaseGetStr(pImage, &pImage->Descriptor, "createType",
2015	&pszCreateType);
2016	if (RT_FAILURE(rc))
2017	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot get image type from descriptor in '%s'"), pImage->pszFilename);
2018	if ( !strcmp(pszCreateType, "twoGbMaxExtentSparse")
2019	\|\| !strcmp(pszCreateType, "twoGbMaxExtentFlat"))
2020	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_SPLIT_2G;
2021	else if ( !strcmp(pszCreateType, "partitionedDevice")
2022	\|\| !strcmp(pszCreateType, "fullDevice"))
2023	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_RAWDISK;
2024	else if (!strcmp(pszCreateType, "streamOptimized"))
2025	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED;
2026	else if (!strcmp(pszCreateType, "vmfs"))
2027	pImage->uImageFlags \|= VD_IMAGE_FLAGS_FIXED \| VD_VMDK_IMAGE_FLAGS_ESX;
2028	RTMemTmpFree(pszCreateType);
2029	/* Count the number of extent config entries. */
2030	for (uLine = pImage->Descriptor.uFirstExtent, cExtents = 0;
2031	uLine != 0;
2032	uLine = pImage->Descriptor.aNextLines[uLine], cExtents++)
2033	/* nothing */;
2034	if (!pImage->pDescData && cExtents != 1)
2035	{
2036	/* Monolithic image, must have only one extent (already opened). */
2037	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);
2038	}
2039	if (pImage->pDescData)
2040	{
2041	/* Non-monolithic image, extents need to be allocated. */
2042	rc = vmdkCreateExtents(pImage, cExtents);
2043	if (RT_FAILURE(rc))
2044	return rc;
2045	}
2046	for (i = 0, uLine = pImage->Descriptor.uFirstExtent;
2047	i < cExtents; i++, uLine = pImage->Descriptor.aNextLines[uLine])
2048	{
2049	char *pszLine = pImage->Descriptor.aLines[uLine];
2050	/* Access type of the extent. */
2051	if (!strncmp(pszLine, "RW", 2))
2052	{
2053	pImage->pExtents[i].enmAccess = VMDKACCESS_READWRITE;
2054	pszLine += 2;
2055	}
2056	else if (!strncmp(pszLine, "RDONLY", 6))
2057	{
2058	pImage->pExtents[i].enmAccess = VMDKACCESS_READONLY;
2059	pszLine += 6;
2060	}
2061	else if (!strncmp(pszLine, "NOACCESS", 8))
2062	{
2063	pImage->pExtents[i].enmAccess = VMDKACCESS_NOACCESS;
2064	pszLine += 8;
2065	}
2066	else
2067	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2068	if (*pszLine++ != ' ')
2069	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2070	/* Nominal size of the extent. */
2071	rc = RTStrToUInt64Ex(pszLine, &pszLine, 10,
2072	&pImage->pExtents[i].cNominalSectors);
2073	if (RT_FAILURE(rc))
2074	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2075	if (*pszLine++ != ' ')
2076	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2077	/* Type of the extent. */
2078	if (!strncmp(pszLine, "SPARSE", 6))
2079	{
2080	pImage->pExtents[i].enmType = VMDKETYPE_HOSTED_SPARSE;
2081	pszLine += 6;
2082	}
2083	else if (!strncmp(pszLine, "FLAT", 4))
2084	{
2085	pImage->pExtents[i].enmType = VMDKETYPE_FLAT;
2086	pszLine += 4;
2087	}
2088	else if (!strncmp(pszLine, "ZERO", 4))
2089	{
2090	pImage->pExtents[i].enmType = VMDKETYPE_ZERO;
2091	pszLine += 4;
2092	}
2093	else if (!strncmp(pszLine, "VMFS", 4))
2094	{
2095	pImage->pExtents[i].enmType = VMDKETYPE_VMFS;
2096	pszLine += 4;
2097	}
2098	else
2099	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2100	if (pImage->pExtents[i].enmType == VMDKETYPE_ZERO)
2101	{
2102	/* This one has no basename or offset. */
2103	if (*pszLine == ' ')
2104	pszLine++;
2105	if (*pszLine != '\0')
2106	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2107	pImage->pExtents[i].pszBasename = NULL;
2108	}
2109	else
2110	{
2111	/* All other extent types have basename and optional offset. */
2112	if (*pszLine++ != ' ')
2113	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2114	/* Basename of the image. Surrounded by quotes. */
2115	char *pszBasename;
2116	rc = vmdkStringUnquote(pImage, pszLine, &pszBasename, &pszLine);
2117	if (RT_FAILURE(rc))
2118	return rc;
2119	pImage->pExtents[i].pszBasename = pszBasename;
2120	if (*pszLine == ' ')
2121	{
2122	pszLine++;
2123	if (*pszLine != '\0')
2124	{
2125	/* Optional offset in extent specified. */
2126	rc = RTStrToUInt64Ex(pszLine, &pszLine, 10,
2127	&pImage->pExtents[i].uSectorOffset);
2128	if (RT_FAILURE(rc))
2129	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2130	}
2131	}
2132	if (*pszLine != '\0')
2133	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2134	}
2135	}
2136	/* Determine PCHS geometry (autogenerate if necessary). */
2137	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2138	VMDK_DDB_GEO_PCHS_CYLINDERS,
2139	&pImage->PCHSGeometry.cCylinders);
2140	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2141	pImage->PCHSGeometry.cCylinders = 0;
2142	else if (RT_FAILURE(rc))
2143	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2144	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2145	VMDK_DDB_GEO_PCHS_HEADS,
2146	&pImage->PCHSGeometry.cHeads);
2147	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2148	pImage->PCHSGeometry.cHeads = 0;
2149	else if (RT_FAILURE(rc))
2150	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2151	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2152	VMDK_DDB_GEO_PCHS_SECTORS,
2153	&pImage->PCHSGeometry.cSectors);
2154	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2155	pImage->PCHSGeometry.cSectors = 0;
2156	else if (RT_FAILURE(rc))
2157	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2158	if ( pImage->PCHSGeometry.cCylinders == 0
2159	\|\| pImage->PCHSGeometry.cHeads == 0
2160	\|\| pImage->PCHSGeometry.cHeads > 16
2161	\|\| pImage->PCHSGeometry.cSectors == 0
2162	\|\| pImage->PCHSGeometry.cSectors > 63)
2163	{
2164	/* Mark PCHS geometry as not yet valid (can't do the calculation here
2165	* as the total image size isn't known yet). */
2166	pImage->PCHSGeometry.cCylinders = 0;
2167	pImage->PCHSGeometry.cHeads = 16;
2168	pImage->PCHSGeometry.cSectors = 63;
2169	}
2170	/* Determine LCHS geometry (set to 0 if not specified). */
2171	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2172	VMDK_DDB_GEO_LCHS_CYLINDERS,
2173	&pImage->LCHSGeometry.cCylinders);
2174	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2175	pImage->LCHSGeometry.cCylinders = 0;
2176	else if (RT_FAILURE(rc))
2177	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2178	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2179	VMDK_DDB_GEO_LCHS_HEADS,
2180	&pImage->LCHSGeometry.cHeads);
2181	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2182	pImage->LCHSGeometry.cHeads = 0;
2183	else if (RT_FAILURE(rc))
2184	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2185	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2186	VMDK_DDB_GEO_LCHS_SECTORS,
2187	&pImage->LCHSGeometry.cSectors);
2188	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2189	pImage->LCHSGeometry.cSectors = 0;
2190	else if (RT_FAILURE(rc))
2191	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2192	if ( pImage->LCHSGeometry.cCylinders == 0
2193	\|\| pImage->LCHSGeometry.cHeads == 0
2194	\|\| pImage->LCHSGeometry.cSectors == 0)
2195	{
2196	pImage->LCHSGeometry.cCylinders = 0;
2197	pImage->LCHSGeometry.cHeads = 0;
2198	pImage->LCHSGeometry.cSectors = 0;
2199	}
2200	/* Get image UUID. */
2201	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor, VMDK_DDB_IMAGE_UUID,
2202	&pImage->ImageUuid);
2203	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2204	{
2205	/* Image without UUID. Probably created by VMware and not yet used
2206	* by VirtualBox. Can only be added for images opened in read/write
2207	* mode, so don't bother producing a sensible UUID otherwise. */
2208	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2209	RTUuidClear(&pImage->ImageUuid);
2210	else
2211	{
2212	rc = RTUuidCreate(&pImage->ImageUuid);
2213	if (RT_FAILURE(rc))
2214	return rc;
2215	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2216	VMDK_DDB_IMAGE_UUID, &pImage->ImageUuid);
2217	if (RT_FAILURE(rc))
2218	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image UUID in descriptor in '%s'"), pImage->pszFilename);
2219	}
2220	}
2221	else if (RT_FAILURE(rc))
2222	return rc;
2223	/* Get image modification UUID. */
2224	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor,
2225	VMDK_DDB_MODIFICATION_UUID,
2226	&pImage->ModificationUuid);
2227	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2228	{
2229	/* Image without UUID. Probably created by VMware and not yet used
2230	* by VirtualBox. Can only be added for images opened in read/write
2231	* mode, so don't bother producing a sensible UUID otherwise. */
2232	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2233	RTUuidClear(&pImage->ModificationUuid);
2234	else
2235	{
2236	rc = RTUuidCreate(&pImage->ModificationUuid);
2237	if (RT_FAILURE(rc))
2238	return rc;
2239	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2240	VMDK_DDB_MODIFICATION_UUID,
2241	&pImage->ModificationUuid);
2242	if (RT_FAILURE(rc))
2243	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image modification UUID in descriptor in '%s'"), pImage->pszFilename);
2244	}
2245	}
2246	else if (RT_FAILURE(rc))
2247	return rc;
2248	/* Get UUID of parent image. */
2249	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor, VMDK_DDB_PARENT_UUID,
2250	&pImage->ParentUuid);
2251	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2252	{
2253	/* Image without UUID. Probably created by VMware and not yet used
2254	* by VirtualBox. Can only be added for images opened in read/write
2255	* mode, so don't bother producing a sensible UUID otherwise. */
2256	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2257	RTUuidClear(&pImage->ParentUuid);
2258	else
2259	{
2260	rc = RTUuidClear(&pImage->ParentUuid);
2261	if (RT_FAILURE(rc))
2262	return rc;
2263	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2264	VMDK_DDB_PARENT_UUID, &pImage->ParentUuid);
2265	if (RT_FAILURE(rc))
2266	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent UUID in descriptor in '%s'"), pImage->pszFilename);
2267	}
2268	}
2269	else if (RT_FAILURE(rc))
2270	return rc;
2271	/* Get parent image modification UUID. */
2272	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor,
2273	VMDK_DDB_PARENT_MODIFICATION_UUID,
2274	&pImage->ParentModificationUuid);
2275	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2276	{
2277	/* Image without UUID. Probably created by VMware and not yet used
2278	* by VirtualBox. Can only be added for images opened in read/write
2279	* mode, so don't bother producing a sensible UUID otherwise. */
2280	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2281	RTUuidClear(&pImage->ParentModificationUuid);
2282	else
2283	{
2284	RTUuidClear(&pImage->ParentModificationUuid);
2285	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2286	VMDK_DDB_PARENT_MODIFICATION_UUID,
2287	&pImage->ParentModificationUuid);
2288	if (RT_FAILURE(rc))
2289	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent modification UUID in descriptor in '%s'"), pImage->pszFilename);
2290	}
2291	}
2292	else if (RT_FAILURE(rc))
2293	return rc;
2294	return VINF_SUCCESS;
2295	}
2296	/**
2297	* Internal : Prepares the descriptor to write to the image.
2298	*/
2299	static int vmdkDescriptorPrepare(PVMDKIMAGE pImage, uint64_t cbLimit,
2300	void *ppvData, size_t pcbData)
2301	{
2302	int rc = VINF_SUCCESS;
2303	/*
2304	* Allocate temporary descriptor buffer.
2305	* In case there is no limit allocate a default
2306	* and increase if required.
2307	*/
2308	size_t cbDescriptor = cbLimit ? cbLimit : 4 * _1K;
2309	char pszDescriptor = (char )RTMemAllocZ(cbDescriptor);
2310	size_t offDescriptor = 0;
2311	if (!pszDescriptor)
2312	return VERR_NO_MEMORY;
2313	for (unsigned i = 0; i < pImage->Descriptor.cLines; i++)
2314	{
2315	const char *psz = pImage->Descriptor.aLines[i];
2316	size_t cb = strlen(psz);
2317	/*
2318	* Increase the descriptor if there is no limit and
2319	* there is not enough room left for this line.
2320	*/
2321	if (offDescriptor + cb + 1 > cbDescriptor)
2322	{
2323	if (cbLimit)
2324	{
2325	rc = vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too long in '%s'"), pImage->pszFilename);
2326	break;
2327	}
2328	else
2329	{
2330	char *pszDescriptorNew = NULL;
2331	LogFlow(("Increasing descriptor cache\n"));
2332	pszDescriptorNew = (char )RTMemRealloc(pszDescriptor, cbDescriptor + cb + 4 _1K);
2333	if (!pszDescriptorNew)
2334	{
2335	rc = VERR_NO_MEMORY;
2336	break;
2337	}
2338	pszDescriptor = pszDescriptorNew;
2339	cbDescriptor += cb + 4 * _1K;
2340	}
2341	}
2342	if (cb > 0)
2343	{
2344	memcpy(pszDescriptor + offDescriptor, psz, cb);
2345	offDescriptor += cb;
2346	}
2347	memcpy(pszDescriptor + offDescriptor, "\n", 1);
2348	offDescriptor++;
2349	}
2350	if (RT_SUCCESS(rc))
2351	{
2352	*ppvData = pszDescriptor;
2353	*pcbData = offDescriptor;
2354	}
2355	else if (pszDescriptor)
2356	RTMemFree(pszDescriptor);
2357	return rc;
2358	}
2359	/**
2360	* Internal: write/update the descriptor part of the image.
2361	*/
2362	static int vmdkWriteDescriptor(PVMDKIMAGE pImage, PVDIOCTX pIoCtx)
2363	{
2364	int rc = VINF_SUCCESS;
2365	uint64_t cbLimit;
2366	uint64_t uOffset;
2367	PVMDKFILE pDescFile;
2368	void *pvDescriptor = NULL;
2369	size_t cbDescriptor;
2370	if (pImage->pDescData)
2371	{
2372	/* Separate descriptor file. */
2373	uOffset = 0;
2374	cbLimit = 0;
2375	pDescFile = pImage->pFile;
2376	}
2377	else
2378	{
2379	/* Embedded descriptor file. */
2380	uOffset = VMDK_SECTOR2BYTE(pImage->pExtents[0].uDescriptorSector);
2381	cbLimit = VMDK_SECTOR2BYTE(pImage->pExtents[0].cDescriptorSectors);
2382	pDescFile = pImage->pExtents[0].pFile;
2383	}
2384	/* Bail out if there is no file to write to. */
2385	if (pDescFile == NULL)
2386	return VERR_INVALID_PARAMETER;
2387	rc = vmdkDescriptorPrepare(pImage, cbLimit, &pvDescriptor, &cbDescriptor);
2388	if (RT_SUCCESS(rc))
2389	{
2390	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pDescFile->pStorage,
2391	uOffset, pvDescriptor,
2392	cbLimit ? cbLimit : cbDescriptor,
2393	pIoCtx, NULL, NULL);
2394	if ( RT_FAILURE(rc)
2395	&& rc != VERR_VD_ASYNC_IO_IN_PROGRESS)
2396	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error writing descriptor in '%s'"), pImage->pszFilename);
2397	}
2398	if (RT_SUCCESS(rc) && !cbLimit)
2399	{
2400	rc = vdIfIoIntFileSetSize(pImage->pIfIo, pDescFile->pStorage, cbDescriptor);
2401	if (RT_FAILURE(rc))
2402	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error truncating descriptor in '%s'"), pImage->pszFilename);
2403	}
2404	if (RT_SUCCESS(rc))
2405	pImage->Descriptor.fDirty = false;
2406	if (pvDescriptor)
2407	RTMemFree(pvDescriptor);
2408	return rc;
2409	}
2410	/**
2411	* Internal: validate the consistency check values in a binary header.
2412	*/
2413	static int vmdkValidateHeader(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, const SparseExtentHeader *pHeader)
2414	{
2415	int rc = VINF_SUCCESS;
2416	if (RT_LE2H_U32(pHeader->magicNumber) != VMDK_SPARSE_MAGICNUMBER)
2417	{
2418	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrect magic in sparse extent header in '%s'"), pExtent->pszFullname);
2419	return rc;
2420	}
2421	if (RT_LE2H_U32(pHeader->version) != 1 && RT_LE2H_U32(pHeader->version) != 3)
2422	{
2423	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);
2424	return rc;
2425	}
2426	if ( (RT_LE2H_U32(pHeader->flags) & 1)
2427	&& ( pHeader->singleEndLineChar != '\n'
2428	\|\| pHeader->nonEndLineChar != ' '
2429	\|\| pHeader->doubleEndLineChar1 != '\r'
2430	\|\| pHeader->doubleEndLineChar2 != '\n') )
2431	{
2432	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: corrupted by CR/LF translation in '%s'"), pExtent->pszFullname);
2433	return rc;
2434	}
2435	if (RT_LE2H_U64(pHeader->descriptorSize) > VMDK_SPARSE_DESCRIPTOR_SIZE_MAX)
2436	{
2437	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor size out of bounds (%llu vs %llu) '%s'"),
2438	pExtent->pszFullname, RT_LE2H_U64(pHeader->descriptorSize), VMDK_SPARSE_DESCRIPTOR_SIZE_MAX);
2439	return rc;
2440	}
2441	return rc;
2442	}
2443	/**
2444	* Internal: read metadata belonging to an extent with binary header, i.e.
2445	* as found in monolithic files.
2446	*/
2447	static int vmdkReadBinaryMetaExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
2448	bool fMagicAlreadyRead)
2449	{
2450	SparseExtentHeader Header;
2451	int rc;
2452	if (!fMagicAlreadyRead)
2453	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage, 0,
2454	&Header, sizeof(Header));
2455	else
2456	{
2457	Header.magicNumber = RT_H2LE_U32(VMDK_SPARSE_MAGICNUMBER);
2458	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
2459	RT_UOFFSETOF(SparseExtentHeader, version),
2460	&Header.version,
2461	sizeof(Header)
2462	- RT_UOFFSETOF(SparseExtentHeader, version));
2463	}
2464	if (RT_SUCCESS(rc))
2465	{
2466	rc = vmdkValidateHeader(pImage, pExtent, &Header);
2467	if (RT_SUCCESS(rc))
2468	{
2469	uint64_t cbFile = 0;
2470	if ( (RT_LE2H_U32(Header.flags) & RT_BIT(17))
2471	&& RT_LE2H_U64(Header.gdOffset) == VMDK_GD_AT_END)
2472	pExtent->fFooter = true;
2473	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2474	\|\| ( pExtent->fFooter
2475	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2476	{
2477	rc = vdIfIoIntFileGetSize(pImage->pIfIo, pExtent->pFile->pStorage, &cbFile);
2478	if (RT_FAILURE(rc))
2479	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot get size of '%s'"), pExtent->pszFullname);
2480	}
2481	if (RT_SUCCESS(rc))
2482	{
2483	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
2484	pExtent->uAppendPosition = RT_ALIGN_64(cbFile, 512);
2485	if ( pExtent->fFooter
2486	&& ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2487	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2488	{
2489	/* Read the footer, which comes before the end-of-stream marker. */
2490	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
2491	cbFile - 2*512, &Header,
2492	sizeof(Header));
2493	if (RT_FAILURE(rc))
2494	{
2495	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading extent footer in '%s'"), pExtent->pszFullname);
2496	rc = VERR_VD_VMDK_INVALID_HEADER;
2497	}
2498	if (RT_SUCCESS(rc))
2499	rc = vmdkValidateHeader(pImage, pExtent, &Header);
2500	/* Prohibit any writes to this extent. */
2501	pExtent->uAppendPosition = 0;
2502	}
2503	if (RT_SUCCESS(rc))
2504	{
2505	pExtent->uVersion = RT_LE2H_U32(Header.version);
2506	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE; /* Just dummy value, changed later. */
2507	pExtent->cSectors = RT_LE2H_U64(Header.capacity);
2508	pExtent->cSectorsPerGrain = RT_LE2H_U64(Header.grainSize);
2509	pExtent->uDescriptorSector = RT_LE2H_U64(Header.descriptorOffset);
2510	pExtent->cDescriptorSectors = RT_LE2H_U64(Header.descriptorSize);
2511	pExtent->cGTEntries = RT_LE2H_U32(Header.numGTEsPerGT);
2512	pExtent->cOverheadSectors = RT_LE2H_U64(Header.overHead);
2513	pExtent->fUncleanShutdown = !!Header.uncleanShutdown;
2514	pExtent->uCompression = RT_LE2H_U16(Header.compressAlgorithm);
2515	if (RT_LE2H_U32(Header.flags) & RT_BIT(1))
2516	{
2517	pExtent->uSectorRGD = RT_LE2H_U64(Header.rgdOffset);
2518	pExtent->uSectorGD = RT_LE2H_U64(Header.gdOffset);
2519	}
2520	else
2521	{
2522	pExtent->uSectorGD = RT_LE2H_U64(Header.gdOffset);
2523	pExtent->uSectorRGD = 0;
2524	}
2525	if (pExtent->uDescriptorSector && !pExtent->cDescriptorSectors)
2526	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2527	N_("VMDK: inconsistent embedded descriptor config in '%s'"), pExtent->pszFullname);
2528	if ( RT_SUCCESS(rc)
2529	&& ( pExtent->uSectorGD == VMDK_GD_AT_END
2530	\|\| pExtent->uSectorRGD == VMDK_GD_AT_END)
2531	&& ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2532	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2533	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2534	N_("VMDK: cannot resolve grain directory offset in '%s'"), pExtent->pszFullname);
2535	if (RT_SUCCESS(rc))
2536	{
2537	uint64_t cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
2538	if (!cSectorsPerGDE \|\| cSectorsPerGDE > UINT32_MAX)
2539	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2540	N_("VMDK: incorrect grain directory size in '%s'"), pExtent->pszFullname);
2541	else
2542	{
2543	pExtent->cSectorsPerGDE = cSectorsPerGDE;
2544	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
2545	/* Fix up the number of descriptor sectors, as some flat images have
2546	* really just one, and this causes failures when inserting the UUID
2547	* values and other extra information. */
2548	if (pExtent->cDescriptorSectors != 0 && pExtent->cDescriptorSectors < 4)
2549	{
2550	/* Do it the easy way - just fix it for flat images which have no
2551	* other complicated metadata which needs space too. */
2552	if ( pExtent->uDescriptorSector + 4 < pExtent->cOverheadSectors
2553	&& pExtent->cGTEntries * pExtent->cGDEntries == 0)
2554	pExtent->cDescriptorSectors = 4;
2555	}
2556	}
2557	}
2558	}
2559	}
2560	}
2561	}
2562	else
2563	{
2564	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading extent header in '%s'"), pExtent->pszFullname);
2565	rc = VERR_VD_VMDK_INVALID_HEADER;
2566	}
2567	if (RT_FAILURE(rc))
2568	vmdkFreeExtentData(pImage, pExtent, false);
2569	return rc;
2570	}
2571	/**
2572	* Internal: read additional metadata belonging to an extent. For those
2573	* extents which have no additional metadata just verify the information.
2574	*/
2575	static int vmdkReadMetaExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
2576	{
2577	int rc = VINF_SUCCESS;
2578	/* disabled the check as there are too many truncated vmdk images out there */
2579	#ifdef VBOX_WITH_VMDK_STRICT_SIZE_CHECK
2580	uint64_t cbExtentSize;
2581	/* The image must be a multiple of a sector in size and contain the data
2582	* area (flat images only). If not, it means the image is at least
2583	* truncated, or even seriously garbled. */
2584	rc = vdIfIoIntFileGetSize(pImage->pIfIo, pExtent->pFile->pStorage, &cbExtentSize);
2585	if (RT_FAILURE(rc))
2586	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting size in '%s'"), pExtent->pszFullname);
2587	else if ( cbExtentSize != RT_ALIGN_64(cbExtentSize, 512)
2588	&& (pExtent->enmType != VMDKETYPE_FLAT \|\| pExtent->cNominalSectors + pExtent->uSectorOffset > VMDK_BYTE2SECTOR(cbExtentSize)))
2589	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2590	N_("VMDK: file size is not a multiple of 512 in '%s', file is truncated or otherwise garbled"), pExtent->pszFullname);
2591	#endif /* VBOX_WITH_VMDK_STRICT_SIZE_CHECK */
2592	if ( RT_SUCCESS(rc)
2593	&& pExtent->enmType == VMDKETYPE_HOSTED_SPARSE)
2594	{
2595	/* The spec says that this must be a power of two and greater than 8,
2596	* but probably they meant not less than 8. */
2597	if ( (pExtent->cSectorsPerGrain & (pExtent->cSectorsPerGrain - 1))
2598	\|\| pExtent->cSectorsPerGrain < 8)
2599	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2600	N_("VMDK: invalid extent grain size %u in '%s'"), pExtent->cSectorsPerGrain, pExtent->pszFullname);
2601	else
2602	{
2603	/* This code requires that a grain table must hold a power of two multiple
2604	* of the number of entries per GT cache entry. */
2605	if ( (pExtent->cGTEntries & (pExtent->cGTEntries - 1))
2606	\|\| pExtent->cGTEntries < VMDK_GT_CACHELINE_SIZE)
2607	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2608	N_("VMDK: grain table cache size problem in '%s'"), pExtent->pszFullname);
2609	else
2610	{
2611	rc = vmdkAllocStreamBuffers(pImage, pExtent);
2612	if (RT_SUCCESS(rc))
2613	{
2614	/* Prohibit any writes to this streamOptimized extent. */
2615	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2616	pExtent->uAppendPosition = 0;
2617	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2618	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2619	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
2620	rc = vmdkReadGrainDirectory(pImage, pExtent);
2621	else
2622	{
2623	pExtent->uGrainSectorAbs = pExtent->cOverheadSectors;
2624	pExtent->cbGrainStreamRead = 0;
2625	}
2626	}
2627	}
2628	}
2629	}
2630	if (RT_FAILURE(rc))
2631	vmdkFreeExtentData(pImage, pExtent, false);
2632	return rc;
2633	}
2634	/**
2635	* Internal: write/update the metadata for a sparse extent.
2636	*/
2637	static int vmdkWriteMetaSparseExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
2638	uint64_t uOffset, PVDIOCTX pIoCtx)
2639	{
2640	SparseExtentHeader Header;
2641	memset(&Header, '\0', sizeof(Header));
2642	Header.magicNumber = RT_H2LE_U32(VMDK_SPARSE_MAGICNUMBER);
2643	Header.version = RT_H2LE_U32(pExtent->uVersion);
2644	Header.flags = RT_H2LE_U32(RT_BIT(0));
2645	if (pExtent->pRGD)
2646	Header.flags \|= RT_H2LE_U32(RT_BIT(1));
2647	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2648	Header.flags \|= RT_H2LE_U32(RT_BIT(16) \| RT_BIT(17));
2649	Header.capacity = RT_H2LE_U64(pExtent->cSectors);
2650	Header.grainSize = RT_H2LE_U64(pExtent->cSectorsPerGrain);
2651	Header.descriptorOffset = RT_H2LE_U64(pExtent->uDescriptorSector);
2652	Header.descriptorSize = RT_H2LE_U64(pExtent->cDescriptorSectors);
2653	Header.numGTEsPerGT = RT_H2LE_U32(pExtent->cGTEntries);
2654	if (pExtent->fFooter && uOffset == 0)
2655	{
2656	if (pExtent->pRGD)
2657	{
2658	Assert(pExtent->uSectorRGD);
2659	Header.rgdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2660	Header.gdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2661	}
2662	else
2663	Header.gdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2664	}
2665	else
2666	{
2667	if (pExtent->pRGD)
2668	{
2669	Assert(pExtent->uSectorRGD);
2670	Header.rgdOffset = RT_H2LE_U64(pExtent->uSectorRGD);
2671	Header.gdOffset = RT_H2LE_U64(pExtent->uSectorGD);
2672	}
2673	else
2674	Header.gdOffset = RT_H2LE_U64(pExtent->uSectorGD);
2675	}
2676	Header.overHead = RT_H2LE_U64(pExtent->cOverheadSectors);
2677	Header.uncleanShutdown = pExtent->fUncleanShutdown;
2678	Header.singleEndLineChar = '\n';
2679	Header.nonEndLineChar = ' ';
2680	Header.doubleEndLineChar1 = '\r';
2681	Header.doubleEndLineChar2 = '\n';
2682	Header.compressAlgorithm = RT_H2LE_U16(pExtent->uCompression);
2683	int rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
2684	uOffset, &Header, sizeof(Header),
2685	pIoCtx, NULL, NULL);
2686	if (RT_FAILURE(rc) && (rc != VERR_VD_ASYNC_IO_IN_PROGRESS))
2687	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error writing extent header in '%s'"), pExtent->pszFullname);
2688	return rc;
2689	}
2690	/**
2691	* Internal: free the buffers used for streamOptimized images.
2692	*/
2693	static void vmdkFreeStreamBuffers(PVMDKEXTENT pExtent)
2694	{
2695	if (pExtent->pvCompGrain)
2696	{
2697	RTMemFree(pExtent->pvCompGrain);
2698	pExtent->pvCompGrain = NULL;
2699	}
2700	if (pExtent->pvGrain)
2701	{
2702	RTMemFree(pExtent->pvGrain);
2703	pExtent->pvGrain = NULL;
2704	}
2705	}
2706	/**
2707	* Internal: free the memory used by the extent data structure, optionally
2708	* deleting the referenced files.
2709	*
2710	* @returns VBox status code.
2711	* @param pImage Pointer to the image instance data.
2712	* @param pExtent The extent to free.
2713	* @param fDelete Flag whether to delete the backing storage.
2714	*/
2715	static int vmdkFreeExtentData(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
2716	bool fDelete)
2717	{
2718	int rc = VINF_SUCCESS;
2719	vmdkFreeGrainDirectory(pExtent);
2720	if (pExtent->pDescData)
2721	{
2722	RTMemFree(pExtent->pDescData);
2723	pExtent->pDescData = NULL;
2724	}
2725	if (pExtent->pFile != NULL)
2726	{
2727	/* Do not delete raw extents, these have full and base names equal. */
2728	rc = vmdkFileClose(pImage, &pExtent->pFile,
2729	fDelete
2730	&& pExtent->pszFullname
2731	&& pExtent->pszBasename
2732	&& strcmp(pExtent->pszFullname, pExtent->pszBasename));
2733	}
2734	if (pExtent->pszBasename)
2735	{
2736	RTMemTmpFree((void *)pExtent->pszBasename);
2737	pExtent->pszBasename = NULL;
2738	}
2739	if (pExtent->pszFullname)
2740	{
2741	RTStrFree((char )(void )pExtent->pszFullname);
2742	pExtent->pszFullname = NULL;
2743	}
2744	vmdkFreeStreamBuffers(pExtent);
2745	return rc;
2746	}
2747	/**
2748	* Internal: allocate grain table cache if necessary for this image.
2749	*/
2750	static int vmdkAllocateGrainTableCache(PVMDKIMAGE pImage)
2751	{
2752	PVMDKEXTENT pExtent;
2753	/* Allocate grain table cache if any sparse extent is present. */
2754	for (unsigned i = 0; i < pImage->cExtents; i++)
2755	{
2756	pExtent = &pImage->pExtents[i];
2757	if (pExtent->enmType == VMDKETYPE_HOSTED_SPARSE)
2758	{
2759	/* Allocate grain table cache. */
2760	pImage->pGTCache = (PVMDKGTCACHE)RTMemAllocZ(sizeof(VMDKGTCACHE));
2761	if (!pImage->pGTCache)
2762	return VERR_NO_MEMORY;
2763	for (unsigned j = 0; j < VMDK_GT_CACHE_SIZE; j++)
2764	{
2765	PVMDKGTCACHEENTRY pGCE = &pImage->pGTCache->aGTCache[j];
2766	pGCE->uExtent = UINT32_MAX;
2767	}
2768	pImage->pGTCache->cEntries = VMDK_GT_CACHE_SIZE;
2769	break;
2770	}
2771	}
2772	return VINF_SUCCESS;
2773	}
2774	/**
2775	* Internal: allocate the given number of extents.
2776	*/
2777	static int vmdkCreateExtents(PVMDKIMAGE pImage, unsigned cExtents)
2778	{
2779	int rc = VINF_SUCCESS;
2780	PVMDKEXTENT pExtents = (PVMDKEXTENT)RTMemAllocZ(cExtents * sizeof(VMDKEXTENT));
2781	if (pExtents)
2782	{
2783	for (unsigned i = 0; i < cExtents; i++)
2784	{
2785	pExtents[i].pFile = NULL;
2786	pExtents[i].pszBasename = NULL;
2787	pExtents[i].pszFullname = NULL;
2788	pExtents[i].pGD = NULL;
2789	pExtents[i].pRGD = NULL;
2790	pExtents[i].pDescData = NULL;
2791	pExtents[i].uVersion = 1;
2792	pExtents[i].uCompression = VMDK_COMPRESSION_NONE;
2793	pExtents[i].uExtent = i;
2794	pExtents[i].pImage = pImage;
2795	}
2796	pImage->pExtents = pExtents;
2797	pImage->cExtents = cExtents;
2798	}
2799	else
2800	rc = VERR_NO_MEMORY;
2801	return rc;
2802	}
2803	/**
2804	* Reads and processes the descriptor embedded in sparse images.
2805	*
2806	* @returns VBox status code.
2807	* @param pImage VMDK image instance.
2808	* @param pFile The sparse file handle.
2809	*/
2810	static int vmdkDescriptorReadSparse(PVMDKIMAGE pImage, PVMDKFILE pFile)
2811	{
2812	/* It's a hosted single-extent image. */
2813	int rc = vmdkCreateExtents(pImage, 1);
2814	if (RT_SUCCESS(rc))
2815	{
2816	/* The opened file is passed to the extent. No separate descriptor
2817	* file, so no need to keep anything open for the image. */
2818	PVMDKEXTENT pExtent = &pImage->pExtents[0];
2819	pExtent->pFile = pFile;
2820	pImage->pFile = NULL;
2821	pExtent->pszFullname = RTPathAbsDup(pImage->pszFilename);
2822	if (RT_LIKELY(pExtent->pszFullname))
2823	{
2824	/* As we're dealing with a monolithic image here, there must
2825	* be a descriptor embedded in the image file. */
2826	rc = vmdkReadBinaryMetaExtent(pImage, pExtent, true /* fMagicAlreadyRead */);
2827	if ( RT_SUCCESS(rc)
2828	&& pExtent->uDescriptorSector
2829	&& pExtent->cDescriptorSectors)
2830	{
2831	/* HACK: extend the descriptor if it is unusually small and it fits in
2832	* the unused space after the image header. Allows opening VMDK files
2833	* with extremely small descriptor in read/write mode.
2834	*
2835	* The previous version introduced a possible regression for VMDK stream
2836	* optimized images from VMware which tend to have only a single sector sized
2837	* descriptor. Increasing the descriptor size resulted in adding the various uuid
2838	* entries required to make it work with VBox but for stream optimized images
2839	* the updated binary header wasn't written to the disk creating a mismatch
2840	* between advertised and real descriptor size.
2841	*
2842	* The descriptor size will be increased even if opened readonly now if there
2843	* enough room but the new value will not be written back to the image.
2844	*/
2845	if ( pExtent->cDescriptorSectors < 3
2846	&& (int64_t)pExtent->uSectorGD - pExtent->uDescriptorSector >= 4
2847	&& (!pExtent->uSectorRGD \|\| (int64_t)pExtent->uSectorRGD - pExtent->uDescriptorSector >= 4))
2848	{
2849	uint64_t cDescriptorSectorsOld = pExtent->cDescriptorSectors;
2850	pExtent->cDescriptorSectors = 4;
2851	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
2852	{
2853	/*
2854	* Update the on disk number now to make sure we don't introduce inconsistencies
2855	* in case of stream optimized images from VMware where the descriptor is just
2856	* one sector big (the binary header is not written to disk for complete
2857	* stream optimized images in vmdkFlushImage()).
2858	*/
2859	uint64_t u64DescSizeNew = RT_H2LE_U64(pExtent->cDescriptorSectors);
2860	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pFile->pStorage,
2861	RT_UOFFSETOF(SparseExtentHeader, descriptorSize),
2862	&u64DescSizeNew, sizeof(u64DescSizeNew));
2863	if (RT_FAILURE(rc))
2864	{
2865	LogFlowFunc(("Increasing the descriptor size failed with %Rrc\n", rc));
2866	/* Restore the old size and carry on. */
2867	pExtent->cDescriptorSectors = cDescriptorSectorsOld;
2868	}
2869	}
2870	}
2871	/* Read the descriptor from the extent. */
2872	pExtent->pDescData = (char *)RTMemAllocZ(VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
2873	if (RT_LIKELY(pExtent->pDescData))
2874	{
2875	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
2876	VMDK_SECTOR2BYTE(pExtent->uDescriptorSector),
2877	pExtent->pDescData,
2878	VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
2879	if (RT_SUCCESS(rc))
2880	{
2881	rc = vmdkParseDescriptor(pImage, pExtent->pDescData,
2882	VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
2883	if ( RT_SUCCESS(rc)
2884	&& ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2885	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_ASYNC_IO)))
2886	{
2887	rc = vmdkReadMetaExtent(pImage, pExtent);
2888	if (RT_SUCCESS(rc))
2889	{
2890	/* Mark the extent as unclean if opened in read-write mode. */
2891	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2892	&& !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
2893	{
2894	pExtent->fUncleanShutdown = true;
2895	pExtent->fMetaDirty = true;
2896	}
2897	}
2898	}
2899	else if (RT_SUCCESS(rc))
2900	rc = VERR_NOT_SUPPORTED;
2901	}
2902	else
2903	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: read error for descriptor in '%s'"), pExtent->pszFullname);
2904	}
2905	else
2906	rc = VERR_NO_MEMORY;
2907	}
2908	else if (RT_SUCCESS(rc))
2909	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: monolithic image without descriptor in '%s'"), pImage->pszFilename);
2910	}
2911	else
2912	rc = VERR_NO_MEMORY;
2913	}
2914	return rc;
2915	}
2916	/**
2917	* Reads the descriptor from a pure text file.
2918	*
2919	* @returns VBox status code.
2920	* @param pImage VMDK image instance.
2921	* @param pFile The descriptor file handle.
2922	*/
2923	static int vmdkDescriptorReadAscii(PVMDKIMAGE pImage, PVMDKFILE pFile)
2924	{
2925	/* Allocate at least 10K, and make sure that there is 5K free space
2926	* in case new entries need to be added to the descriptor. Never
2927	* allocate more than 128K, because that's no valid descriptor file
2928	* and will result in the correct "truncated read" error handling. */
2929	uint64_t cbFileSize;
2930	int rc = vdIfIoIntFileGetSize(pImage->pIfIo, pFile->pStorage, &cbFileSize);
2931	if ( RT_SUCCESS(rc)
2932	&& cbFileSize >= 50)
2933	{
2934	uint64_t cbSize = cbFileSize;
2935	if (cbSize % VMDK_SECTOR2BYTE(10))
2936	cbSize += VMDK_SECTOR2BYTE(20) - cbSize % VMDK_SECTOR2BYTE(10);
2937	else
2938	cbSize += VMDK_SECTOR2BYTE(10);
2939	cbSize = RT_MIN(cbSize, _128K);
2940	pImage->cbDescAlloc = RT_MAX(VMDK_SECTOR2BYTE(20), cbSize);
2941	pImage->pDescData = (char *)RTMemAllocZ(pImage->cbDescAlloc);
2942	if (RT_LIKELY(pImage->pDescData))
2943	{
2944	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pFile->pStorage, 0, pImage->pDescData,
2945	RT_MIN(pImage->cbDescAlloc, cbFileSize));
2946	if (RT_SUCCESS(rc))
2947	{
2948	#if 0 /** @todo Revisit */
2949	cbRead += sizeof(u32Magic);
2950	if (cbRead == pImage->cbDescAlloc)
2951	{
2952	/* Likely the read is truncated. Better fail a bit too early
2953	* (normally the descriptor is much smaller than our buffer). */
2954	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: cannot read descriptor in '%s'"), pImage->pszFilename);
2955	goto out;
2956	}
2957	#endif
2958	rc = vmdkParseDescriptor(pImage, pImage->pDescData,
2959	pImage->cbDescAlloc);
2960	if (RT_SUCCESS(rc))
2961	{
2962	for (unsigned i = 0; i < pImage->cExtents && RT_SUCCESS(rc); i++)
2963	{
2964	PVMDKEXTENT pExtent = &pImage->pExtents[i];
2965	if (pExtent->pszBasename)
2966	{
2967	/* Hack to figure out whether the specified name in the
2968	* extent descriptor is absolute. Doesn't always work, but
2969	* should be good enough for now. */
2970	char *pszFullname;
2971	/** @todo implement proper path absolute check. */
2972	if (pExtent->pszBasename[0] == RTPATH_SLASH)
2973	{
2974	pszFullname = RTStrDup(pExtent->pszBasename);
2975	if (!pszFullname)
2976	{
2977	rc = VERR_NO_MEMORY;
2978	break;
2979	}
2980	}
2981	else
2982	{
2983	char *pszDirname = RTStrDup(pImage->pszFilename);
2984	if (!pszDirname)
2985	{
2986	rc = VERR_NO_MEMORY;
2987	break;
2988	}
2989	RTPathStripFilename(pszDirname);
2990	pszFullname = RTPathJoinA(pszDirname, pExtent->pszBasename);
2991	RTStrFree(pszDirname);
2992	if (!pszFullname)
2993	{
2994	rc = VERR_NO_STR_MEMORY;
2995	break;
2996	}
2997	}
2998	pExtent->pszFullname = pszFullname;
2999	}
3000	else
3001	pExtent->pszFullname = NULL;
3002	unsigned uOpenFlags = pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0);
3003	switch (pExtent->enmType)
3004	{
3005	case VMDKETYPE_HOSTED_SPARSE:
3006	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
3007	VDOpenFlagsToFileOpenFlags(uOpenFlags, false /* fCreate */));
3008	if (RT_FAILURE(rc))
3009	{
3010	/* Do NOT signal an appropriate error here, as the VD
3011	* layer has the choice of retrying the open if it
3012	* failed. */
3013	break;
3014	}
3015	rc = vmdkReadBinaryMetaExtent(pImage, pExtent,
3016	false /* fMagicAlreadyRead */);
3017	if (RT_FAILURE(rc))
3018	break;
3019	rc = vmdkReadMetaExtent(pImage, pExtent);
3020	if (RT_FAILURE(rc))
3021	break;
3022	/* Mark extent as unclean if opened in read-write mode. */
3023	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
3024	{
3025	pExtent->fUncleanShutdown = true;
3026	pExtent->fMetaDirty = true;
3027	}
3028	break;
3029	case VMDKETYPE_VMFS:
3030	case VMDKETYPE_FLAT:
3031	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
3032	VDOpenFlagsToFileOpenFlags(uOpenFlags, false /* fCreate */));
3033	if (RT_FAILURE(rc))
3034	{
3035	/* Do NOT signal an appropriate error here, as the VD
3036	* layer has the choice of retrying the open if it
3037	* failed. */
3038	break;
3039	}
3040	break;
3041	case VMDKETYPE_ZERO:
3042	/* Nothing to do. */
3043	break;
3044	default:
3045	AssertMsgFailed(("unknown vmdk extent type %d\n", pExtent->enmType));
3046	}
3047	}
3048	}
3049	}
3050	else
3051	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: read error for descriptor in '%s'"), pImage->pszFilename);
3052	}
3053	else
3054	rc = VERR_NO_MEMORY;
3055	}
3056	else if (RT_SUCCESS(rc))
3057	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor in '%s' is too short"), pImage->pszFilename);
3058	return rc;
3059	}
3060	/**
3061	* Read and process the descriptor based on the image type.
3062	*
3063	* @returns VBox status code.
3064	* @param pImage VMDK image instance.
3065	* @param pFile VMDK file handle.
3066	*/
3067	static int vmdkDescriptorRead(PVMDKIMAGE pImage, PVMDKFILE pFile)
3068	{
3069	uint32_t u32Magic;
3070	/* Read magic (if present). */
3071	int rc = vdIfIoIntFileReadSync(pImage->pIfIo, pFile->pStorage, 0,
3072	&u32Magic, sizeof(u32Magic));
3073	if (RT_SUCCESS(rc))
3074	{
3075	/* Handle the file according to its magic number. */
3076	if (RT_LE2H_U32(u32Magic) == VMDK_SPARSE_MAGICNUMBER)
3077	rc = vmdkDescriptorReadSparse(pImage, pFile);
3078	else
3079	rc = vmdkDescriptorReadAscii(pImage, pFile);
3080	}
3081	else
3082	{
3083	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading the magic number in '%s'"), pImage->pszFilename);
3084	rc = VERR_VD_VMDK_INVALID_HEADER;
3085	}
3086	return rc;
3087	}
3088	/**
3089	* Internal: Open an image, constructing all necessary data structures.
3090	*/
3091	static int vmdkOpenImage(PVMDKIMAGE pImage, unsigned uOpenFlags)
3092	{
3093	pImage->uOpenFlags = uOpenFlags;
3094	pImage->pIfError = VDIfErrorGet(pImage->pVDIfsDisk);
3095	pImage->pIfIo = VDIfIoIntGet(pImage->pVDIfsImage);
3096	AssertPtrReturn(pImage->pIfIo, VERR_INVALID_PARAMETER);
3097	/*
3098	* Open the image.
3099	* We don't have to check for asynchronous access because
3100	* we only support raw access and the opened file is a description
3101	* file were no data is stored.
3102	*/
3103	PVMDKFILE pFile;
3104	int rc = vmdkFileOpen(pImage, &pFile, NULL, pImage->pszFilename,
3105	VDOpenFlagsToFileOpenFlags(uOpenFlags, false /* fCreate */));
3106	if (RT_SUCCESS(rc))
3107	{
3108	pImage->pFile = pFile;
3109	rc = vmdkDescriptorRead(pImage, pFile);
3110	if (RT_SUCCESS(rc))
3111	{
3112	/* Determine PCHS geometry if not set. */
3113	if (pImage->PCHSGeometry.cCylinders == 0)
3114	{
3115	uint64_t cCylinders = VMDK_BYTE2SECTOR(pImage->cbSize)
3116	/ pImage->PCHSGeometry.cHeads
3117	/ pImage->PCHSGeometry.cSectors;
3118	pImage->PCHSGeometry.cCylinders = (unsigned)RT_MIN(cCylinders, 16383);
3119	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3120	&& !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
3121	{
3122	rc = vmdkDescSetPCHSGeometry(pImage, &pImage->PCHSGeometry);
3123	AssertRC(rc);
3124	}
3125	}
3126	/* Update the image metadata now in case has changed. */
3127	rc = vmdkFlushImage(pImage, NULL);
3128	if (RT_SUCCESS(rc))
3129	{
3130	/* Figure out a few per-image constants from the extents. */
3131	pImage->cbSize = 0;
3132	for (unsigned i = 0; i < pImage->cExtents; i++)
3133	{
3134	PVMDKEXTENT pExtent = &pImage->pExtents[i];
3135	if (pExtent->enmType == VMDKETYPE_HOSTED_SPARSE)
3136	{
3137	/* Here used to be a check whether the nominal size of an extent
3138	* is a multiple of the grain size. The spec says that this is
3139	* always the case, but unfortunately some files out there in the
3140	* wild violate the spec (e.g. ReactOS 0.3.1). */
3141	}
3142	else if ( pExtent->enmType == VMDKETYPE_FLAT
3143	\|\| pExtent->enmType == VMDKETYPE_ZERO)
3144	pImage->uImageFlags \|= VD_IMAGE_FLAGS_FIXED;
3145	pImage->cbSize += VMDK_SECTOR2BYTE(pExtent->cNominalSectors);
3146	}
3147	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
3148	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3149	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
3150	rc = vmdkAllocateGrainTableCache(pImage);
3151	}
3152	}
3153	}
3154	/* else: Do NOT signal an appropriate error here, as the VD layer has the
3155	* choice of retrying the open if it failed. */
3156	if (RT_SUCCESS(rc))
3157	{
3158	PVDREGIONDESC pRegion = &pImage->RegionList.aRegions[0];
3159	pImage->RegionList.fFlags = 0;
3160	pImage->RegionList.cRegions = 1;
3161	pRegion->offRegion = 0; /* Disk start. */
3162	pRegion->cbBlock = 512;
3163	pRegion->enmDataForm = VDREGIONDATAFORM_RAW;
3164	pRegion->enmMetadataForm = VDREGIONMETADATAFORM_NONE;
3165	pRegion->cbData = 512;
3166	pRegion->cbMetadata = 0;
3167	pRegion->cRegionBlocksOrBytes = pImage->cbSize;
3168	}
3169	else
3170	vmdkFreeImage(pImage, false, false /fFlush/); /* Don't try to flush anything if opening failed. */
3171	return rc;
3172	}
3173	/**
3174	* Frees a raw descriptor.
3175	* @internal
3176	*/
3177	static int vmdkRawDescFree(PVDISKRAW pRawDesc)
3178	{
3179	if (!pRawDesc)
3180	return VINF_SUCCESS;
3181	RTStrFree(pRawDesc->pszRawDisk);
3182	pRawDesc->pszRawDisk = NULL;
3183	/* Partitions: */
3184	for (unsigned i = 0; i < pRawDesc->cPartDescs; i++)
3185	{
3186	RTStrFree(pRawDesc->pPartDescs[i].pszRawDevice);
3187	pRawDesc->pPartDescs[i].pszRawDevice = NULL;
3188	RTMemFree(pRawDesc->pPartDescs[i].pvPartitionData);
3189	pRawDesc->pPartDescs[i].pvPartitionData = NULL;
3190	}
3191	RTMemFree(pRawDesc->pPartDescs);
3192	pRawDesc->pPartDescs = NULL;
3193	RTMemFree(pRawDesc);
3194	return VINF_SUCCESS;
3195	}
3196	/**
3197	* Helper that grows the raw partition descriptor table by @a cToAdd entries,
3198	* returning the pointer to the first new entry.
3199	* @internal
3200	*/
3201	static int vmdkRawDescAppendPartDesc(PVMDKIMAGE pImage, PVDISKRAW pRawDesc, uint32_t cToAdd, PVDISKRAWPARTDESC *ppRet)
3202	{
3203	uint32_t const cOld = pRawDesc->cPartDescs;
3204	uint32_t const cNew = cOld + cToAdd;
3205	PVDISKRAWPARTDESC paNew = (PVDISKRAWPARTDESC)RTMemReallocZ(pRawDesc->pPartDescs,
3206	cOld * sizeof(pRawDesc->pPartDescs[0]),
3207	cNew * sizeof(pRawDesc->pPartDescs[0]));
3208	if (paNew)
3209	{
3210	pRawDesc->cPartDescs = cNew;
3211	pRawDesc->pPartDescs = paNew;
3212	*ppRet = &paNew[cOld];
3213	return VINF_SUCCESS;
3214	}
3215	*ppRet = NULL;
3216	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS,
3217	N_("VMDK: Image path: '%s'. Out of memory growing the partition descriptors (%u -> %u)."),
3218	pImage->pszFilename, cOld, cNew);
3219	}
3220	/**
3221	* @callback_method_impl{FNRTSORTCMP}
3222	*/
3223	static DECLCALLBACK(int) vmdkRawDescPartComp(void const pvElement1, void const pvElement2, void *pvUser)
3224	{
3225	RT_NOREF(pvUser);
3226	int64_t const iDelta = ((PVDISKRAWPARTDESC)pvElement1)->offStartInVDisk - ((PVDISKRAWPARTDESC)pvElement2)->offStartInVDisk;
3227	return iDelta < 0 ? -1 : iDelta > 0 ? 1 : 0;
3228	}
3229	/**
3230	* Post processes the partition descriptors.
3231	*
3232	* Sorts them and check that they don't overlap.
3233	*/
3234	static int vmdkRawDescPostProcessPartitions(PVMDKIMAGE pImage, PVDISKRAW pRawDesc, uint64_t cbSize)
3235	{
3236	/*
3237	* Sort data areas in ascending order of start.
3238	*/
3239	RTSortShell(pRawDesc->pPartDescs, pRawDesc->cPartDescs, sizeof(pRawDesc->pPartDescs[0]), vmdkRawDescPartComp, NULL);
3240	/*
3241	* Check that we don't have overlapping descriptors. If we do, that's an
3242	* indication that the drive is corrupt or that the RTDvm code is buggy.
3243	*/
3244	VDISKRAWPARTDESC const *paPartDescs = pRawDesc->pPartDescs;
3245	for (uint32_t i = 0; i < pRawDesc->cPartDescs; i++)
3246	{
3247	uint64_t offLast = paPartDescs[i].offStartInVDisk + paPartDescs[i].cbData;
3248	if (offLast <= paPartDescs[i].offStartInVDisk)
3249	return vdIfError(pImage->pIfError, VERR_FILESYSTEM_CORRUPT /?/, RT_SRC_POS,
3250	N_("VMDK: Image path: '%s'. Bogus partition descriptor #%u (%#RX64 LB %#RX64%s): Wrap around or zero"),
3251	pImage->pszFilename, i, paPartDescs[i].offStartInVDisk, paPartDescs[i].cbData,
3252	paPartDescs[i].pvPartitionData ? " (data)" : "");
3253	offLast -= 1;
3254	if (i + 1 < pRawDesc->cPartDescs && offLast >= paPartDescs[i + 1].offStartInVDisk)
3255	return vdIfError(pImage->pIfError, VERR_FILESYSTEM_CORRUPT /?/, RT_SRC_POS,
3256	N_("VMDK: Image path: '%s'. Partition descriptor #%u (%#RX64 LB %#RX64%s) overlaps with the next (%#RX64 LB %#RX64%s)"),
3257	pImage->pszFilename, i, paPartDescs[i].offStartInVDisk, paPartDescs[i].cbData,
3258	paPartDescs[i].pvPartitionData ? " (data)" : "", paPartDescs[i + 1].offStartInVDisk,
3259	paPartDescs[i + 1].cbData, paPartDescs[i + 1].pvPartitionData ? " (data)" : "");
3260	if (offLast >= cbSize)
3261	return vdIfError(pImage->pIfError, VERR_FILESYSTEM_CORRUPT /?/, RT_SRC_POS,
3262	N_("VMDK: Image path: '%s'. Partition descriptor #%u (%#RX64 LB %#RX64%s) goes beyond the end of the drive (%#RX64)"),
3263	pImage->pszFilename, i, paPartDescs[i].offStartInVDisk, paPartDescs[i].cbData,
3264	paPartDescs[i].pvPartitionData ? " (data)" : "", cbSize);
3265	}
3266	return VINF_SUCCESS;
3267	}
3268	#ifdef RT_OS_LINUX
3269	/**
3270	* Searches the dir specified in @a pszBlockDevDir for subdirectories with a
3271	* 'dev' file matching @a uDevToLocate.
3272	*
3273	* This is used both
3274	*
3275	* @returns IPRT status code, errors have been reported properly.
3276	* @param pImage For error reporting.
3277	* @param pszBlockDevDir Input: Path to the directory search under.
3278	* Output: Path to the directory containing information
3279	* for @a uDevToLocate.
3280	* @param cbBlockDevDir The size of the buffer @a pszBlockDevDir points to.
3281	* @param uDevToLocate The device number of the block device info dir to
3282	* locate.
3283	* @param pszDevToLocate For error reporting.
3284	*/
3285	static int vmdkFindSysBlockDevPath(PVMDKIMAGE pImage, char *pszBlockDevDir, size_t cbBlockDevDir,
3286	dev_t uDevToLocate, const char *pszDevToLocate)
3287	{
3288	size_t const cchDir = RTPathEnsureTrailingSeparator(pszBlockDevDir, cbBlockDevDir);
3289	AssertReturn(cchDir > 0, VERR_BUFFER_OVERFLOW);
3290	RTDIR hDir = NIL_RTDIR;
3291	int rc = RTDirOpen(&hDir, pszBlockDevDir);
3292	if (RT_SUCCESS(rc))
3293	{
3294	for (;;)
3295	{
3296	RTDIRENTRY Entry;
3297	rc = RTDirRead(hDir, &Entry, NULL);
3298	if (RT_SUCCESS(rc))
3299	{
3300	/* We're interested in directories and symlinks. */
3301	if ( Entry.enmType == RTDIRENTRYTYPE_DIRECTORY
3302	\|\| Entry.enmType == RTDIRENTRYTYPE_SYMLINK
3303	\|\| Entry.enmType == RTDIRENTRYTYPE_UNKNOWN)
3304	{
3305	rc = RTStrCopy(&pszBlockDevDir[cchDir], cbBlockDevDir - cchDir, Entry.szName);
3306	AssertContinue(RT_SUCCESS(rc)); /* should not happen! */
3307	dev_t uThisDevNo = ~uDevToLocate;
3308	rc = RTLinuxSysFsReadDevNumFile(&uThisDevNo, "%s/dev", pszBlockDevDir);
3309	if (RT_SUCCESS(rc) && uThisDevNo == uDevToLocate)
3310	break;
3311	}
3312	}
3313	else
3314	{
3315	pszBlockDevDir[cchDir] = '\0';
3316	if (rc == VERR_NO_MORE_FILES)
3317	rc = vdIfError(pImage->pIfError, VERR_NOT_FOUND, RT_SRC_POS,
3318	N_("VMDK: Image path: '%s'. Failed to locate device corresponding to '%s' under '%s'"),
3319	pImage->pszFilename, pszDevToLocate, pszBlockDevDir);
3320	else
3321	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
3322	N_("VMDK: Image path: '%s'. RTDirRead failed enumerating '%s': %Rrc"),
3323	pImage->pszFilename, pszBlockDevDir, rc);
3324	break;
3325	}
3326	}
3327	RTDirClose(hDir);
3328	}
3329	else
3330	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
3331	N_("VMDK: Image path: '%s'. Failed to open dir '%s' for listing: %Rrc"),
3332	pImage->pszFilename, pszBlockDevDir, rc);
3333	return rc;
3334	}
3335	#endif /* RT_OS_LINUX */
3336	#ifdef RT_OS_FREEBSD
3337	/**
3338	* Reads the config data from the provider and returns offset and size
3339	*
3340	* @return IPRT status code
3341	* @param pProvider GEOM provider representing partition
3342	* @param pcbOffset Placeholder for the offset of the partition
3343	* @param pcbSize Placeholder for the size of the partition
3344	*/
3345	static int vmdkReadPartitionsParamsFromProvider(gprovider pProvider, uint64_t pcbOffset, uint64_t *pcbSize)
3346	{
3347	gconfig *pConfEntry;
3348	int rc = VERR_NOT_FOUND;
3349	/*
3350	* Required parameters are located in the list containing key/value pairs.
3351	* Both key and value are in text form. Manuals tells nothing about the fact
3352	* that the both parameters should be present in the list. Thus, there are
3353	* cases when only one parameter is presented. To handle such cases we treat
3354	* absent params as zero allowing the caller decide the case is either correct
3355	* or an error.
3356	*/
3357	uint64_t cbOffset = 0;
3358	uint64_t cbSize = 0;
3359	LIST_FOREACH(pConfEntry, &pProvider->lg_config, lg_config)
3360	{
3361	if (RTStrCmp(pConfEntry->lg_name, "offset") == 0)
3362	{
3363	cbOffset = RTStrToUInt64(pConfEntry->lg_val);
3364	rc = VINF_SUCCESS;
3365	}
3366	else if (RTStrCmp(pConfEntry->lg_name, "length") == 0)
3367	{
3368	cbSize = RTStrToUInt64(pConfEntry->lg_val);
3369	rc = VINF_SUCCESS;
3370	}
3371	}
3372	if (RT_SUCCESS(rc))
3373	{
3374	*pcbOffset = cbOffset;
3375	*pcbSize = cbSize;
3376	}
3377	return rc;
3378	}
3379	/**
3380	* Searches the partition specified by name and calculates its size and absolute offset.
3381	*
3382	* @return IPRT status code.
3383	* @param pParentClass Class containing pParentGeom
3384	* @param pszParentGeomName Name of the parent geom where we are looking for provider
3385	* @param pszProviderName Name of the provider we are looking for
3386	* @param pcbAbsoluteOffset Placeholder for the absolute offset of the partition, i.e. offset from the beginning of the disk
3387	* @param psbSize Placeholder for the size of the partition.
3388	*/
3389	static int vmdkFindPartitionParamsByName(gclass pParentClass, const char pszParentGeomName, const char *pszProviderName,
3390	uint64_t pcbAbsoluteOffset, uint64_t pcbSize)
3391	{
3392	AssertReturn(pParentClass, VERR_INVALID_PARAMETER);
3393	AssertReturn(pszParentGeomName, VERR_INVALID_PARAMETER);
3394	AssertReturn(pszProviderName, VERR_INVALID_PARAMETER);
3395	AssertReturn(pcbAbsoluteOffset, VERR_INVALID_PARAMETER);
3396	AssertReturn(pcbSize, VERR_INVALID_PARAMETER);
3397	ggeom *pParentGeom;
3398	int rc = VERR_NOT_FOUND;
3399	LIST_FOREACH(pParentGeom, &pParentClass->lg_geom, lg_geom)
3400	{
3401	if (RTStrCmp(pParentGeom->lg_name, pszParentGeomName) == 0)
3402	{
3403	rc = VINF_SUCCESS;
3404	break;
3405	}
3406	}
3407	if (RT_FAILURE(rc))
3408	return rc;
3409	gprovider *pProvider;
3410	/*
3411	* First, go over providers without handling EBR or BSDLabel
3412	* partitions for case when looking provider is child
3413	* of the givng geom, to reduce searching time
3414	*/
3415	LIST_FOREACH(pProvider, &pParentGeom->lg_provider, lg_provider)
3416	{
3417	if (RTStrCmp(pProvider->lg_name, pszProviderName) == 0)
3418	return vmdkReadPartitionsParamsFromProvider(pProvider, pcbAbsoluteOffset, pcbSize);
3419	}
3420	/*
3421	* No provider found. Go over the parent geom again
3422	* and make recursions if geom represents EBR or BSDLabel.
3423	* In this case given parent geom contains only EBR or BSDLabel
3424	* partition itself and their own partitions are in the separate
3425	* geoms. Also, partition offsets are relative to geom, so
3426	* we have to add offset from child provider with parent geoms
3427	* provider
3428	*/
3429	LIST_FOREACH(pProvider, &pParentGeom->lg_provider, lg_provider)
3430	{
3431	uint64_t cbOffset = 0;
3432	uint64_t cbSize = 0;
3433	rc = vmdkReadPartitionsParamsFromProvider(pProvider, &cbOffset, &cbSize);
3434	if (RT_FAILURE(rc))
3435	return rc;
3436	uint64_t cbProviderOffset = 0;
3437	uint64_t cbProviderSize = 0;
3438	rc = vmdkFindPartitionParamsByName(pParentClass, pProvider->lg_name, pszProviderName, &cbProviderOffset, &cbProviderSize);
3439	if (RT_SUCCESS(rc))
3440	{
3441	*pcbAbsoluteOffset = cbOffset + cbProviderOffset;
3442	*pcbSize = cbProviderSize;
3443	return rc;
3444	}
3445	}
3446	return VERR_NOT_FOUND;
3447	}
3448	#endif
3449	/**
3450	* Attempts to verify the raw partition path.
3451	*
3452	* We don't want to trust RTDvm and the partition device node morphing blindly.
3453	*/
3454	static int vmdkRawDescVerifyPartitionPath(PVMDKIMAGE pImage, PVDISKRAWPARTDESC pPartDesc, uint32_t idxPartition,
3455	const char *pszRawDrive, RTFILE hRawDrive, uint32_t cbSector, RTDVMVOLUME hVol)
3456	{
3457	RT_NOREF(pImage, pPartDesc, idxPartition, pszRawDrive, hRawDrive, cbSector, hVol);
3458	/*
3459	* Try open the raw partition device.
3460	*/
3461	RTFILE hRawPart = NIL_RTFILE;
3462	int rc = RTFileOpen(&hRawPart, pPartDesc->pszRawDevice, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE);
3463	if (RT_FAILURE(rc))
3464	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
3465	N_("VMDK: Image path: '%s'. Failed to open partition #%u on '%s' via '%s' (%Rrc)"),
3466	pImage->pszFilename, idxPartition, pszRawDrive, pPartDesc->pszRawDevice, rc);
3467	/*
3468	* Compare the partition UUID if we can get it.
3469	*/
3470	#ifdef RT_OS_WINDOWS
3471	DWORD cbReturned;
3472	/* 1. Get the device numbers for both handles, they should have the same disk. */
3473	STORAGE_DEVICE_NUMBER DevNum1;
3474	RT_ZERO(DevNum1);
3475	if (!DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_STORAGE_GET_DEVICE_NUMBER,
3476	NULL /pvInBuffer/, 0 /cbInBuffer/, &DevNum1, sizeof(DevNum1), &cbReturned, NULL /pOverlapped/))
3477	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
3478	N_("VMDK: Image path: '%s'. IOCTL_STORAGE_GET_DEVICE_NUMBER failed on '%s': %u"),
3479	pImage->pszFilename, pszRawDrive, GetLastError());
3480	STORAGE_DEVICE_NUMBER DevNum2;
3481	RT_ZERO(DevNum2);
3482	if (!DeviceIoControl((HANDLE)RTFileToNative(hRawPart), IOCTL_STORAGE_GET_DEVICE_NUMBER,
3483	NULL /pvInBuffer/, 0 /cbInBuffer/, &DevNum2, sizeof(DevNum2), &cbReturned, NULL /pOverlapped/))
3484	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
3485	N_("VMDK: Image path: '%s'. IOCTL_STORAGE_GET_DEVICE_NUMBER failed on '%s': %u"),
3486	pImage->pszFilename, pPartDesc->pszRawDevice, GetLastError());
3487	if ( RT_SUCCESS(rc)
3488	&& ( DevNum1.DeviceNumber != DevNum2.DeviceNumber
3489	\|\| DevNum1.DeviceType != DevNum2.DeviceType))
3490	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3491	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s' (%#x != %#x \|\| %#x != %#x)"),
3492	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
3493	DevNum1.DeviceNumber, DevNum2.DeviceNumber, DevNum1.DeviceType, DevNum2.DeviceType);
3494	if (RT_SUCCESS(rc))
3495	{
3496	/* Get the partitions from the raw drive and match up with the volume info
3497	from RTDvm. The partition number is found in DevNum2. */
3498	DWORD cbNeeded = 0;
3499	if ( DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_DISK_GET_DRIVE_LAYOUT_EX,
3500	NULL /pvInBuffer/, 0 /cbInBuffer/, NULL, 0, &cbNeeded, NULL /pOverlapped/)
3501	\|\| cbNeeded < RT_UOFFSETOF_DYN(DRIVE_LAYOUT_INFORMATION_EX, PartitionEntry[1]))
3502	cbNeeded = RT_UOFFSETOF_DYN(DRIVE_LAYOUT_INFORMATION_EX, PartitionEntry[64]);
3503	cbNeeded += sizeof(PARTITION_INFORMATION_EX) * 2; /* just in case */
3504	DRIVE_LAYOUT_INFORMATION_EX pLayout = (DRIVE_LAYOUT_INFORMATION_EX )RTMemTmpAllocZ(cbNeeded);
3505	if (pLayout)
3506	{
3507	cbReturned = 0;
3508	if (DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_DISK_GET_DRIVE_LAYOUT_EX,
3509	NULL /pvInBuffer/, 0 /cbInBuffer/, pLayout, cbNeeded, &cbReturned, NULL /pOverlapped/))
3510	{
3511	/* Find the entry with the given partition number (it's not an index, array contains empty MBR entries ++). */
3512	unsigned iEntry = 0;
3513	while ( iEntry < pLayout->PartitionCount
3514	&& pLayout->PartitionEntry[iEntry].PartitionNumber != DevNum2.PartitionNumber)
3515	iEntry++;
3516	if (iEntry < pLayout->PartitionCount)
3517	{
3518	/* Compare the basics */
3519	PARTITION_INFORMATION_EX const * const pLayoutEntry = &pLayout->PartitionEntry[iEntry];
3520	if (pLayoutEntry->StartingOffset.QuadPart != (int64_t)pPartDesc->offStartInVDisk)
3521	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3522	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': StartingOffset %RU64, expected %RU64"),
3523	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
3524	pLayoutEntry->StartingOffset.QuadPart, pPartDesc->offStartInVDisk);
3525	else if (pLayoutEntry->PartitionLength.QuadPart != (int64_t)pPartDesc->cbData)
3526	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3527	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': PartitionLength %RU64, expected %RU64"),
3528	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
3529	pLayoutEntry->PartitionLength.QuadPart, pPartDesc->cbData);
3530	/** @todo We could compare the MBR type, GPT type and ID. */
3531	RT_NOREF(hVol);
3532	}
3533	else
3534	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3535	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': PartitionCount (%#x vs %#x)"),
3536	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
3537	DevNum2.PartitionNumber, pLayout->PartitionCount);
3538	# ifndef LOG_ENABLED
3539	if (RT_FAILURE(rc))
3540	# endif
3541	{
3542	LogRel(("VMDK: Windows reports %u partitions for '%s':\n", pLayout->PartitionCount, pszRawDrive));
3543	PARTITION_INFORMATION_EX const *pEntry = &pLayout->PartitionEntry[0];
3544	for (DWORD i = 0; i < pLayout->PartitionCount; i++, pEntry++)
3545	{
3546	LogRel(("VMDK: #%u/%u: %016RU64 LB %016RU64 style=%d rewrite=%d",
3547	i, pEntry->PartitionNumber, pEntry->StartingOffset.QuadPart, pEntry->PartitionLength.QuadPart,
3548	pEntry->PartitionStyle, pEntry->RewritePartition));
3549	if (pEntry->PartitionStyle == PARTITION_STYLE_MBR)
3550	LogRel((" type=%#x boot=%d rec=%d hidden=%u\n", pEntry->Mbr.PartitionType, pEntry->Mbr.BootIndicator,
3551	pEntry->Mbr.RecognizedPartition, pEntry->Mbr.HiddenSectors));
3552	else if (pEntry->PartitionStyle == PARTITION_STYLE_GPT)
3553	LogRel((" type=%RTuuid id=%RTuuid aatrib=%RX64 name=%.36ls\n", &pEntry->Gpt.PartitionType,
3554	&pEntry->Gpt.PartitionId, pEntry->Gpt.Attributes, &pEntry->Gpt.Name[0]));
3555	else
3556	LogRel(("\n"));
3557	}
3558	LogRel(("VMDK: Looked for partition #%u (%u, '%s') at %RU64 LB %RU64\n", DevNum2.PartitionNumber,
3559	idxPartition, pPartDesc->pszRawDevice, pPartDesc->offStartInVDisk, pPartDesc->cbData));
3560	}
3561	}
3562	else
3563	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
3564	N_("VMDK: Image path: '%s'. IOCTL_DISK_GET_DRIVE_LAYOUT_EX failed on '%s': %u (cb %u, cbRet %u)"),
3565	pImage->pszFilename, pPartDesc->pszRawDevice, GetLastError(), cbNeeded, cbReturned);
3566	RTMemTmpFree(pLayout);
3567	}
3568	else
3569	rc = VERR_NO_TMP_MEMORY;
3570	}
3571	#elif defined(RT_OS_LINUX)
3572	RT_NOREF(hVol);
3573	/* Stat the two devices first to get their device numbers. (We probably
3574	could make some assumptions here about the major & minor number assignments
3575	for legacy nodes, but it doesn't hold up for nvme, so we'll skip that.) */
3576	struct stat StDrive, StPart;
3577	if (fstat((int)RTFileToNative(hRawDrive), &StDrive) != 0)
3578	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
3579	N_("VMDK: Image path: '%s'. fstat failed on '%s': %d"), pImage->pszFilename, pszRawDrive, errno);
3580	else if (fstat((int)RTFileToNative(hRawPart), &StPart) != 0)
3581	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
3582	N_("VMDK: Image path: '%s'. fstat failed on '%s': %d"), pImage->pszFilename, pPartDesc->pszRawDevice, errno);
3583	else
3584	{
3585	/* Scan the directories immediately under /sys/block/ for one with a
3586	'dev' file matching the drive's device number: */
3587	char szSysPath[RTPATH_MAX];
3588	rc = RTLinuxConstructPath(szSysPath, sizeof(szSysPath), "block/");
3589	AssertRCReturn(rc, rc); /* this shall not fail */
3590	if (RTDirExists(szSysPath))
3591	{
3592	rc = vmdkFindSysBlockDevPath(pImage, szSysPath, sizeof(szSysPath), StDrive.st_rdev, pszRawDrive);
3593	/* Now, scan the directories under that again for a partition device
3594	matching the hRawPart device's number: */
3595	if (RT_SUCCESS(rc))
3596	rc = vmdkFindSysBlockDevPath(pImage, szSysPath, sizeof(szSysPath), StPart.st_rdev, pPartDesc->pszRawDevice);
3597	/* Having found the /sys/block/device/partition/ path, we can finally
3598	read the partition attributes and compare with hVol. */
3599	if (RT_SUCCESS(rc))
3600	{
3601	/* partition number: */
3602	int64_t iLnxPartition = 0;
3603	rc = RTLinuxSysFsReadIntFile(10, &iLnxPartition, "%s/partition", szSysPath);
3604	if (RT_SUCCESS(rc) && iLnxPartition != idxPartition)
3605	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3606	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Partition number %RI64, expected %RU32"),
3607	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, iLnxPartition, idxPartition);
3608	/* else: ignore failure? */
3609	/* start offset: */
3610	uint32_t const cbLnxSector = 512; /* It's hardcoded in the Linux kernel */
3611	if (RT_SUCCESS(rc))
3612	{
3613	int64_t offLnxStart = -1;
3614	rc = RTLinuxSysFsReadIntFile(10, &offLnxStart, "%s/start", szSysPath);
3615	offLnxStart *= cbLnxSector;
3616	if (RT_SUCCESS(rc) && offLnxStart != (int64_t)pPartDesc->offStartInVDisk)
3617	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3618	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RI64, expected %RU64"),
3619	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, offLnxStart, pPartDesc->offStartInVDisk);
3620	/* else: ignore failure? */
3621	}
3622	/* the size: */
3623	if (RT_SUCCESS(rc))
3624	{
3625	int64_t cbLnxData = -1;
3626	rc = RTLinuxSysFsReadIntFile(10, &cbLnxData, "%s/size", szSysPath);
3627	cbLnxData *= cbLnxSector;
3628	if (RT_SUCCESS(rc) && cbLnxData != (int64_t)pPartDesc->cbData)
3629	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3630	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RI64, expected %RU64"),
3631	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbLnxData, pPartDesc->cbData);
3632	/* else: ignore failure? */
3633	}
3634	}
3635	}
3636	/* else: We've got nothing to work on, so only do content comparison. */
3637	}
3638	#elif defined(RT_OS_FREEBSD)
3639	char szDriveDevName[256];
3640	char* pszDevName = fdevname_r(RTFileToNative(hRawDrive), szDriveDevName, 256);
3641	if (pszDevName == NULL)
3642	rc = vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
3643	N_("VMDK: Image path: '%s'. '%s' is not a drive path"), pImage->pszFilename, pszRawDrive);
3644	char szPartDevName[256];
3645	if (RT_SUCCESS(rc))
3646	{
3647	pszDevName = fdevname_r(RTFileToNative(hRawPart), szPartDevName, 256);
3648	if (pszDevName == NULL)
3649	rc = vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
3650	N_("VMDK: Image path: '%s'. '%s' is not a partition path"), pImage->pszFilename, pPartDesc->pszRawDevice);
3651	}
3652	if (RT_SUCCESS(rc))
3653	{
3654	gmesh geomMesh;
3655	int err = geom_gettree(&geomMesh);
3656	if (err == 0)
3657	{
3658	/* Find root class containg partitions info */
3659	gclass* pPartClass;
3660	LIST_FOREACH(pPartClass, &geomMesh.lg_class, lg_class)
3661	{
3662	if (RTStrCmp(pPartClass->lg_name, "PART") == 0)
3663	break;
3664	}
3665	if (pPartClass == NULL \|\| RTStrCmp(pPartClass->lg_name, "PART") != 0)
3666	rc = vdIfError(pImage->pIfError, VERR_GENERAL_FAILURE, RT_SRC_POS,
3667	N_("VMDK: Image path: '%s'. 'PART' class not found in the GEOM tree"), pImage->pszFilename);
3668	if (RT_SUCCESS(rc))
3669	{
3670	/* Find provider representing partition device */
3671	uint64_t cbOffset;
3672	uint64_t cbSize;
3673	rc = vmdkFindPartitionParamsByName(pPartClass, szDriveDevName, szPartDevName, &cbOffset, &cbSize);
3674	if (RT_SUCCESS(rc))
3675	{
3676	if (cbOffset != pPartDesc->offStartInVDisk)
3677	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3678	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RU64, expected %RU64"),
3679	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbOffset, pPartDesc->offStartInVDisk);
3680	if (cbSize != pPartDesc->cbData)
3681	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3682	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RU64, expected %RU64"),
3683	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbSize, pPartDesc->cbData);
3684	}
3685	else
3686	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
3687	N_("VMDK: Image path: '%s'. Error getting geom provider for the partition '%s' of the drive '%s' in the GEOM tree: %Rrc"),
3688	pImage->pszFilename, pPartDesc->pszRawDevice, pszRawDrive, rc);
3689	}
3690	geom_deletetree(&geomMesh);
3691	}
3692	else
3693	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(err), RT_SRC_POS,
3694	N_("VMDK: Image path: '%s'. geom_gettree failed: %d"), pImage->pszFilename, err);
3695	}
3696	#elif defined(RT_OS_SOLARIS)
3697	RT_NOREF(hVol);
3698	dk_cinfo dkiDriveInfo;
3699	dk_cinfo dkiPartInfo;
3700	if (ioctl(RTFileToNative(hRawDrive), DKIOCINFO, (caddr_t)&dkiDriveInfo) == -1)
3701	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
3702	N_("VMDK: Image path: '%s'. DKIOCINFO failed on '%s': %d"), pImage->pszFilename, pszRawDrive, errno);
3703	else if (ioctl(RTFileToNative(hRawPart), DKIOCINFO, (caddr_t)&dkiPartInfo) == -1)
3704	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
3705	N_("VMDK: Image path: '%s'. DKIOCINFO failed on '%s': %d"), pImage->pszFilename, pszRawDrive, errno);
3706	else if ( dkiDriveInfo.dki_ctype != dkiPartInfo.dki_ctype
3707	\|\| dkiDriveInfo.dki_cnum != dkiPartInfo.dki_cnum
3708	\|\| dkiDriveInfo.dki_addr != dkiPartInfo.dki_addr
3709	\|\| dkiDriveInfo.dki_unit != dkiPartInfo.dki_unit
3710	\|\| dkiDriveInfo.dki_slave != dkiPartInfo.dki_slave)
3711	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3712	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s' (%#x != %#x \|\| %#x != %#x \|\| %#x != %#x \|\| %#x != %#x \|\| %#x != %#x)"),
3713	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
3714	dkiDriveInfo.dki_ctype, dkiPartInfo.dki_ctype, dkiDriveInfo.dki_cnum, dkiPartInfo.dki_cnum,
3715	dkiDriveInfo.dki_addr, dkiPartInfo.dki_addr, dkiDriveInfo.dki_unit, dkiPartInfo.dki_unit,
3716	dkiDriveInfo.dki_slave, dkiPartInfo.dki_slave);
3717	else
3718	{
3719	uint64_t cbOffset = 0;
3720	uint64_t cbSize = 0;
3721	dk_gpt *pEfi = NULL;
3722	int idxEfiPart = efi_alloc_and_read(RTFileToNative(hRawPart), &pEfi);
3723	if (idxEfiPart >= 0)
3724	{
3725	if ((uint32_t)dkiPartInfo.dki_partition + 1 == idxPartition)
3726	{
3727	cbOffset = pEfi->efi_parts[idxEfiPart].p_start * pEfi->efi_lbasize;
3728	cbSize = pEfi->efi_parts[idxEfiPart].p_size * pEfi->efi_lbasize;
3729	}
3730	else
3731	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3732	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s' (%#x != %#x)"),
3733	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
3734	idxPartition, (uint32_t)dkiPartInfo.dki_partition + 1);
3735	efi_free(pEfi);
3736	}
3737	else
3738	{
3739	/*
3740	* Manual says the efi_alloc_and_read returns VT_EINVAL if no EFI partition table found.
3741	* Actually, the function returns any error, e.g. VT_ERROR. Thus, we are not sure, is it
3742	* real error or just no EFI table found. Therefore, let's try to obtain partition info
3743	* using another way. If there is an error, it returns errno which will be handled below.
3744	*/
3745	uint32_t numPartition = (uint32_t)dkiPartInfo.dki_partition;
3746	if (numPartition > NDKMAP)
3747	numPartition -= NDKMAP;
3748	if (numPartition != idxPartition)
3749	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3750	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s' (%#x != %#x)"),
3751	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
3752	idxPartition, numPartition);
3753	else
3754	{
3755	dk_minfo_ext mediaInfo;
3756	if (ioctl(RTFileToNative(hRawPart), DKIOCGMEDIAINFOEXT, (caddr_t)&mediaInfo) == -1)
3757	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
3758	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s'. Can not obtain partition info: %d"),
3759	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
3760	else
3761	{
3762	extpart_info extPartInfo;
3763	if (ioctl(RTFileToNative(hRawPart), DKIOCEXTPARTINFO, (caddr_t)&extPartInfo) != -1)
3764	{
3765	cbOffset = (uint64_t)extPartInfo.p_start * mediaInfo.dki_lbsize;
3766	cbSize = (uint64_t)extPartInfo.p_length * mediaInfo.dki_lbsize;
3767	}
3768	else
3769	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
3770	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s'. Can not obtain partition info: %d"),
3771	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
3772	}
3773	}
3774	}
3775	if (RT_SUCCESS(rc) && cbOffset != pPartDesc->offStartInVDisk)
3776	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3777	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RI64, expected %RU64"),
3778	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbOffset, pPartDesc->offStartInVDisk);
3779	if (RT_SUCCESS(rc) && cbSize != pPartDesc->cbData)
3780	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3781	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RI64, expected %RU64"),
3782	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbSize, pPartDesc->cbData);
3783	}
3784
3785	#elif defined(RT_OS_DARWIN)
3786	/* Stat the drive get its device number. */
3787	struct stat StDrive;
3788	if (fstat((int)RTFileToNative(hRawDrive), &StDrive) != 0)
3789	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
3790	N_("VMDK: Image path: '%s'. fstat failed on '%s' (errno=%d)"), pImage->pszFilename, pszRawDrive, errno);
3791	else
3792	{
3793	if (ioctl(RTFileToNative(hRawPart), DKIOCLOCKPHYSICALEXTENTS, NULL) == -1)
3794	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
3795	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to lock the partition (errno=%d)"),
3796	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
3797	else
3798	{
3799	uint32_t cbBlockSize = 0;
3800	uint64_t cbOffset = 0;
3801	uint64_t cbSize = 0;
3802	if (ioctl(RTFileToNative(hRawPart), DKIOCGETBLOCKSIZE, (caddr_t)&cbBlockSize) == -1)
3803	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
3804	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain the sector size of the partition (errno=%d)"),
3805	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
3806	else if (ioctl(RTFileToNative(hRawPart), DKIOCGETBASE, (caddr_t)&cbOffset) == -1)
3807	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
3808	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain the start offset of the partition (errno=%d)"),
3809	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
3810	else if (ioctl(RTFileToNative(hRawPart), DKIOCGETBLOCKCOUNT, (caddr_t)&cbSize) == -1)
3811	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
3812	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain the size of the partition (errno=%d)"),
3813	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
3814	else
3815	{
3816	cbSize *= (uint64_t)cbBlockSize;
3817	dk_physical_extent_t dkPartExtent = {0};
3818	dkPartExtent.offset = 0;
3819	dkPartExtent.length = cbSize;
3820	if (ioctl(RTFileToNative(hRawPart), DKIOCGETPHYSICALEXTENT, (caddr_t)&dkPartExtent) == -1)
3821	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
3822	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain partition info (errno=%d)"),
3823	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
3824	else
3825	{
3826	if (dkPartExtent.dev != StDrive.st_rdev)
3827	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3828	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Drive does not contain the partition"),
3829	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive);
3830	else if (cbOffset != pPartDesc->offStartInVDisk)
3831	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3832	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RU64, expected %RU64"),
3833	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbOffset, pPartDesc->offStartInVDisk);
3834	else if (cbSize != pPartDesc->cbData)
3835	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3836	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RU64, expected %RU64"),
3837	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbSize, pPartDesc->cbData);
3838	}
3839	}
3840
3841	if (ioctl(RTFileToNative(hRawPart), DKIOCUNLOCKPHYSICALEXTENTS, NULL) == -1)
3842	{
3843	int rc2 = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
3844	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to unlock the partition (errno=%d)"),
3845	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
3846	if (RT_SUCCESS(rc))
3847	rc = rc2;
3848	}
3849	}
3850	}
3851
3852	#else
3853	RT_NOREF(hVol); /* PORTME */
3854	#endif
3855	if (RT_SUCCESS(rc))
3856	{
3857	/*
3858	* Compare the first 32 sectors of the partition.
3859	*
3860	* This might not be conclusive, but for partitions formatted with the more
3861	* common file systems it should be as they have a superblock copy at or near
3862	* the start of the partition (fat, fat32, ntfs, and ext4 does at least).
3863	*/
3864	size_t const cbToCompare = (size_t)RT_MIN(pPartDesc->cbData / cbSector, 32) * cbSector;
3865	uint8_t pbSector1 = (uint8_t )RTMemTmpAlloc(cbToCompare * 2);
3866	if (pbSector1 != NULL)
3867	{
3868	uint8_t *pbSector2 = pbSector1 + cbToCompare;
3869	/* Do the comparing, we repeat if it fails and the data might be volatile. */
3870	uint64_t uPrevCrc1 = 0;
3871	uint64_t uPrevCrc2 = 0;
3872	uint32_t cStable = 0;
3873	for (unsigned iTry = 0; iTry < 256; iTry++)
3874	{
3875	rc = RTFileReadAt(hRawDrive, pPartDesc->offStartInVDisk, pbSector1, cbToCompare, NULL);
3876	if (RT_SUCCESS(rc))
3877	{
3878	rc = RTFileReadAt(hRawPart, pPartDesc->offStartInDevice, pbSector2, cbToCompare, NULL);
3879	if (RT_SUCCESS(rc))
3880	{
3881	if (memcmp(pbSector1, pbSector2, cbToCompare) != 0)
3882	{
3883	rc = VERR_MISMATCH;
3884	/* Do data stability checks before repeating: */
3885	uint64_t const uCrc1 = RTCrc64(pbSector1, cbToCompare);
3886	uint64_t const uCrc2 = RTCrc64(pbSector2, cbToCompare);
3887	if ( uPrevCrc1 != uCrc1
3888	\|\| uPrevCrc2 != uCrc2)
3889	cStable = 0;
3890	else if (++cStable > 4)
3891	break;
3892	uPrevCrc1 = uCrc1;
3893	uPrevCrc2 = uCrc2;
3894	continue;
3895	}
3896	rc = VINF_SUCCESS;
3897	}
3898	else
3899	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
3900	N_("VMDK: Image path: '%s'. Error reading %zu bytes from '%s' at offset %RU64 (%Rrc)"),
3901	pImage->pszFilename, cbToCompare, pPartDesc->pszRawDevice, pPartDesc->offStartInDevice, rc);
3902	}
3903	else
3904	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
3905	N_("VMDK: Image path: '%s'. Error reading %zu bytes from '%s' at offset %RU64 (%Rrc)"),
3906	pImage->pszFilename, cbToCompare, pszRawDrive, pPartDesc->offStartInVDisk, rc);
3907	break;
3908	}
3909	if (rc == VERR_MISMATCH)
3910	{
3911	/* Find the first mismatching bytes: */
3912	size_t offMissmatch = 0;
3913	while (offMissmatch < cbToCompare && pbSector1[offMissmatch] == pbSector2[offMissmatch])
3914	offMissmatch++;
3915	int cbSample = (int)RT_MIN(cbToCompare - offMissmatch, 16);
3916	if (cStable > 0)
3917	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
3918	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s' (cStable=%d @%#zx: %.Rhxs vs %.Rhxs)"),
3919	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cStable,
3920	offMissmatch, cbSample, &pbSector1[offMissmatch], cbSample, &pbSector2[offMissmatch]);
3921	else
3922	{
3923	LogRel(("VMDK: Image path: '%s'. Partition #%u path ('%s') verification undecided on '%s' because of unstable data! (@%#zx: %.Rhxs vs %.Rhxs)\n",
3924	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
3925	offMissmatch, cbSample, &pbSector1[offMissmatch], cbSample, &pbSector2[offMissmatch]));
3926	rc = -rc;
3927	}
3928	}
3929	RTMemTmpFree(pbSector1);
3930	}
3931	else
3932	rc = vdIfError(pImage->pIfError, VERR_NO_TMP_MEMORY, RT_SRC_POS,
3933	N_("VMDK: Image path: '%s'. Failed to allocate %zu bytes for a temporary read buffer\n"),
3934	pImage->pszFilename, cbToCompare * 2);
3935	}
3936	RTFileClose(hRawPart);
3937	return rc;
3938	}
3939	#ifdef RT_OS_WINDOWS
3940	/**
3941	* Construct the device name for the given partition number.
3942	*/
3943	static int vmdkRawDescWinMakePartitionName(PVMDKIMAGE pImage, const char *pszRawDrive, RTFILE hRawDrive, uint32_t idxPartition,
3944	char **ppszRawPartition)
3945	{
3946	int rc = VINF_SUCCESS;
3947	DWORD cbReturned = 0;
3948	STORAGE_DEVICE_NUMBER DevNum;
3949	RT_ZERO(DevNum);
3950	if (DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_STORAGE_GET_DEVICE_NUMBER,
3951	NULL /pvInBuffer/, 0 /cbInBuffer/, &DevNum, sizeof(DevNum), &cbReturned, NULL /pOverlapped/))
3952	RTStrAPrintf(ppszRawPartition, "\\\\.\\Harddisk%uPartition%u", DevNum.DeviceNumber, idxPartition);
3953	else
3954	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
3955	N_("VMDK: Image path: '%s'. IOCTL_STORAGE_GET_DEVICE_NUMBER failed on '%s': %u"),
3956	pImage->pszFilename, pszRawDrive, GetLastError());
3957	return rc;
3958	}
3959	#endif /* RT_OS_WINDOWS */
3960	/**
3961	* Worker for vmdkMakeRawDescriptor that adds partition descriptors when the
3962	* 'Partitions' configuration value is present.
3963	*
3964	* @returns VBox status code, error message has been set on failure.
3965	*
3966	* @note Caller is assumed to clean up @a pRawDesc and release
3967	* @a *phVolToRelease.
3968	* @internal
3969	*/
3970	static int vmdkRawDescDoPartitions(PVMDKIMAGE pImage, RTDVM hVolMgr, PVDISKRAW pRawDesc,
3971	RTFILE hRawDrive, const char *pszRawDrive, uint32_t cbSector,
3972	uint32_t fPartitions, uint32_t fPartitionsReadOnly, bool fRelative,
3973	PRTDVMVOLUME phVolToRelease)
3974	{
3975	*phVolToRelease = NIL_RTDVMVOLUME;
3976	/* Check sanity/understanding. */
3977	Assert(fPartitions);
3978	Assert((fPartitions & fPartitionsReadOnly) == fPartitionsReadOnly); /* RO should be a sub-set */
3979	/*
3980	* Allocate on descriptor for each volume up front.
3981	*/
3982	uint32_t const cVolumes = RTDvmMapGetValidVolumes(hVolMgr);
3983	PVDISKRAWPARTDESC paPartDescs = NULL;
3984	int rc = vmdkRawDescAppendPartDesc(pImage, pRawDesc, cVolumes, &paPartDescs);
3985	AssertRCReturn(rc, rc);
3986	/*
3987	* Enumerate the partitions (volumes) on the disk and create descriptors for each of them.
3988	*/
3989	uint32_t fPartitionsLeft = fPartitions;
3990	RTDVMVOLUME hVol = NIL_RTDVMVOLUME; /* the current volume, needed for getting the next. */
3991	for (uint32_t i = 0; i < cVolumes; i++)
3992	{
3993	/*
3994	* Get the next/first volume and release the current.
3995	*/
3996	RTDVMVOLUME hVolNext = NIL_RTDVMVOLUME;
3997	if (i == 0)
3998	rc = RTDvmMapQueryFirstVolume(hVolMgr, &hVolNext);
3999	else
4000	rc = RTDvmMapQueryNextVolume(hVolMgr, hVol, &hVolNext);
4001	if (RT_FAILURE(rc))
4002	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4003	N_("VMDK: Image path: '%s'. Volume enumeration failed at volume #%u on '%s' (%Rrc)"),
4004	pImage->pszFilename, i, pszRawDrive, rc);
4005	uint32_t cRefs = RTDvmVolumeRelease(hVol);
4006	Assert(cRefs != UINT32_MAX); RT_NOREF(cRefs);
4007	*phVolToRelease = hVol = hVolNext;
4008	/*
4009	* Depending on the fPartitions selector and associated read-only mask,
4010	* the guest either gets read-write or read-only access (bits set)
4011	* or no access (selector bit clear, access directed to the VMDK).
4012	*/
4013	paPartDescs[i].cbData = RTDvmVolumeGetSize(hVol);
4014	uint64_t offVolumeEndIgnored = 0;
4015	rc = RTDvmVolumeQueryRange(hVol, &paPartDescs[i].offStartInVDisk, &offVolumeEndIgnored);
4016	if (RT_FAILURE(rc))
4017	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4018	N_("VMDK: Image path: '%s'. Failed to get location of volume #%u on '%s' (%Rrc)"),
4019	pImage->pszFilename, i, pszRawDrive, rc);
4020	Assert(paPartDescs[i].cbData == offVolumeEndIgnored + 1 - paPartDescs[i].offStartInVDisk);
4021	/* Note! The index must match IHostDrivePartition::number. */
4022	uint32_t idxPartition = RTDvmVolumeGetIndex(hVol, RTDVMVOLIDX_HOST);
4023	if ( idxPartition < 32
4024	&& (fPartitions & RT_BIT_32(idxPartition)))
4025	{
4026	fPartitionsLeft &= ~RT_BIT_32(idxPartition);
4027	if (fPartitionsReadOnly & RT_BIT_32(idxPartition))
4028	paPartDescs[i].uFlags \|= VDISKRAW_READONLY;
4029	if (!fRelative)
4030	{
4031	/*
4032	* Accessing the drive thru the main device node (pRawDesc->pszRawDisk).
4033	*/
4034	paPartDescs[i].offStartInDevice = paPartDescs[i].offStartInVDisk;
4035	paPartDescs[i].pszRawDevice = RTStrDup(pszRawDrive);
4036	AssertPtrReturn(paPartDescs[i].pszRawDevice, VERR_NO_STR_MEMORY);
4037	}
4038	else
4039	{
4040	/*
4041	* Relative means access the partition data via the device node for that
4042	* partition, allowing the sysadmin/OS to allow a user access to individual
4043	* partitions without necessarily being able to compromise the host OS.
4044	* Obviously, the creation of the VMDK requires read access to the main
4045	* device node for the drive, but that's a one-time thing and can be done
4046	* by the sysadmin. Here data starts at offset zero in the device node.
4047	*/
4048	paPartDescs[i].offStartInDevice = 0;
4049	#if defined(RT_OS_DARWIN) \|\| defined(RT_OS_FREEBSD)
4050	/* /dev/rdisk1 -> /dev/rdisk1s2 (s=slice) */
4051	RTStrAPrintf(&paPartDescs[i].pszRawDevice, "%ss%u", pszRawDrive, idxPartition);
4052	#elif defined(RT_OS_LINUX)
4053	/* Two naming schemes here: /dev/nvme0n1 -> /dev/nvme0n1p1; /dev/sda -> /dev/sda1 */
4054	RTStrAPrintf(&paPartDescs[i].pszRawDevice,
4055	RT_C_IS_DIGIT(pszRawDrive[strlen(pszRawDrive) - 1]) ? "%sp%u" : "%s%u", pszRawDrive, idxPartition);
4056	#elif defined(RT_OS_WINDOWS)
4057	rc = vmdkRawDescWinMakePartitionName(pImage, pszRawDrive, hRawDrive, idxPartition, &paPartDescs[i].pszRawDevice);
4058	AssertRCReturn(rc, rc);
4059	#elif defined(RT_OS_SOLARIS)
4060	if (pRawDesc->enmPartitioningType == VDISKPARTTYPE_MBR)
4061	{
4062	/*
4063	* MBR partitions have device nodes in form /dev/(r)dsk/cXtYdZpK
4064	* where X is the controller,
4065	* Y is target (SCSI device number),
4066	* Z is disk number,
4067	* K is partition number,
4068	* where p0 is the whole disk
4069	* p1-pN are the partitions of the disk
4070	*/
4071	const char *pszRawDrivePath = pszRawDrive;
4072	char szDrivePath[RTPATH_MAX];
4073	size_t cbRawDrive = strlen(pszRawDrive);
4074	if ( cbRawDrive > 1 && strcmp(&pszRawDrive[cbRawDrive - 2], "p0") == 0)
4075	{
4076	memcpy(szDrivePath, pszRawDrive, cbRawDrive - 2);
4077	szDrivePath[cbRawDrive - 2] = '\0';
4078	pszRawDrivePath = szDrivePath;
4079	}
4080	RTStrAPrintf(&paPartDescs[i].pszRawDevice, "%sp%u", pszRawDrivePath, idxPartition);
4081	}
4082	else /* GPT */
4083	{
4084	/*
4085	* GPT partitions have device nodes in form /dev/(r)dsk/cXtYdZsK
4086	* where X is the controller,
4087	* Y is target (SCSI device number),
4088	* Z is disk number,
4089	* K is partition number, zero based. Can be only from 0 to 6.
4090	* Thus, only partitions numbered 0 through 6 have device nodes.
4091	*/
4092	if (idxPartition > 7)
4093	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4094	N_("VMDK: Image path: '%s'. the partition #%u on '%s' has no device node and can not be specified with 'Relative' property"),
4095	pImage->pszFilename, idxPartition, pszRawDrive);
4096	RTStrAPrintf(&paPartDescs[i].pszRawDevice, "%ss%u", pszRawDrive, idxPartition - 1);
4097	}
4098	#else
4099	AssertFailedReturn(VERR_INTERNAL_ERROR_4); /* The option parsing code should have prevented this - PORTME */
4100	#endif
4101	AssertPtrReturn(paPartDescs[i].pszRawDevice, VERR_NO_STR_MEMORY);
4102	rc = vmdkRawDescVerifyPartitionPath(pImage, &paPartDescs[i], idxPartition, pszRawDrive, hRawDrive, cbSector, hVol);
4103	AssertRCReturn(rc, rc);
4104	}
4105	}
4106	else
4107	{
4108	/* Not accessible to the guest. */
4109	paPartDescs[i].offStartInDevice = 0;
4110	paPartDescs[i].pszRawDevice = NULL;
4111	}
4112	} /* for each volume */
4113	RTDvmVolumeRelease(hVol);
4114	*phVolToRelease = NIL_RTDVMVOLUME;
4115	/*
4116	* Check that we found all the partitions the user selected.
4117	*/
4118	if (fPartitionsLeft)
4119	{
4120	char szLeft[3 * sizeof(fPartitions) * 8];
4121	size_t cchLeft = 0;
4122	for (unsigned i = 0; i < sizeof(fPartitions) * 8; i++)
4123	if (fPartitionsLeft & RT_BIT_32(i))
4124	cchLeft += RTStrPrintf(&szLeft[cchLeft], sizeof(szLeft) - cchLeft, cchLeft ? "%u" : ",%u", i);
4125	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4126	N_("VMDK: Image path: '%s'. Not all the specified partitions for drive '%s' was found: %s"),
4127	pImage->pszFilename, pszRawDrive, szLeft);
4128	}
4129	return VINF_SUCCESS;
4130	}
4131	/**
4132	* Worker for vmdkMakeRawDescriptor that adds partition descriptors with copies
4133	* of the partition tables and associated padding areas when the 'Partitions'
4134	* configuration value is present.
4135	*
4136	* The guest is not allowed access to the partition tables, however it needs
4137	* them to be able to access the drive. So, create descriptors for each of the
4138	* tables and attach the current disk content. vmdkCreateRawImage() will later
4139	* write the content to the VMDK. Any changes the guest later makes to the
4140	* partition tables will then go to the VMDK copy, rather than the host drive.
4141	*
4142	* @returns VBox status code, error message has been set on failure.
4143	*
4144	* @note Caller is assumed to clean up @a pRawDesc
4145	* @internal
4146	*/
4147	static int vmdkRawDescDoCopyPartitionTables(PVMDKIMAGE pImage, RTDVM hVolMgr, PVDISKRAW pRawDesc,
4148	const char pszRawDrive, RTFILE hRawDrive, void pvBootSector, size_t cbBootSector)
4149	{
4150	/*
4151	* Query the locations.
4152	*/
4153	/* Determin how many locations there are: */
4154	size_t cLocations = 0;
4155	int rc = RTDvmMapQueryTableLocations(hVolMgr, RTDVMMAPQTABLOC_F_INCLUDE_LEGACY, NULL, 0, &cLocations);
4156	if (rc != VERR_BUFFER_OVERFLOW)
4157	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4158	N_("VMDK: Image path: '%s'. RTDvmMapQueryTableLocations failed on '%s' (%Rrc)"),
4159	pImage->pszFilename, pszRawDrive, rc);
4160	AssertReturn(cLocations > 0 && cLocations < _16M, VERR_INTERNAL_ERROR_5);
4161	/* We can allocate the partition descriptors here to save an intentation level. */
4162	PVDISKRAWPARTDESC paPartDescs = NULL;
4163	rc = vmdkRawDescAppendPartDesc(pImage, pRawDesc, (uint32_t)cLocations, &paPartDescs);
4164	AssertRCReturn(rc, rc);
4165	/* Allocate the result table and repeat the location table query: */
4166	PRTDVMTABLELOCATION paLocations = (PRTDVMTABLELOCATION)RTMemAllocZ(sizeof(paLocations[0]) * cLocations);
4167	if (!paLocations)
4168	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS, N_("VMDK: Image path: '%s'. Failed to allocate %zu bytes"),
4169	pImage->pszFilename, sizeof(paLocations[0]) * cLocations);
4170	rc = RTDvmMapQueryTableLocations(hVolMgr, RTDVMMAPQTABLOC_F_INCLUDE_LEGACY, paLocations, cLocations, NULL);
4171	if (RT_SUCCESS(rc))
4172	{
4173	/*
4174	* Translate them into descriptors.
4175	*
4176	* We restrict the amount of partition alignment padding to 4MiB as more
4177	* will just be a waste of space. The use case for including the padding
4178	* are older boot loaders and boot manager (including one by a team member)
4179	* that put data and code in the 62 sectors between the MBR and the first
4180	* partition (total of 63). Later CHS was abandond and partition started
4181	* being aligned on power of two sector boundraries (typically 64KiB or
4182	* 1MiB depending on the media size).
4183	*/
4184	for (size_t i = 0; i < cLocations && RT_SUCCESS(rc); i++)
4185	{
4186	Assert(paLocations[i].cb > 0);
4187	if (paLocations[i].cb <= _64M)
4188	{
4189	/* Create the partition descriptor entry: */
4190	//paPartDescs[i].pszRawDevice = NULL;
4191	//paPartDescs[i].offStartInDevice = 0;
4192	//paPartDescs[i].uFlags = 0;
4193	paPartDescs[i].offStartInVDisk = paLocations[i].off;
4194	paPartDescs[i].cbData = paLocations[i].cb;
4195	if (paPartDescs[i].cbData < _4M)
4196	paPartDescs[i].cbData = RT_MIN(paPartDescs[i].cbData + paLocations[i].cbPadding, _4M);
4197	paPartDescs[i].pvPartitionData = RTMemAllocZ((size_t)paPartDescs[i].cbData);
4198	if (paPartDescs[i].pvPartitionData)
4199	{
4200	/* Read the content from the drive: */
4201	rc = RTFileReadAt(hRawDrive, paPartDescs[i].offStartInVDisk, paPartDescs[i].pvPartitionData,
4202	(size_t)paPartDescs[i].cbData, NULL);
4203	if (RT_SUCCESS(rc))
4204	{
4205	/* Do we have custom boot sector code? */
4206	if (pvBootSector && cbBootSector && paPartDescs[i].offStartInVDisk == 0)
4207	{
4208	/* Note! Old code used to quietly drop the bootsector if it was considered too big.
4209	Instead we fail as we weren't able to do what the user requested us to do.
4210	Better if the user knows than starts questioning why the guest isn't
4211	booting as expected. */
4212	if (cbBootSector <= paPartDescs[i].cbData)
4213	memcpy(paPartDescs[i].pvPartitionData, pvBootSector, cbBootSector);
4214	else
4215	rc = vdIfError(pImage->pIfError, VERR_TOO_MUCH_DATA, RT_SRC_POS,
4216	N_("VMDK: Image path: '%s'. The custom boot sector is too big: %zu bytes, %RU64 bytes available"),
4217	pImage->pszFilename, cbBootSector, paPartDescs[i].cbData);
4218	}
4219	}
4220	else
4221	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4222	N_("VMDK: Image path: '%s'. Failed to read partition at off %RU64 length %zu from '%s' (%Rrc)"),
4223	pImage->pszFilename, paPartDescs[i].offStartInVDisk,
4224	(size_t)paPartDescs[i].cbData, pszRawDrive, rc);
4225	}
4226	else
4227	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4228	N_("VMDK: Image path: '%s'. Failed to allocate %zu bytes for copying the partition table at off %RU64"),
4229	pImage->pszFilename, (size_t)paPartDescs[i].cbData, paPartDescs[i].offStartInVDisk);
4230	}
4231	else
4232	rc = vdIfError(pImage->pIfError, VERR_TOO_MUCH_DATA, RT_SRC_POS,
4233	N_("VMDK: Image path: '%s'. Partition table #%u at offset %RU64 in '%s' is to big: %RU64 bytes"),
4234	pImage->pszFilename, i, paLocations[i].off, pszRawDrive, paLocations[i].cb);
4235	}
4236	}
4237	else
4238	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4239	N_("VMDK: Image path: '%s'. RTDvmMapQueryTableLocations failed on '%s' (%Rrc)"),
4240	pImage->pszFilename, pszRawDrive, rc);
4241	RTMemFree(paLocations);
4242	return rc;
4243	}
4244	/**
4245	* Opens the volume manager for the raw drive when in selected-partition mode.
4246	*
4247	* @param pImage The VMDK image (for errors).
4248	* @param hRawDrive The raw drive handle.
4249	* @param pszRawDrive The raw drive device path (for errors).
4250	* @param cbSector The sector size.
4251	* @param phVolMgr Where to return the handle to the volume manager on
4252	* success.
4253	* @returns VBox status code, errors have been reported.
4254	* @internal
4255	*/
4256	static int vmdkRawDescOpenVolMgr(PVMDKIMAGE pImage, RTFILE hRawDrive, const char *pszRawDrive, uint32_t cbSector, PRTDVM phVolMgr)
4257	{
4258	*phVolMgr = NIL_RTDVM;
4259	RTVFSFILE hVfsFile = NIL_RTVFSFILE;
4260	int rc = RTVfsFileFromRTFile(hRawDrive, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE, true /fLeaveOpen/, &hVfsFile);
4261	if (RT_FAILURE(rc))
4262	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4263	N_("VMDK: Image path: '%s'. RTVfsFileFromRTFile failed for '%s' handle (%Rrc)"),
4264	pImage->pszFilename, pszRawDrive, rc);
4265	RTDVM hVolMgr = NIL_RTDVM;
4266	rc = RTDvmCreate(&hVolMgr, hVfsFile, cbSector, 0 /fFlags/);
4267	RTVfsFileRelease(hVfsFile);
4268	if (RT_FAILURE(rc))
4269	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4270	N_("VMDK: Image path: '%s'. Failed to create volume manager instance for '%s' (%Rrc)"),
4271	pImage->pszFilename, pszRawDrive, rc);
4272	rc = RTDvmMapOpen(hVolMgr);
4273	if (RT_SUCCESS(rc))
4274	{
4275	*phVolMgr = hVolMgr;
4276	return VINF_SUCCESS;
4277	}
4278	RTDvmRelease(hVolMgr);
4279	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: Image path: '%s'. RTDvmMapOpen failed for '%s' (%Rrc)"),
4280	pImage->pszFilename, pszRawDrive, rc);
4281	}
4282	/**
4283	* Opens the raw drive device and get the sizes for it.
4284	*
4285	* @param pImage The image (for error reporting).
4286	* @param pszRawDrive The device/whatever to open.
4287	* @param phRawDrive Where to return the file handle.
4288	* @param pcbRawDrive Where to return the size.
4289	* @param pcbSector Where to return the sector size.
4290	* @returns IPRT status code, errors have been reported.
4291	* @internal
4292	*/
4293	static int vmkdRawDescOpenDevice(PVMDKIMAGE pImage, const char *pszRawDrive,
4294	PRTFILE phRawDrive, uint64_t pcbRawDrive, uint32_t pcbSector)
4295	{
4296	/*
4297	* Open the device for the raw drive.
4298	*/
4299	RTFILE hRawDrive = NIL_RTFILE;
4300	int rc = RTFileOpen(&hRawDrive, pszRawDrive, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE);
4301	if (RT_FAILURE(rc))
4302	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4303	N_("VMDK: Image path: '%s'. Failed to open the raw drive '%s' for reading (%Rrc)"),
4304	pImage->pszFilename, pszRawDrive, rc);
4305	/*
4306	* Get the sector size.
4307	*/
4308	uint32_t cbSector = 0;
4309	rc = RTFileQuerySectorSize(hRawDrive, &cbSector);
4310	if (RT_SUCCESS(rc))
4311	{
4312	/* sanity checks */
4313	if ( cbSector >= 512
4314	&& cbSector <= _64K
4315	&& RT_IS_POWER_OF_TWO(cbSector))
4316	{
4317	/*
4318	* Get the size.
4319	*/
4320	uint64_t cbRawDrive = 0;
4321	rc = RTFileQuerySize(hRawDrive, &cbRawDrive);
4322	if (RT_SUCCESS(rc))
4323	{
4324	/* Check whether cbSize is actually sensible. */
4325	if (cbRawDrive > cbSector && (cbRawDrive % cbSector) == 0)
4326	{
4327	*phRawDrive = hRawDrive;
4328	*pcbRawDrive = cbRawDrive;
4329	*pcbSector = cbSector;
4330	return VINF_SUCCESS;
4331	}
4332	rc = vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4333	N_("VMDK: Image path: '%s'. Got a bogus size for the raw drive '%s': %RU64 (sector size %u)"),
4334	pImage->pszFilename, pszRawDrive, cbRawDrive, cbSector);
4335	}
4336	else
4337	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4338	N_("VMDK: Image path: '%s'. Failed to query size of the drive '%s' (%Rrc)"),
4339	pImage->pszFilename, pszRawDrive, rc);
4340	}
4341	else
4342	rc = vdIfError(pImage->pIfError, VERR_OUT_OF_RANGE, RT_SRC_POS,
4343	N_("VMDK: Image path: '%s'. Unsupported sector size for '%s': %u (%#x)"),
4344	pImage->pszFilename, pszRawDrive, cbSector, cbSector);
4345	}
4346	else
4347	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4348	N_("VMDK: Image path: '%s'. Failed to get the sector size for '%s' (%Rrc)"),
4349	pImage->pszFilename, pszRawDrive, rc);
4350	RTFileClose(hRawDrive);
4351	return rc;
4352	}
4353	/**
4354	* Reads the raw disk configuration, leaving initalization and cleanup to the
4355	* caller (regardless of return status).
4356	*
4357	* @returns VBox status code, errors properly reported.
4358	* @internal
4359	*/
4360	static int vmdkRawDescParseConfig(PVMDKIMAGE pImage, char **ppszRawDrive,
4361	uint32_t pfPartitions, uint32_t pfPartitionsReadOnly,
4362	void *ppvBootSector, size_t pcbBootSector, bool *pfRelative,
4363	char **ppszFreeMe)
4364	{
4365	PVDINTERFACECONFIG pImgCfg = VDIfConfigGet(pImage->pVDIfsImage);
4366	if (!pImgCfg)
4367	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4368	N_("VMDK: Image path: '%s'. Getting config interface failed"), pImage->pszFilename);
4369	/*
4370	* RawDrive = path
4371	*/
4372	int rc = VDCFGQueryStringAlloc(pImgCfg, "RawDrive", ppszRawDrive);
4373	if (RT_FAILURE(rc))
4374	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4375	N_("VMDK: Image path: '%s'. Getting 'RawDrive' configuration failed (%Rrc)"), pImage->pszFilename, rc);
4376	AssertPtrReturn(*ppszRawDrive, VERR_INTERNAL_ERROR_3);
4377	/*
4378	* Partitions=n[r][,...]
4379	*/
4380	uint32_t const cMaxPartitionBits = sizeof(pfPartitions) 8 /* ASSUMES 8 bits per char */;
4381	pfPartitions = pfPartitionsReadOnly = 0;
4382	rc = VDCFGQueryStringAlloc(pImgCfg, "Partitions", ppszFreeMe);
4383	if (RT_SUCCESS(rc))
4384	{
4385	char psz = ppszFreeMe;
4386	while (*psz != '\0')
4387	{
4388	char *pszNext;
4389	uint32_t u32;
4390	rc = RTStrToUInt32Ex(psz, &pszNext, 0, &u32);
4391	if (rc == VWRN_NUMBER_TOO_BIG \|\| rc == VWRN_NEGATIVE_UNSIGNED)
4392	rc = -rc;
4393	if (RT_FAILURE(rc))
4394	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4395	N_("VMDK: Image path: '%s'. Parsing 'Partitions' config value failed. Incorrect value (%Rrc): %s"),
4396	pImage->pszFilename, rc, psz);
4397	if (u32 >= cMaxPartitionBits)
4398	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4399	N_("VMDK: Image path: '%s'. 'Partitions' config sub-value out of range: %RU32, max %RU32"),
4400	pImage->pszFilename, u32, cMaxPartitionBits);
4401	*pfPartitions \|= RT_BIT_32(u32);
4402	psz = pszNext;
4403	if (*psz == 'r')
4404	{
4405	*pfPartitionsReadOnly \|= RT_BIT_32(u32);
4406	psz++;
4407	}
4408	if (*psz == ',')
4409	psz++;
4410	else if (*psz != '\0')
4411	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4412	N_("VMDK: Image path: '%s'. Malformed 'Partitions' config value, expected separator: %s"),
4413	pImage->pszFilename, psz);
4414	}
4415	RTStrFree(*ppszFreeMe);
4416	*ppszFreeMe = NULL;
4417	}
4418	else if (rc != VERR_CFGM_VALUE_NOT_FOUND)
4419	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4420	N_("VMDK: Image path: '%s'. Getting 'Partitions' configuration failed (%Rrc)"), pImage->pszFilename, rc);
4421	/*
4422	* BootSector=base64
4423	*/
4424	rc = VDCFGQueryStringAlloc(pImgCfg, "BootSector", ppszFreeMe);
4425	if (RT_SUCCESS(rc))
4426	{
4427	ssize_t cbBootSector = RTBase64DecodedSize(*ppszFreeMe, NULL);
4428	if (cbBootSector < 0)
4429	return vdIfError(pImage->pIfError, VERR_INVALID_BASE64_ENCODING, RT_SRC_POS,
4430	N_("VMDK: Image path: '%s'. BASE64 decoding failed on the custom bootsector for '%s'"),
4431	pImage->pszFilename, *ppszRawDrive);
4432	if (cbBootSector == 0)
4433	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4434	N_("VMDK: Image path: '%s'. Custom bootsector for '%s' is zero bytes big"),
4435	pImage->pszFilename, *ppszRawDrive);
4436	if (cbBootSector > _4M) /* this is just a preliminary max */
4437	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4438	N_("VMDK: Image path: '%s'. Custom bootsector for '%s' is way too big: %zu bytes, max 4MB"),
4439	pImage->pszFilename, *ppszRawDrive, cbBootSector);
4440	/* Refuse the boot sector if whole-drive. This used to be done quietly,
4441	however, bird disagrees and thinks the user should be told that what
4442	he/she/it tries to do isn't possible. There should be less head
4443	scratching this way when the guest doesn't do the expected thing. */
4444	if (!*pfPartitions)
4445	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4446	N_("VMDK: Image path: '%s'. Custom bootsector for '%s' is not supported for whole-drive configurations, only when selecting partitions"),
4447	pImage->pszFilename, *ppszRawDrive);
4448	*pcbBootSector = (size_t)cbBootSector;
4449	*ppvBootSector = RTMemAlloc((size_t)cbBootSector);
4450	if (!*ppvBootSector)
4451	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS,
4452	N_("VMDK: Image path: '%s'. Failed to allocate %zd bytes for the custom bootsector for '%s'"),
4453	pImage->pszFilename, cbBootSector, *ppszRawDrive);
4454	rc = RTBase64Decode(ppszFreeMe, ppvBootSector, cbBootSector, NULL /pcbActual/, NULL /ppszEnd/);
4455	if (RT_FAILURE(rc))
4456	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS,
4457	N_("VMDK: Image path: '%s'. Base64 decoding of the custom boot sector for '%s' failed (%Rrc)"),
4458	pImage->pszFilename, *ppszRawDrive, rc);
4459	RTStrFree(*ppszFreeMe);
4460	*ppszFreeMe = NULL;
4461	}
4462	else if (rc != VERR_CFGM_VALUE_NOT_FOUND)
4463	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4464	N_("VMDK: Image path: '%s'. Getting 'BootSector' configuration failed (%Rrc)"), pImage->pszFilename, rc);
4465	/*
4466	* Relative=0/1
4467	*/
4468	*pfRelative = false;
4469	rc = VDCFGQueryBool(pImgCfg, "Relative", pfRelative);
4470	if (RT_SUCCESS(rc))
4471	{
4472	if (!pfPartitions && pfRelative != false)
4473	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4474	N_("VMDK: Image path: '%s'. The 'Relative' option is not supported for whole-drive configurations, only when selecting partitions"),
4475	pImage->pszFilename);
4476	#if !defined(RT_OS_DARWIN) && !defined(RT_OS_LINUX) && !defined(RT_OS_FREEBSD) && !defined(RT_OS_WINDOWS) && !defined(RT_OS_SOLARIS) /* PORTME */
4477	if (*pfRelative == true)
4478	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4479	N_("VMDK: Image path: '%s'. The 'Relative' option is not supported on this host OS"),
4480	pImage->pszFilename);
4481	#endif
4482	}
4483	else if (rc != VERR_CFGM_VALUE_NOT_FOUND)
4484	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4485	N_("VMDK: Image path: '%s'. Getting 'Relative' configuration failed (%Rrc)"), pImage->pszFilename, rc);
4486	else
4487	#ifdef RT_OS_DARWIN /* different default on macOS, see ticketref:1461 (comment 20). */
4488	*pfRelative = true;
4489	#else
4490	*pfRelative = false;
4491	#endif
4492	return VINF_SUCCESS;
4493	}
4494	/**
4495	* Creates a raw drive (nee disk) descriptor.
4496	*
4497	* This was originally done in VBoxInternalManage.cpp, but was copied (not move)
4498	* here much later. That's one of the reasons why we produce a descriptor just
4499	* like it does, rather than mixing directly into the vmdkCreateRawImage code.
4500	*
4501	* @returns VBox status code.
4502	* @param pImage The image.
4503	* @param ppRaw Where to return the raw drive descriptor. Caller must
4504	* free it using vmdkRawDescFree regardless of the status
4505	* code.
4506	* @internal
4507	*/
4508	static int vmdkMakeRawDescriptor(PVMDKIMAGE pImage, PVDISKRAW *ppRaw)
4509	{
4510	/* Make sure it's NULL. */
4511	*ppRaw = NULL;
4512	/*
4513	* Read the configuration.
4514	*/
4515	char *pszRawDrive = NULL;
4516	uint32_t fPartitions = 0; /* zero if whole-drive */
4517	uint32_t fPartitionsReadOnly = 0; /* (subset of fPartitions) */
4518	void *pvBootSector = NULL;
4519	size_t cbBootSector = 0;
4520	bool fRelative = false;
4521	char pszFreeMe = NULL; / lazy bird cleanup. */
4522	int rc = vmdkRawDescParseConfig(pImage, &pszRawDrive, &fPartitions, &fPartitionsReadOnly,
4523	&pvBootSector, &cbBootSector, &fRelative, &pszFreeMe);
4524	RTStrFree(pszFreeMe);
4525	if (RT_SUCCESS(rc))
4526	{
4527	/*
4528	* Open the device, getting the sector size and drive size.
4529	*/
4530	uint64_t cbSize = 0;
4531	uint32_t cbSector = 0;
4532	RTFILE hRawDrive = NIL_RTFILE;
4533	rc = vmkdRawDescOpenDevice(pImage, pszRawDrive, &hRawDrive, &cbSize, &cbSector);
4534	if (RT_SUCCESS(rc))
4535	{
4536	/*
4537	* Create the raw-drive descriptor
4538	*/
4539	PVDISKRAW pRawDesc = (PVDISKRAW)RTMemAllocZ(sizeof(*pRawDesc));
4540	if (pRawDesc)
4541	{
4542	pRawDesc->szSignature[0] = 'R';
4543	pRawDesc->szSignature[1] = 'A';
4544	pRawDesc->szSignature[2] = 'W';
4545	//pRawDesc->szSignature[3] = '\0';
4546	if (!fPartitions)
4547	{
4548	/*
4549	* It's simple for when doing the whole drive.
4550	*/
4551	pRawDesc->uFlags = VDISKRAW_DISK;
4552	rc = RTStrDupEx(&pRawDesc->pszRawDisk, pszRawDrive);
4553	}
4554	else
4555	{
4556	/*
4557	* In selected partitions mode we've got a lot more work ahead of us.
4558	*/
4559	pRawDesc->uFlags = VDISKRAW_NORMAL;
4560	//pRawDesc->pszRawDisk = NULL;
4561	//pRawDesc->cPartDescs = 0;
4562	//pRawDesc->pPartDescs = NULL;
4563	/* We need to parse the partition map to complete the descriptor: */
4564	RTDVM hVolMgr = NIL_RTDVM;
4565	rc = vmdkRawDescOpenVolMgr(pImage, hRawDrive, pszRawDrive, cbSector, &hVolMgr);
4566	if (RT_SUCCESS(rc))
4567	{
4568	RTDVMFORMATTYPE enmFormatType = RTDvmMapGetFormatType(hVolMgr);
4569	if ( enmFormatType == RTDVMFORMATTYPE_MBR
4570	\|\| enmFormatType == RTDVMFORMATTYPE_GPT)
4571	{
4572	pRawDesc->enmPartitioningType = enmFormatType == RTDVMFORMATTYPE_MBR
4573	? VDISKPARTTYPE_MBR : VDISKPARTTYPE_GPT;
4574	/* Add copies of the partition tables: */
4575	rc = vmdkRawDescDoCopyPartitionTables(pImage, hVolMgr, pRawDesc, pszRawDrive, hRawDrive,
4576	pvBootSector, cbBootSector);
4577	if (RT_SUCCESS(rc))
4578	{
4579	/* Add descriptors for the partitions/volumes, indicating which
4580	should be accessible and how to access them: */
4581	RTDVMVOLUME hVolRelease = NIL_RTDVMVOLUME;
4582	rc = vmdkRawDescDoPartitions(pImage, hVolMgr, pRawDesc, hRawDrive, pszRawDrive, cbSector,
4583	fPartitions, fPartitionsReadOnly, fRelative, &hVolRelease);
4584	RTDvmVolumeRelease(hVolRelease);
4585	/* Finally, sort the partition and check consistency (overlaps, etc): */
4586	if (RT_SUCCESS(rc))
4587	rc = vmdkRawDescPostProcessPartitions(pImage, pRawDesc, cbSize);
4588	}
4589	}
4590	else
4591	rc = vdIfError(pImage->pIfError, VERR_NOT_SUPPORTED, RT_SRC_POS,
4592	N_("VMDK: Image path: '%s'. Unsupported partitioning for the disk '%s': %s"),
4593	pImage->pszFilename, pszRawDrive, RTDvmMapGetFormatType(hVolMgr));
4594	RTDvmRelease(hVolMgr);
4595	}
4596	}
4597	if (RT_SUCCESS(rc))
4598	{
4599	/*
4600	* We succeeded.
4601	*/
4602	*ppRaw = pRawDesc;
4603	Log(("vmdkMakeRawDescriptor: fFlags=%#x enmPartitioningType=%d cPartDescs=%u pszRawDisk=%s\n",
4604	pRawDesc->uFlags, pRawDesc->enmPartitioningType, pRawDesc->cPartDescs, pRawDesc->pszRawDisk));
4605	if (pRawDesc->cPartDescs)
4606	{
4607	Log(("# VMDK offset Length Device offset PartDataPtr Device\n"));
4608	for (uint32_t i = 0; i < pRawDesc->cPartDescs; i++)
4609	Log(("%2u %14RU64 %14RU64 %14RU64 %#18p %s\n", i, pRawDesc->pPartDescs[i].offStartInVDisk,
4610	pRawDesc->pPartDescs[i].cbData, pRawDesc->pPartDescs[i].offStartInDevice,
4611	pRawDesc->pPartDescs[i].pvPartitionData, pRawDesc->pPartDescs[i].pszRawDevice));
4612	}
4613	}
4614	else
4615	vmdkRawDescFree(pRawDesc);
4616	}
4617	else
4618	rc = vdIfError(pImage->pIfError, VERR_NOT_SUPPORTED, RT_SRC_POS,
4619	N_("VMDK: Image path: '%s'. Failed to allocate %u bytes for the raw drive descriptor"),
4620	pImage->pszFilename, sizeof(*pRawDesc));
4621	RTFileClose(hRawDrive);
4622	}
4623	}
4624	RTStrFree(pszRawDrive);
4625	RTMemFree(pvBootSector);
4626	return rc;
4627	}
4628	/**
4629	* Internal: create VMDK images for raw disk/partition access.
4630	*/
4631	static int vmdkCreateRawImage(PVMDKIMAGE pImage, const PVDISKRAW pRaw,
4632	uint64_t cbSize)
4633	{
4634	int rc = VINF_SUCCESS;
4635	PVMDKEXTENT pExtent;
4636	if (pRaw->uFlags & VDISKRAW_DISK)
4637	{
4638	/* Full raw disk access. This requires setting up a descriptor
4639	* file and open the (flat) raw disk. */
4640	rc = vmdkCreateExtents(pImage, 1);
4641	if (RT_FAILURE(rc))
4642	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
4643	pExtent = &pImage->pExtents[0];
4644	/* Create raw disk descriptor file. */
4645	rc = vmdkFileOpen(pImage, &pImage->pFile, NULL, pImage->pszFilename,
4646	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
4647	true /* fCreate */));
4648	if (RT_FAILURE(rc))
4649	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pImage->pszFilename);
4650	/* Set up basename for extent description. Cannot use StrDup. */
4651	size_t cbBasename = strlen(pRaw->pszRawDisk) + 1;
4652	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
4653	if (!pszBasename)
4654	return VERR_NO_MEMORY;
4655	memcpy(pszBasename, pRaw->pszRawDisk, cbBasename);
4656	pExtent->pszBasename = pszBasename;
4657	/* For raw disks the full name is identical to the base name. */
4658	pExtent->pszFullname = RTStrDup(pszBasename);
4659	if (!pExtent->pszFullname)
4660	return VERR_NO_MEMORY;
4661	pExtent->enmType = VMDKETYPE_FLAT;
4662	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
4663	pExtent->uSectorOffset = 0;
4664	pExtent->enmAccess = (pRaw->uFlags & VDISKRAW_READONLY) ? VMDKACCESS_READONLY : VMDKACCESS_READWRITE;
4665	pExtent->fMetaDirty = false;
4666	/* Open flat image, the raw disk. */
4667	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
4668	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
4669	false /* fCreate */));
4670	if (RT_FAILURE(rc))
4671	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not open raw disk file '%s'"), pExtent->pszFullname);
4672	}
4673	else
4674	{
4675	/* Raw partition access. This requires setting up a descriptor
4676	* file, write the partition information to a flat extent and
4677	* open all the (flat) raw disk partitions. */
4678	/* First pass over the partition data areas to determine how many
4679	* extents we need. One data area can require up to 2 extents, as
4680	* it might be necessary to skip over unpartitioned space. */
4681	unsigned cExtents = 0;
4682	uint64_t uStart = 0;
4683	for (unsigned i = 0; i < pRaw->cPartDescs; i++)
4684	{
4685	PVDISKRAWPARTDESC pPart = &pRaw->pPartDescs[i];
4686	if (uStart > pPart->offStartInVDisk)
4687	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4688	N_("VMDK: incorrect partition data area ordering set up by the caller in '%s'"), pImage->pszFilename);
4689	if (uStart < pPart->offStartInVDisk)
4690	cExtents++;
4691	uStart = pPart->offStartInVDisk + pPart->cbData;
4692	cExtents++;
4693	}
4694	/* Another extent for filling up the rest of the image. */
4695	if (uStart != cbSize)
4696	cExtents++;
4697	rc = vmdkCreateExtents(pImage, cExtents);
4698	if (RT_FAILURE(rc))
4699	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
4700	/* Create raw partition descriptor file. */
4701	rc = vmdkFileOpen(pImage, &pImage->pFile, NULL, pImage->pszFilename,
4702	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
4703	true /* fCreate */));
4704	if (RT_FAILURE(rc))
4705	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pImage->pszFilename);
4706	/* Create base filename for the partition table extent. */
4707	/** @todo remove fixed buffer without creating memory leaks. */
4708	char pszPartition[1024];
4709	const char *pszBase = RTPathFilename(pImage->pszFilename);
4710	const char *pszSuff = RTPathSuffix(pszBase);
4711	if (pszSuff == NULL)
4712	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: invalid filename '%s'"), pImage->pszFilename);
4713	char *pszBaseBase = RTStrDup(pszBase);
4714	if (!pszBaseBase)
4715	return VERR_NO_MEMORY;
4716	RTPathStripSuffix(pszBaseBase);
4717	RTStrPrintf(pszPartition, sizeof(pszPartition), "%s-pt%s",
4718	pszBaseBase, pszSuff);
4719	RTStrFree(pszBaseBase);
4720	/* Second pass over the partitions, now define all extents. */
4721	uint64_t uPartOffset = 0;
4722	cExtents = 0;
4723	uStart = 0;
4724	for (unsigned i = 0; i < pRaw->cPartDescs; i++)
4725	{
4726	PVDISKRAWPARTDESC pPart = &pRaw->pPartDescs[i];
4727	pExtent = &pImage->pExtents[cExtents++];
4728	if (uStart < pPart->offStartInVDisk)
4729	{
4730	pExtent->pszBasename = NULL;
4731	pExtent->pszFullname = NULL;
4732	pExtent->enmType = VMDKETYPE_ZERO;
4733	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->offStartInVDisk - uStart);
4734	pExtent->uSectorOffset = 0;
4735	pExtent->enmAccess = VMDKACCESS_READWRITE;
4736	pExtent->fMetaDirty = false;
4737	/* go to next extent */
4738	pExtent = &pImage->pExtents[cExtents++];
4739	}
4740	uStart = pPart->offStartInVDisk + pPart->cbData;
4741	if (pPart->pvPartitionData)
4742	{
4743	/* Set up basename for extent description. Can't use StrDup. */
4744	size_t cbBasename = strlen(pszPartition) + 1;
4745	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
4746	if (!pszBasename)
4747	return VERR_NO_MEMORY;
4748	memcpy(pszBasename, pszPartition, cbBasename);
4749	pExtent->pszBasename = pszBasename;
4750	/* Set up full name for partition extent. */
4751	char *pszDirname = RTStrDup(pImage->pszFilename);
4752	if (!pszDirname)
4753	return VERR_NO_STR_MEMORY;
4754	RTPathStripFilename(pszDirname);
4755	char *pszFullname = RTPathJoinA(pszDirname, pExtent->pszBasename);
4756	RTStrFree(pszDirname);
4757	if (!pszFullname)
4758	return VERR_NO_STR_MEMORY;
4759	pExtent->pszFullname = pszFullname;
4760	pExtent->enmType = VMDKETYPE_FLAT;
4761	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
4762	pExtent->uSectorOffset = uPartOffset;
4763	pExtent->enmAccess = VMDKACCESS_READWRITE;
4764	pExtent->fMetaDirty = false;
4765	/* Create partition table flat image. */
4766	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
4767	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
4768	true /* fCreate */));
4769	if (RT_FAILURE(rc))
4770	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new partition data file '%s'"), pExtent->pszFullname);
4771	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
4772	VMDK_SECTOR2BYTE(uPartOffset),
4773	pPart->pvPartitionData,
4774	pPart->cbData);
4775	if (RT_FAILURE(rc))
4776	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not write partition data to '%s'"), pExtent->pszFullname);
4777	uPartOffset += VMDK_BYTE2SECTOR(pPart->cbData);
4778	}
4779	else
4780	{
4781	if (pPart->pszRawDevice)
4782	{
4783	/* Set up basename for extent descr. Can't use StrDup. */
4784	size_t cbBasename = strlen(pPart->pszRawDevice) + 1;
4785	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
4786	if (!pszBasename)
4787	return VERR_NO_MEMORY;
4788	memcpy(pszBasename, pPart->pszRawDevice, cbBasename);
4789	pExtent->pszBasename = pszBasename;
4790	/* For raw disks full name is identical to base name. */
4791	pExtent->pszFullname = RTStrDup(pszBasename);
4792	if (!pExtent->pszFullname)
4793	return VERR_NO_MEMORY;
4794	pExtent->enmType = VMDKETYPE_FLAT;
4795	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
4796	pExtent->uSectorOffset = VMDK_BYTE2SECTOR(pPart->offStartInDevice);
4797	pExtent->enmAccess = (pPart->uFlags & VDISKRAW_READONLY) ? VMDKACCESS_READONLY : VMDKACCESS_READWRITE;
4798	pExtent->fMetaDirty = false;
4799	/* Open flat image, the raw partition. */
4800	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
4801	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
4802	false /* fCreate */));
4803	if (RT_FAILURE(rc))
4804	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not open raw partition file '%s'"), pExtent->pszFullname);
4805	}
4806	else
4807	{
4808	pExtent->pszBasename = NULL;
4809	pExtent->pszFullname = NULL;
4810	pExtent->enmType = VMDKETYPE_ZERO;
4811	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
4812	pExtent->uSectorOffset = 0;
4813	pExtent->enmAccess = VMDKACCESS_READWRITE;
4814	pExtent->fMetaDirty = false;
4815	}
4816	}
4817	}
4818	/* Another extent for filling up the rest of the image. */
4819	if (uStart != cbSize)
4820	{
4821	pExtent = &pImage->pExtents[cExtents++];
4822	pExtent->pszBasename = NULL;
4823	pExtent->pszFullname = NULL;
4824	pExtent->enmType = VMDKETYPE_ZERO;
4825	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize - uStart);
4826	pExtent->uSectorOffset = 0;
4827	pExtent->enmAccess = VMDKACCESS_READWRITE;
4828	pExtent->fMetaDirty = false;
4829	}
4830	}
4831	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
4832	(pRaw->uFlags & VDISKRAW_DISK) ?
4833	"fullDevice" : "partitionedDevice");
4834	if (RT_FAILURE(rc))
4835	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
4836	return rc;
4837	}
4838	/**
4839	* Internal: create a regular (i.e. file-backed) VMDK image.
4840	*/
4841	static int vmdkCreateRegularImage(PVMDKIMAGE pImage, uint64_t cbSize,
4842	unsigned uImageFlags, PVDINTERFACEPROGRESS pIfProgress,
4843	unsigned uPercentStart, unsigned uPercentSpan)
4844	{
4845	int rc = VINF_SUCCESS;
4846	unsigned cExtents = 1;
4847	uint64_t cbOffset = 0;
4848	uint64_t cbRemaining = cbSize;
4849	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
4850	{
4851	cExtents = cbSize / VMDK_2G_SPLIT_SIZE;
4852	/* Do proper extent computation: need one smaller extent if the total
4853	* size isn't evenly divisible by the split size. */
4854	if (cbSize % VMDK_2G_SPLIT_SIZE)
4855	cExtents++;
4856	}
4857	rc = vmdkCreateExtents(pImage, cExtents);
4858	if (RT_FAILURE(rc))
4859	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
4860	/* Basename strings needed for constructing the extent names. */
4861	char *pszBasenameSubstr = RTPathFilename(pImage->pszFilename);
4862	AssertPtr(pszBasenameSubstr);
4863	size_t cbBasenameSubstr = strlen(pszBasenameSubstr) + 1;
4864	/* Create separate descriptor file if necessary. */
4865	if (cExtents != 1 \|\| (uImageFlags & VD_IMAGE_FLAGS_FIXED))
4866	{
4867	rc = vmdkFileOpen(pImage, &pImage->pFile, NULL, pImage->pszFilename,
4868	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
4869	true /* fCreate */));
4870	if (RT_FAILURE(rc))
4871	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new sparse descriptor file '%s'"), pImage->pszFilename);
4872	}
4873	else
4874	pImage->pFile = NULL;
4875	/* Set up all extents. */
4876	for (unsigned i = 0; i < cExtents; i++)
4877	{
4878	PVMDKEXTENT pExtent = &pImage->pExtents[i];
4879	uint64_t cbExtent = cbRemaining;
4880	/* Set up fullname/basename for extent description. Cannot use StrDup
4881	* for basename, as it is not guaranteed that the memory can be freed
4882	* with RTMemTmpFree, which must be used as in other code paths
4883	* StrDup is not usable. */
4884	if (cExtents == 1 && !(uImageFlags & VD_IMAGE_FLAGS_FIXED))
4885	{
4886	char pszBasename = (char )RTMemTmpAlloc(cbBasenameSubstr);
4887	if (!pszBasename)
4888	return VERR_NO_MEMORY;
4889	memcpy(pszBasename, pszBasenameSubstr, cbBasenameSubstr);
4890	pExtent->pszBasename = pszBasename;
4891	}
4892	else
4893	{
4894	char *pszBasenameSuff = RTPathSuffix(pszBasenameSubstr);
4895	char *pszBasenameBase = RTStrDup(pszBasenameSubstr);
4896	RTPathStripSuffix(pszBasenameBase);
4897	char *pszTmp;
4898	size_t cbTmp;
4899	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
4900	{
4901	if (cExtents == 1)
4902	RTStrAPrintf(&pszTmp, "%s-flat%s", pszBasenameBase,
4903	pszBasenameSuff);
4904	else
4905	RTStrAPrintf(&pszTmp, "%s-f%03d%s", pszBasenameBase,
4906	i+1, pszBasenameSuff);
4907	}
4908	else
4909	RTStrAPrintf(&pszTmp, "%s-s%03d%s", pszBasenameBase, i+1,
4910	pszBasenameSuff);
4911	RTStrFree(pszBasenameBase);
4912	if (!pszTmp)
4913	return VERR_NO_STR_MEMORY;
4914	cbTmp = strlen(pszTmp) + 1;
4915	char pszBasename = (char )RTMemTmpAlloc(cbTmp);
4916	if (!pszBasename)
4917	{
4918	RTStrFree(pszTmp);
4919	return VERR_NO_MEMORY;
4920	}
4921	memcpy(pszBasename, pszTmp, cbTmp);
4922	RTStrFree(pszTmp);
4923	pExtent->pszBasename = pszBasename;
4924	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
4925	cbExtent = RT_MIN(cbRemaining, VMDK_2G_SPLIT_SIZE);
4926	}
4927	char *pszBasedirectory = RTStrDup(pImage->pszFilename);
4928	if (!pszBasedirectory)
4929	return VERR_NO_STR_MEMORY;
4930	RTPathStripFilename(pszBasedirectory);
4931	char *pszFullname = RTPathJoinA(pszBasedirectory, pExtent->pszBasename);
4932	RTStrFree(pszBasedirectory);
4933	if (!pszFullname)
4934	return VERR_NO_STR_MEMORY;
4935	pExtent->pszFullname = pszFullname;
4936	/* Create file for extent. */
4937	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
4938	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
4939	true /* fCreate */));
4940	if (RT_FAILURE(rc))
4941	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pExtent->pszFullname);
4942	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
4943	{
4944	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage, cbExtent,
4945	0 /* fFlags */, pIfProgress,
4946	uPercentStart + cbOffset * uPercentSpan / cbSize,
4947	cbExtent * uPercentSpan / cbSize);
4948	if (RT_FAILURE(rc))
4949	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
4950	}
4951	/* Place descriptor file information (where integrated). */
4952	if (cExtents == 1 && !(uImageFlags & VD_IMAGE_FLAGS_FIXED))
4953	{
4954	pExtent->uDescriptorSector = 1;
4955	pExtent->cDescriptorSectors = VMDK_BYTE2SECTOR(pImage->cbDescAlloc);
4956	/* The descriptor is part of the (only) extent. */
4957	pExtent->pDescData = pImage->pDescData;
4958	pImage->pDescData = NULL;
4959	}
4960	if (!(uImageFlags & VD_IMAGE_FLAGS_FIXED))
4961	{
4962	uint64_t cSectorsPerGDE, cSectorsPerGD;
4963	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE;
4964	pExtent->cSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64(cbExtent, _64K));
4965	pExtent->cSectorsPerGrain = VMDK_BYTE2SECTOR(_64K);
4966	pExtent->cGTEntries = 512;
4967	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
4968	pExtent->cSectorsPerGDE = cSectorsPerGDE;
4969	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
4970	cSectorsPerGD = (pExtent->cGDEntries + (512 / sizeof(uint32_t) - 1)) / (512 / sizeof(uint32_t));
4971	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
4972	{
4973	/* The spec says version is 1 for all VMDKs, but the vast
4974	* majority of streamOptimized VMDKs actually contain
4975	* version 3 - so go with the majority. Both are accepted. */
4976	pExtent->uVersion = 3;
4977	pExtent->uCompression = VMDK_COMPRESSION_DEFLATE;
4978	}
4979	}
4980	else
4981	{
4982	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
4983	pExtent->enmType = VMDKETYPE_VMFS;
4984	else
4985	pExtent->enmType = VMDKETYPE_FLAT;
4986	}
4987	pExtent->enmAccess = VMDKACCESS_READWRITE;
4988	pExtent->fUncleanShutdown = true;
4989	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbExtent);
4990	pExtent->uSectorOffset = 0;
4991	pExtent->fMetaDirty = true;
4992	if (!(uImageFlags & VD_IMAGE_FLAGS_FIXED))
4993	{
4994	/* fPreAlloc should never be false because VMware can't use such images. */
4995	rc = vmdkCreateGrainDirectory(pImage, pExtent,
4996	RT_MAX( pExtent->uDescriptorSector
4997	+ pExtent->cDescriptorSectors,
4998	1),
4999	true /* fPreAlloc */);
5000	if (RT_FAILURE(rc))
5001	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new grain directory in '%s'"), pExtent->pszFullname);
5002	}
5003	cbOffset += cbExtent;
5004	if (RT_SUCCESS(rc))
5005	vdIfProgress(pIfProgress, uPercentStart + cbOffset * uPercentSpan / cbSize);
5006	cbRemaining -= cbExtent;
5007	}
5008	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
5009	{
5010	/* VirtualBox doesn't care, but VMWare ESX freaks out if the wrong
5011	* controller type is set in an image. */
5012	rc = vmdkDescDDBSetStr(pImage, &pImage->Descriptor, "ddb.adapterType", "lsilogic");
5013	if (RT_FAILURE(rc))
5014	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set controller type to lsilogic in '%s'"), pImage->pszFilename);
5015	}
5016	const char *pszDescType = NULL;
5017	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
5018	{
5019	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
5020	pszDescType = "vmfs";
5021	else
5022	pszDescType = (cExtents == 1)
5023	? "monolithicFlat" : "twoGbMaxExtentFlat";
5024	}
5025	else
5026	{
5027	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5028	pszDescType = "streamOptimized";
5029	else
5030	{
5031	pszDescType = (cExtents == 1)
5032	? "monolithicSparse" : "twoGbMaxExtentSparse";
5033	}
5034	}
5035	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
5036	pszDescType);
5037	if (RT_FAILURE(rc))
5038	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
5039	return rc;
5040	}
5041	/**
5042	* Internal: Create a real stream optimized VMDK using only linear writes.
5043	*/
5044	static int vmdkCreateStreamImage(PVMDKIMAGE pImage, uint64_t cbSize)
5045	{
5046	int rc = vmdkCreateExtents(pImage, 1);
5047	if (RT_FAILURE(rc))
5048	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
5049	/* Basename strings needed for constructing the extent names. */
5050	const char *pszBasenameSubstr = RTPathFilename(pImage->pszFilename);
5051	AssertPtr(pszBasenameSubstr);
5052	size_t cbBasenameSubstr = strlen(pszBasenameSubstr) + 1;
5053	/* No separate descriptor file. */
5054	pImage->pFile = NULL;
5055	/* Set up all extents. */
5056	PVMDKEXTENT pExtent = &pImage->pExtents[0];
5057	/* Set up fullname/basename for extent description. Cannot use StrDup
5058	* for basename, as it is not guaranteed that the memory can be freed
5059	* with RTMemTmpFree, which must be used as in other code paths
5060	* StrDup is not usable. */
5061	char pszBasename = (char )RTMemTmpAlloc(cbBasenameSubstr);
5062	if (!pszBasename)
5063	return VERR_NO_MEMORY;
5064	memcpy(pszBasename, pszBasenameSubstr, cbBasenameSubstr);
5065	pExtent->pszBasename = pszBasename;
5066	char *pszBasedirectory = RTStrDup(pImage->pszFilename);
5067	RTPathStripFilename(pszBasedirectory);
5068	char *pszFullname = RTPathJoinA(pszBasedirectory, pExtent->pszBasename);
5069	RTStrFree(pszBasedirectory);
5070	if (!pszFullname)
5071	return VERR_NO_STR_MEMORY;
5072	pExtent->pszFullname = pszFullname;
5073	/* Create file for extent. Make it write only, no reading allowed. */
5074	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5075	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5076	true /* fCreate */)
5077	& ~RTFILE_O_READ);
5078	if (RT_FAILURE(rc))
5079	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pExtent->pszFullname);
5080	/* Place descriptor file information. */
5081	pExtent->uDescriptorSector = 1;
5082	pExtent->cDescriptorSectors = VMDK_BYTE2SECTOR(pImage->cbDescAlloc);
5083	/* The descriptor is part of the (only) extent. */
5084	pExtent->pDescData = pImage->pDescData;
5085	pImage->pDescData = NULL;
5086	uint64_t cSectorsPerGDE, cSectorsPerGD;
5087	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE;
5088	pExtent->cSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64(cbSize, _64K));
5089	pExtent->cSectorsPerGrain = VMDK_BYTE2SECTOR(_64K);
5090	pExtent->cGTEntries = 512;
5091	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
5092	pExtent->cSectorsPerGDE = cSectorsPerGDE;
5093	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
5094	cSectorsPerGD = (pExtent->cGDEntries + (512 / sizeof(uint32_t) - 1)) / (512 / sizeof(uint32_t));
5095	/* The spec says version is 1 for all VMDKs, but the vast
5096	* majority of streamOptimized VMDKs actually contain
5097	* version 3 - so go with the majority. Both are accepted. */
5098	pExtent->uVersion = 3;
5099	pExtent->uCompression = VMDK_COMPRESSION_DEFLATE;
5100	pExtent->fFooter = true;
5101	pExtent->enmAccess = VMDKACCESS_READONLY;
5102	pExtent->fUncleanShutdown = false;
5103	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
5104	pExtent->uSectorOffset = 0;
5105	pExtent->fMetaDirty = true;
5106	/* Create grain directory, without preallocating it straight away. It will
5107	* be constructed on the fly when writing out the data and written when
5108	* closing the image. The end effect is that the full grain directory is
5109	* allocated, which is a requirement of the VMDK specs. */
5110	rc = vmdkCreateGrainDirectory(pImage, pExtent, VMDK_GD_AT_END,
5111	false /* fPreAlloc */);
5112	if (RT_FAILURE(rc))
5113	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new grain directory in '%s'"), pExtent->pszFullname);
5114	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
5115	"streamOptimized");
5116	if (RT_FAILURE(rc))
5117	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
5118	return rc;
5119	}
5120	/**
5121	* Initializes the UUID fields in the DDB.
5122	*
5123	* @returns VBox status code.
5124	* @param pImage The VMDK image instance.
5125	*/
5126	static int vmdkCreateImageDdbUuidsInit(PVMDKIMAGE pImage)
5127	{
5128	int rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_IMAGE_UUID, &pImage->ImageUuid);
5129	if (RT_SUCCESS(rc))
5130	{
5131	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_PARENT_UUID, &pImage->ParentUuid);
5132	if (RT_SUCCESS(rc))
5133	{
5134	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_MODIFICATION_UUID,
5135	&pImage->ModificationUuid);
5136	if (RT_SUCCESS(rc))
5137	{
5138	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_PARENT_MODIFICATION_UUID,
5139	&pImage->ParentModificationUuid);
5140	if (RT_FAILURE(rc))
5141	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5142	N_("VMDK: error storing parent modification UUID in new descriptor in '%s'"), pImage->pszFilename);
5143	}
5144	else
5145	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5146	N_("VMDK: error storing modification UUID in new descriptor in '%s'"), pImage->pszFilename);
5147	}
5148	else
5149	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5150	N_("VMDK: error storing parent image UUID in new descriptor in '%s'"), pImage->pszFilename);
5151	}
5152	else
5153	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5154	N_("VMDK: error storing image UUID in new descriptor in '%s'"), pImage->pszFilename);
5155	return rc;
5156	}
5157	/**
5158	* Internal: The actual code for creating any VMDK variant currently in
5159	* existence on hosted environments.
5160	*/
5161	static int vmdkCreateImage(PVMDKIMAGE pImage, uint64_t cbSize,
5162	unsigned uImageFlags, const char *pszComment,
5163	PCVDGEOMETRY pPCHSGeometry,
5164	PCVDGEOMETRY pLCHSGeometry, PCRTUUID pUuid,
5165	PVDINTERFACEPROGRESS pIfProgress,
5166	unsigned uPercentStart, unsigned uPercentSpan)
5167	{
5168	pImage->uImageFlags = uImageFlags;
5169	pImage->pIfError = VDIfErrorGet(pImage->pVDIfsDisk);
5170	pImage->pIfIo = VDIfIoIntGet(pImage->pVDIfsImage);
5171	AssertPtrReturn(pImage->pIfIo, VERR_INVALID_PARAMETER);
5172	int rc = vmdkCreateDescriptor(pImage, pImage->pDescData, pImage->cbDescAlloc,
5173	&pImage->Descriptor);
5174	if (RT_SUCCESS(rc))
5175	{
5176	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
5177	{
5178	/* Raw disk image (includes raw partition). */
5179	PVDISKRAW pRaw = NULL;
5180	rc = vmdkMakeRawDescriptor(pImage, &pRaw);
5181	if (RT_FAILURE(rc))
5182	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could get raw descriptor for '%s'"), pImage->pszFilename);
5183	rc = vmdkCreateRawImage(pImage, pRaw, cbSize);
5184	vmdkRawDescFree(pRaw);
5185	}
5186	else if (uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5187	{
5188	/* Stream optimized sparse image (monolithic). */
5189	rc = vmdkCreateStreamImage(pImage, cbSize);
5190	}
5191	else
5192	{
5193	/* Regular fixed or sparse image (monolithic or split). */
5194	rc = vmdkCreateRegularImage(pImage, cbSize, uImageFlags,
5195	pIfProgress, uPercentStart,
5196	uPercentSpan * 95 / 100);
5197	}
5198	if (RT_SUCCESS(rc))
5199	{
5200	vdIfProgress(pIfProgress, uPercentStart + uPercentSpan * 98 / 100);
5201	pImage->cbSize = cbSize;
5202	for (unsigned i = 0; i < pImage->cExtents; i++)
5203	{
5204	PVMDKEXTENT pExtent = &pImage->pExtents[i];
5205	rc = vmdkDescExtInsert(pImage, &pImage->Descriptor, pExtent->enmAccess,
5206	pExtent->cNominalSectors, pExtent->enmType,
5207	pExtent->pszBasename, pExtent->uSectorOffset);
5208	if (RT_FAILURE(rc))
5209	{
5210	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not insert the extent list into descriptor in '%s'"), pImage->pszFilename);
5211	break;
5212	}
5213	}
5214	if (RT_SUCCESS(rc))
5215	vmdkDescExtRemoveDummy(pImage, &pImage->Descriptor);
5216	if ( RT_SUCCESS(rc)
5217	&& pPCHSGeometry->cCylinders != 0
5218	&& pPCHSGeometry->cHeads != 0
5219	&& pPCHSGeometry->cSectors != 0)
5220	rc = vmdkDescSetPCHSGeometry(pImage, pPCHSGeometry);
5221	if ( RT_SUCCESS(rc)
5222	&& pLCHSGeometry->cCylinders != 0
5223	&& pLCHSGeometry->cHeads != 0
5224	&& pLCHSGeometry->cSectors != 0)
5225	rc = vmdkDescSetLCHSGeometry(pImage, pLCHSGeometry);
5226	pImage->LCHSGeometry = *pLCHSGeometry;
5227	pImage->PCHSGeometry = *pPCHSGeometry;
5228	pImage->ImageUuid = *pUuid;
5229	RTUuidClear(&pImage->ParentUuid);
5230	RTUuidClear(&pImage->ModificationUuid);
5231	RTUuidClear(&pImage->ParentModificationUuid);
5232	if (RT_SUCCESS(rc))
5233	rc = vmdkCreateImageDdbUuidsInit(pImage);
5234	if (RT_SUCCESS(rc))
5235	rc = vmdkAllocateGrainTableCache(pImage);
5236	if (RT_SUCCESS(rc))
5237	{
5238	rc = vmdkSetImageComment(pImage, pszComment);
5239	if (RT_FAILURE(rc))
5240	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot set image comment in '%s'"), pImage->pszFilename);
5241	}
5242	if (RT_SUCCESS(rc))
5243	{
5244	vdIfProgress(pIfProgress, uPercentStart + uPercentSpan * 99 / 100);
5245	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5246	{
5247	/* streamOptimized is a bit special, we cannot trigger the flush
5248	* until all data has been written. So we write the necessary
5249	* information explicitly. */
5250	pImage->pExtents[0].cDescriptorSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64( pImage->Descriptor.aLines[pImage->Descriptor.cLines]
5251	- pImage->Descriptor.aLines[0], 512));
5252	rc = vmdkWriteMetaSparseExtent(pImage, &pImage->pExtents[0], 0, NULL);
5253	if (RT_SUCCESS(rc))
5254	{
5255	rc = vmdkWriteDescriptor(pImage, NULL);
5256	if (RT_FAILURE(rc))
5257	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write VMDK descriptor in '%s'"), pImage->pszFilename);
5258	}
5259	else
5260	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write VMDK header in '%s'"), pImage->pszFilename);
5261	}
5262	else
5263	rc = vmdkFlushImage(pImage, NULL);
5264	}
5265	}
5266	}
5267	else
5268	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new descriptor in '%s'"), pImage->pszFilename);
5269	if (RT_SUCCESS(rc))
5270	{
5271	PVDREGIONDESC pRegion = &pImage->RegionList.aRegions[0];
5272	pImage->RegionList.fFlags = 0;
5273	pImage->RegionList.cRegions = 1;
5274	pRegion->offRegion = 0; /* Disk start. */
5275	pRegion->cbBlock = 512;
5276	pRegion->enmDataForm = VDREGIONDATAFORM_RAW;
5277	pRegion->enmMetadataForm = VDREGIONMETADATAFORM_NONE;
5278	pRegion->cbData = 512;
5279	pRegion->cbMetadata = 0;
5280	pRegion->cRegionBlocksOrBytes = pImage->cbSize;
5281	vdIfProgress(pIfProgress, uPercentStart + uPercentSpan);
5282	}
5283	else
5284	vmdkFreeImage(pImage, rc != VERR_ALREADY_EXISTS, false /fFlush/);
5285	return rc;
5286	}
5287	/**
5288	* Internal: Update image comment.
5289	*/
5290	static int vmdkSetImageComment(PVMDKIMAGE pImage, const char *pszComment)
5291	{
5292	char *pszCommentEncoded = NULL;
5293	if (pszComment)
5294	{
5295	pszCommentEncoded = vmdkEncodeString(pszComment);
5296	if (!pszCommentEncoded)
5297	return VERR_NO_MEMORY;
5298	}
5299	int rc = vmdkDescDDBSetStr(pImage, &pImage->Descriptor,
5300	"ddb.comment", pszCommentEncoded);
5301	if (pszCommentEncoded)
5302	RTStrFree(pszCommentEncoded);
5303	if (RT_FAILURE(rc))
5304	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image comment in descriptor in '%s'"), pImage->pszFilename);
5305	return VINF_SUCCESS;
5306	}
5307	/**
5308	* Internal. Clear the grain table buffer for real stream optimized writing.
5309	*/
5310	static void vmdkStreamClearGT(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
5311	{
5312	uint32_t cCacheLines = RT_ALIGN(pExtent->cGTEntries, VMDK_GT_CACHELINE_SIZE) / VMDK_GT_CACHELINE_SIZE;
5313	for (uint32_t i = 0; i < cCacheLines; i++)
5314	memset(&pImage->pGTCache->aGTCache[i].aGTData[0], '\0',
5315	VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t));
5316	}
5317	/**
5318	* Internal. Flush the grain table buffer for real stream optimized writing.
5319	*/
5320	static int vmdkStreamFlushGT(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
5321	uint32_t uGDEntry)
5322	{
5323	int rc = VINF_SUCCESS;
5324	uint32_t cCacheLines = RT_ALIGN(pExtent->cGTEntries, VMDK_GT_CACHELINE_SIZE) / VMDK_GT_CACHELINE_SIZE;
5325	/* VMware does not write out completely empty grain tables in the case
5326	* of streamOptimized images, which according to my interpretation of
5327	* the VMDK 1.1 spec is bending the rules. Since they do it and we can
5328	* handle it without problems do it the same way and save some bytes. */
5329	bool fAllZero = true;
5330	for (uint32_t i = 0; i < cCacheLines; i++)
5331	{
5332	/* Convert the grain table to little endian in place, as it will not
5333	* be used at all after this function has been called. */
5334	uint32_t *pGTTmp = &pImage->pGTCache->aGTCache[i].aGTData[0];
5335	for (uint32_t j = 0; j < VMDK_GT_CACHELINE_SIZE; j++, pGTTmp++)
5336	if (*pGTTmp)
5337	{
5338	fAllZero = false;
5339	break;
5340	}
5341	if (!fAllZero)
5342	break;
5343	}
5344	if (fAllZero)
5345	return VINF_SUCCESS;
5346	uint64_t uFileOffset = pExtent->uAppendPosition;
5347	if (!uFileOffset)
5348	return VERR_INTERNAL_ERROR;
5349	/* Align to sector, as the previous write could have been any size. */
5350	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
5351	/* Grain table marker. */
5352	uint8_t aMarker[512];
5353	PVMDKMARKER pMarker = (PVMDKMARKER)&aMarker[0];
5354	memset(pMarker, '\0', sizeof(aMarker));
5355	pMarker->uSector = RT_H2LE_U64(VMDK_BYTE2SECTOR((uint64_t)pExtent->cGTEntries * sizeof(uint32_t)));
5356	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_GT);
5357	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
5358	aMarker, sizeof(aMarker));
5359	AssertRC(rc);
5360	uFileOffset += 512;
5361	if (!pExtent->pGD \|\| pExtent->pGD[uGDEntry])
5362	return VERR_INTERNAL_ERROR;
5363	pExtent->pGD[uGDEntry] = VMDK_BYTE2SECTOR(uFileOffset);
5364	for (uint32_t i = 0; i < cCacheLines; i++)
5365	{
5366	/* Convert the grain table to little endian in place, as it will not
5367	* be used at all after this function has been called. */
5368	uint32_t *pGTTmp = &pImage->pGTCache->aGTCache[i].aGTData[0];
5369	for (uint32_t j = 0; j < VMDK_GT_CACHELINE_SIZE; j++, pGTTmp++)
5370	pGTTmp = RT_H2LE_U32(pGTTmp);
5371	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
5372	&pImage->pGTCache->aGTCache[i].aGTData[0],
5373	VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t));
5374	uFileOffset += VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t);
5375	if (RT_FAILURE(rc))
5376	break;
5377	}
5378	Assert(!(uFileOffset % 512));
5379	pExtent->uAppendPosition = RT_ALIGN_64(uFileOffset, 512);
5380	return rc;
5381	}
5382	/**
5383	* Internal. Free all allocated space for representing an image, and optionally
5384	* delete the image from disk.
5385	*/
5386	static int vmdkFreeImage(PVMDKIMAGE pImage, bool fDelete, bool fFlush)
5387	{
5388	int rc = VINF_SUCCESS;
5389	/* Freeing a never allocated image (e.g. because the open failed) is
5390	* not signalled as an error. After all nothing bad happens. */
5391	if (pImage)
5392	{
5393	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
5394	{
5395	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5396	{
5397	/* Check if all extents are clean. */
5398	for (unsigned i = 0; i < pImage->cExtents; i++)
5399	{
5400	Assert(!pImage->pExtents[i].fUncleanShutdown);
5401	}
5402	}
5403	else
5404	{
5405	/* Mark all extents as clean. */
5406	for (unsigned i = 0; i < pImage->cExtents; i++)
5407	{
5408	if ( pImage->pExtents[i].enmType == VMDKETYPE_HOSTED_SPARSE
5409	&& pImage->pExtents[i].fUncleanShutdown)
5410	{
5411	pImage->pExtents[i].fUncleanShutdown = false;
5412	pImage->pExtents[i].fMetaDirty = true;
5413	}
5414	/* From now on it's not safe to append any more data. */
5415	pImage->pExtents[i].uAppendPosition = 0;
5416	}
5417	}
5418	}
5419	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5420	{
5421	/* No need to write any pending data if the file will be deleted
5422	* or if the new file wasn't successfully created. */
5423	if ( !fDelete && pImage->pExtents
5424	&& pImage->pExtents[0].cGTEntries
5425	&& pImage->pExtents[0].uAppendPosition)
5426	{
5427	PVMDKEXTENT pExtent = &pImage->pExtents[0];
5428	uint32_t uLastGDEntry = pExtent->uLastGrainAccess / pExtent->cGTEntries;
5429	rc = vmdkStreamFlushGT(pImage, pExtent, uLastGDEntry);
5430	AssertRC(rc);
5431	vmdkStreamClearGT(pImage, pExtent);
5432	for (uint32_t i = uLastGDEntry + 1; i < pExtent->cGDEntries; i++)
5433	{
5434	rc = vmdkStreamFlushGT(pImage, pExtent, i);
5435	AssertRC(rc);
5436	}
5437	uint64_t uFileOffset = pExtent->uAppendPosition;
5438	if (!uFileOffset)
5439	return VERR_INTERNAL_ERROR;
5440	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
5441	/* From now on it's not safe to append any more data. */
5442	pExtent->uAppendPosition = 0;
5443	/* Grain directory marker. */
5444	uint8_t aMarker[512];
5445	PVMDKMARKER pMarker = (PVMDKMARKER)&aMarker[0];
5446	memset(pMarker, '\0', sizeof(aMarker));
5447	pMarker->uSector = VMDK_BYTE2SECTOR(RT_ALIGN_64(RT_H2LE_U64((uint64_t)pExtent->cGDEntries * sizeof(uint32_t)), 512));
5448	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_GD);
5449	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
5450	aMarker, sizeof(aMarker));
5451	AssertRC(rc);
5452	uFileOffset += 512;
5453	/* Write grain directory in little endian style. The array will
5454	* not be used after this, so convert in place. */
5455	uint32_t *pGDTmp = pExtent->pGD;
5456	for (uint32_t i = 0; i < pExtent->cGDEntries; i++, pGDTmp++)
5457	pGDTmp = RT_H2LE_U32(pGDTmp);
5458	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
5459	uFileOffset, pExtent->pGD,
5460	pExtent->cGDEntries * sizeof(uint32_t));
5461	AssertRC(rc);
5462	pExtent->uSectorGD = VMDK_BYTE2SECTOR(uFileOffset);
5463	pExtent->uSectorRGD = VMDK_BYTE2SECTOR(uFileOffset);
5464	uFileOffset = RT_ALIGN_64( uFileOffset
5465	+ pExtent->cGDEntries * sizeof(uint32_t),
5466	512);
5467	/* Footer marker. */
5468	memset(pMarker, '\0', sizeof(aMarker));
5469	pMarker->uSector = VMDK_BYTE2SECTOR(512);
5470	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_FOOTER);
5471	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
5472	uFileOffset, aMarker, sizeof(aMarker));
5473	AssertRC(rc);
5474	uFileOffset += 512;
5475	rc = vmdkWriteMetaSparseExtent(pImage, pExtent, uFileOffset, NULL);
5476	AssertRC(rc);
5477	uFileOffset += 512;
5478	/* End-of-stream marker. */
5479	memset(pMarker, '\0', sizeof(aMarker));
5480	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
5481	uFileOffset, aMarker, sizeof(aMarker));
5482	AssertRC(rc);
5483	}
5484	}
5485	else if (!fDelete && fFlush)
5486	vmdkFlushImage(pImage, NULL);
5487	if (pImage->pExtents != NULL)
5488	{
5489	for (unsigned i = 0 ; i < pImage->cExtents; i++)
5490	{
5491	int rc2 = vmdkFreeExtentData(pImage, &pImage->pExtents[i], fDelete);
5492	if (RT_SUCCESS(rc))
5493	rc = rc2; /* Propogate any error when closing the file. */
5494	}
5495	RTMemFree(pImage->pExtents);
5496	pImage->pExtents = NULL;
5497	}
5498	pImage->cExtents = 0;
5499	if (pImage->pFile != NULL)
5500	{
5501	int rc2 = vmdkFileClose(pImage, &pImage->pFile, fDelete);
5502	if (RT_SUCCESS(rc))
5503	rc = rc2; /* Propogate any error when closing the file. */
5504	}
5505	int rc2 = vmdkFileCheckAllClose(pImage);
5506	if (RT_SUCCESS(rc))
5507	rc = rc2; /* Propogate any error when closing the file. */
5508	if (pImage->pGTCache)
5509	{
5510	RTMemFree(pImage->pGTCache);
5511	pImage->pGTCache = NULL;
5512	}
5513	if (pImage->pDescData)
5514	{
5515	RTMemFree(pImage->pDescData);
5516	pImage->pDescData = NULL;
5517	}
5518	}
5519	LogFlowFunc(("returns %Rrc\n", rc));
5520	return rc;
5521	}
5522	/**
5523	* Internal. Flush image data (and metadata) to disk.
5524	*/
5525	static int vmdkFlushImage(PVMDKIMAGE pImage, PVDIOCTX pIoCtx)
5526	{
5527	PVMDKEXTENT pExtent;
5528	int rc = VINF_SUCCESS;
5529	/* Update descriptor if changed. */
5530	if (pImage->Descriptor.fDirty)
5531	rc = vmdkWriteDescriptor(pImage, pIoCtx);
5532	if (RT_SUCCESS(rc))
5533	{
5534	for (unsigned i = 0; i < pImage->cExtents; i++)
5535	{
5536	pExtent = &pImage->pExtents[i];
5537	if (pExtent->pFile != NULL && pExtent->fMetaDirty)
5538	{
5539	switch (pExtent->enmType)
5540	{
5541	case VMDKETYPE_HOSTED_SPARSE:
5542	if (!pExtent->fFooter)
5543	rc = vmdkWriteMetaSparseExtent(pImage, pExtent, 0, pIoCtx);
5544	else
5545	{
5546	uint64_t uFileOffset = pExtent->uAppendPosition;
5547	/* Simply skip writing anything if the streamOptimized
5548	* image hasn't been just created. */
5549	if (!uFileOffset)
5550	break;
5551	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
5552	rc = vmdkWriteMetaSparseExtent(pImage, pExtent,
5553	uFileOffset, pIoCtx);
5554	}
5555	break;
5556	case VMDKETYPE_VMFS:
5557	case VMDKETYPE_FLAT:
5558	/* Nothing to do. */
5559	break;
5560	case VMDKETYPE_ZERO:
5561	default:
5562	AssertMsgFailed(("extent with type %d marked as dirty\n",
5563	pExtent->enmType));
5564	break;
5565	}
5566	}
5567	if (RT_FAILURE(rc))
5568	break;
5569	switch (pExtent->enmType)
5570	{
5571	case VMDKETYPE_HOSTED_SPARSE:
5572	case VMDKETYPE_VMFS:
5573	case VMDKETYPE_FLAT:
5574	/** @todo implement proper path absolute check. */
5575	if ( pExtent->pFile != NULL
5576	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
5577	&& !(pExtent->pszBasename[0] == RTPATH_SLASH))
5578	rc = vdIfIoIntFileFlush(pImage->pIfIo, pExtent->pFile->pStorage, pIoCtx,
5579	NULL, NULL);
5580	break;
5581	case VMDKETYPE_ZERO:
5582	/* No need to do anything for this extent. */
5583	break;
5584	default:
5585	AssertMsgFailed(("unknown extent type %d\n", pExtent->enmType));
5586	break;
5587	}
5588	}
5589	}
5590	return rc;
5591	}
5592	/**
5593	* Internal. Find extent corresponding to the sector number in the disk.
5594	*/
5595	static int vmdkFindExtent(PVMDKIMAGE pImage, uint64_t offSector,
5596	PVMDKEXTENT ppExtent, uint64_t puSectorInExtent)
5597	{
5598	PVMDKEXTENT pExtent = NULL;
5599	int rc = VINF_SUCCESS;
5600	for (unsigned i = 0; i < pImage->cExtents; i++)
5601	{
5602	if (offSector < pImage->pExtents[i].cNominalSectors)
5603	{
5604	pExtent = &pImage->pExtents[i];
5605	*puSectorInExtent = offSector + pImage->pExtents[i].uSectorOffset;
5606	break;
5607	}
5608	offSector -= pImage->pExtents[i].cNominalSectors;
5609	}
5610	if (pExtent)
5611	*ppExtent = pExtent;
5612	else
5613	rc = VERR_IO_SECTOR_NOT_FOUND;
5614	return rc;
5615	}
5616	/**
5617	* Internal. Hash function for placing the grain table hash entries.
5618	*/
5619	static uint32_t vmdkGTCacheHash(PVMDKGTCACHE pCache, uint64_t uSector,
5620	unsigned uExtent)
5621	{
5622	/** @todo this hash function is quite simple, maybe use a better one which
5623	* scrambles the bits better. */
5624	return (uSector + uExtent) % pCache->cEntries;
5625	}
5626	/**
5627	* Internal. Get sector number in the extent file from the relative sector
5628	* number in the extent.
5629	*/
5630	static int vmdkGetSector(PVMDKIMAGE pImage, PVDIOCTX pIoCtx,
5631	PVMDKEXTENT pExtent, uint64_t uSector,
5632	uint64_t *puExtentSector)
5633	{
5634	PVMDKGTCACHE pCache = pImage->pGTCache;
5635	uint64_t uGDIndex, uGTSector, uGTBlock;
5636	uint32_t uGTHash, uGTBlockIndex;
5637	PVMDKGTCACHEENTRY pGTCacheEntry;
5638	uint32_t aGTDataTmp[VMDK_GT_CACHELINE_SIZE];
5639	int rc;
5640	/* For newly created and readonly/sequentially opened streamOptimized
5641	* images this must be a no-op, as the grain directory is not there. */
5642	if ( ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
5643	&& pExtent->uAppendPosition)
5644	\|\| ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
5645	&& pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY
5646	&& pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
5647	{
5648	*puExtentSector = 0;
5649	return VINF_SUCCESS;
5650	}
5651	uGDIndex = uSector / pExtent->cSectorsPerGDE;
5652	if (uGDIndex >= pExtent->cGDEntries)
5653	return VERR_OUT_OF_RANGE;
5654	uGTSector = pExtent->pGD[uGDIndex];
5655	if (!uGTSector)
5656	{
5657	/* There is no grain table referenced by this grain directory
5658	* entry. So there is absolutely no data in this area. */
5659	*puExtentSector = 0;
5660	return VINF_SUCCESS;
5661	}
5662	uGTBlock = uSector / (pExtent->cSectorsPerGrain * VMDK_GT_CACHELINE_SIZE);
5663	uGTHash = vmdkGTCacheHash(pCache, uGTBlock, pExtent->uExtent);
5664	pGTCacheEntry = &pCache->aGTCache[uGTHash];
5665	if ( pGTCacheEntry->uExtent != pExtent->uExtent
5666	\|\| pGTCacheEntry->uGTBlock != uGTBlock)
5667	{
5668	/* Cache miss, fetch data from disk. */
5669	PVDMETAXFER pMetaXfer;
5670	rc = vdIfIoIntFileReadMeta(pImage->pIfIo, pExtent->pFile->pStorage,
5671	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
5672	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx, &pMetaXfer, NULL, NULL);
5673	if (RT_FAILURE(rc))
5674	return rc;
5675	/* We can release the metadata transfer immediately. */
5676	vdIfIoIntMetaXferRelease(pImage->pIfIo, pMetaXfer);
5677	pGTCacheEntry->uExtent = pExtent->uExtent;
5678	pGTCacheEntry->uGTBlock = uGTBlock;
5679	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
5680	pGTCacheEntry->aGTData[i] = RT_LE2H_U32(aGTDataTmp[i]);
5681	}
5682	uGTBlockIndex = (uSector / pExtent->cSectorsPerGrain) % VMDK_GT_CACHELINE_SIZE;
5683	uint32_t uGrainSector = pGTCacheEntry->aGTData[uGTBlockIndex];
5684	if (uGrainSector)
5685	*puExtentSector = uGrainSector + uSector % pExtent->cSectorsPerGrain;
5686	else
5687	*puExtentSector = 0;
5688	return VINF_SUCCESS;
5689	}
5690	/**
5691	* Internal. Writes the grain and also if necessary the grain tables.
5692	* Uses the grain table cache as a true grain table.
5693	*/
5694	static int vmdkStreamAllocGrain(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
5695	uint64_t uSector, PVDIOCTX pIoCtx,
5696	uint64_t cbWrite)
5697	{
5698	uint32_t uGrain;
5699	uint32_t uGDEntry, uLastGDEntry;
5700	uint32_t cbGrain = 0;
5701	uint32_t uCacheLine, uCacheEntry;
5702	const void *pData;
5703	int rc;
5704	/* Very strict requirements: always write at least one full grain, with
5705	* proper alignment. Everything else would require reading of already
5706	* written data, which we don't support for obvious reasons. The only
5707	* exception is the last grain, and only if the image size specifies
5708	* that only some portion holds data. In any case the write must be
5709	* within the image limits, no "overshoot" allowed. */
5710	if ( cbWrite == 0
5711	\|\| ( cbWrite < VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain)
5712	&& pExtent->cNominalSectors - uSector >= pExtent->cSectorsPerGrain)
5713	\|\| uSector % pExtent->cSectorsPerGrain
5714	\|\| uSector + VMDK_BYTE2SECTOR(cbWrite) > pExtent->cNominalSectors)
5715	return VERR_INVALID_PARAMETER;
5716	/* Clip write range to at most the rest of the grain. */
5717	cbWrite = RT_MIN(cbWrite, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain - uSector % pExtent->cSectorsPerGrain));
5718	/* Do not allow to go back. */
5719	uGrain = uSector / pExtent->cSectorsPerGrain;
5720	uCacheLine = uGrain % pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE;
5721	uCacheEntry = uGrain % VMDK_GT_CACHELINE_SIZE;
5722	uGDEntry = uGrain / pExtent->cGTEntries;
5723	uLastGDEntry = pExtent->uLastGrainAccess / pExtent->cGTEntries;
5724	if (uGrain < pExtent->uLastGrainAccess)
5725	return VERR_VD_VMDK_INVALID_WRITE;
5726	/* Zero byte write optimization. Since we don't tell VBoxHDD that we need
5727	* to allocate something, we also need to detect the situation ourself. */
5728	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_HONOR_ZEROES)
5729	&& vdIfIoIntIoCtxIsZero(pImage->pIfIo, pIoCtx, cbWrite, true /* fAdvance */))
5730	return VINF_SUCCESS;
5731	if (uGDEntry != uLastGDEntry)
5732	{
5733	rc = vmdkStreamFlushGT(pImage, pExtent, uLastGDEntry);
5734	if (RT_FAILURE(rc))
5735	return rc;
5736	vmdkStreamClearGT(pImage, pExtent);
5737	for (uint32_t i = uLastGDEntry + 1; i < uGDEntry; i++)
5738	{
5739	rc = vmdkStreamFlushGT(pImage, pExtent, i);
5740	if (RT_FAILURE(rc))
5741	return rc;
5742	}
5743	}
5744	uint64_t uFileOffset;
5745	uFileOffset = pExtent->uAppendPosition;
5746	if (!uFileOffset)
5747	return VERR_INTERNAL_ERROR;
5748	/* Align to sector, as the previous write could have been any size. */
5749	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
5750	/* Paranoia check: extent type, grain table buffer presence and
5751	* grain table buffer space. Also grain table entry must be clear. */
5752	if ( pExtent->enmType != VMDKETYPE_HOSTED_SPARSE
5753	\|\| !pImage->pGTCache
5754	\|\| pExtent->cGTEntries > VMDK_GT_CACHE_SIZE * VMDK_GT_CACHELINE_SIZE
5755	\|\| pImage->pGTCache->aGTCache[uCacheLine].aGTData[uCacheEntry])
5756	return VERR_INTERNAL_ERROR;
5757	/* Update grain table entry. */
5758	pImage->pGTCache->aGTCache[uCacheLine].aGTData[uCacheEntry] = VMDK_BYTE2SECTOR(uFileOffset);
5759	if (cbWrite != VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
5760	{
5761	vdIfIoIntIoCtxCopyFrom(pImage->pIfIo, pIoCtx, pExtent->pvGrain, cbWrite);
5762	memset((char *)pExtent->pvGrain + cbWrite, '\0',
5763	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain) - cbWrite);
5764	pData = pExtent->pvGrain;
5765	}
5766	else
5767	{
5768	RTSGSEG Segment;
5769	unsigned cSegments = 1;
5770	size_t cbSeg = 0;
5771	cbSeg = vdIfIoIntIoCtxSegArrayCreate(pImage->pIfIo, pIoCtx, &Segment,
5772	&cSegments, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
5773	Assert(cbSeg == VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
5774	pData = Segment.pvSeg;
5775	}
5776	rc = vmdkFileDeflateSync(pImage, pExtent, uFileOffset, pData,
5777	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
5778	uSector, &cbGrain);
5779	if (RT_FAILURE(rc))
5780	{
5781	pExtent->uGrainSectorAbs = 0;
5782	AssertRC(rc);
5783	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write compressed data block in '%s'"), pExtent->pszFullname);
5784	}
5785	pExtent->uLastGrainAccess = uGrain;
5786	pExtent->uAppendPosition += cbGrain;
5787	return rc;
5788	}
5789	/**
5790	* Internal: Updates the grain table during grain allocation.
5791	*/
5792	static int vmdkAllocGrainGTUpdate(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, PVDIOCTX pIoCtx,
5793	PVMDKGRAINALLOCASYNC pGrainAlloc)
5794	{
5795	int rc = VINF_SUCCESS;
5796	PVMDKGTCACHE pCache = pImage->pGTCache;
5797	uint32_t aGTDataTmp[VMDK_GT_CACHELINE_SIZE];
5798	uint32_t uGTHash, uGTBlockIndex;
5799	uint64_t uGTSector, uRGTSector, uGTBlock;
5800	uint64_t uSector = pGrainAlloc->uSector;
5801	PVMDKGTCACHEENTRY pGTCacheEntry;
5802	LogFlowFunc(("pImage=%#p pExtent=%#p pCache=%#p pIoCtx=%#p pGrainAlloc=%#p\n",
5803	pImage, pExtent, pCache, pIoCtx, pGrainAlloc));
5804	uGTSector = pGrainAlloc->uGTSector;
5805	uRGTSector = pGrainAlloc->uRGTSector;
5806	LogFlow(("uGTSector=%llu uRGTSector=%llu\n", uGTSector, uRGTSector));
5807	/* Update the grain table (and the cache). */
5808	uGTBlock = uSector / (pExtent->cSectorsPerGrain * VMDK_GT_CACHELINE_SIZE);
5809	uGTHash = vmdkGTCacheHash(pCache, uGTBlock, pExtent->uExtent);
5810	pGTCacheEntry = &pCache->aGTCache[uGTHash];
5811	if ( pGTCacheEntry->uExtent != pExtent->uExtent
5812	\|\| pGTCacheEntry->uGTBlock != uGTBlock)
5813	{
5814	/* Cache miss, fetch data from disk. */
5815	LogFlow(("Cache miss, fetch data from disk\n"));
5816	PVDMETAXFER pMetaXfer = NULL;
5817	rc = vdIfIoIntFileReadMeta(pImage->pIfIo, pExtent->pFile->pStorage,
5818	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
5819	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
5820	&pMetaXfer, vmdkAllocGrainComplete, pGrainAlloc);
5821	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
5822	{
5823	pGrainAlloc->cIoXfersPending++;
5824	pGrainAlloc->fGTUpdateNeeded = true;
5825	/* Leave early, we will be called again after the read completed. */
5826	LogFlowFunc(("Metadata read in progress, leaving\n"));
5827	return rc;
5828	}
5829	else if (RT_FAILURE(rc))
5830	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot read allocated grain table entry in '%s'"), pExtent->pszFullname);
5831	vdIfIoIntMetaXferRelease(pImage->pIfIo, pMetaXfer);
5832	pGTCacheEntry->uExtent = pExtent->uExtent;
5833	pGTCacheEntry->uGTBlock = uGTBlock;
5834	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
5835	pGTCacheEntry->aGTData[i] = RT_LE2H_U32(aGTDataTmp[i]);
5836	}
5837	else
5838	{
5839	/* Cache hit. Convert grain table block back to disk format, otherwise
5840	* the code below will write garbage for all but the updated entry. */
5841	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
5842	aGTDataTmp[i] = RT_H2LE_U32(pGTCacheEntry->aGTData[i]);
5843	}
5844	pGrainAlloc->fGTUpdateNeeded = false;
5845	uGTBlockIndex = (uSector / pExtent->cSectorsPerGrain) % VMDK_GT_CACHELINE_SIZE;
5846	aGTDataTmp[uGTBlockIndex] = RT_H2LE_U32(VMDK_BYTE2SECTOR(pGrainAlloc->uGrainOffset));
5847	pGTCacheEntry->aGTData[uGTBlockIndex] = VMDK_BYTE2SECTOR(pGrainAlloc->uGrainOffset);
5848	/* Update grain table on disk. */
5849	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
5850	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
5851	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
5852	vmdkAllocGrainComplete, pGrainAlloc);
5853	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
5854	pGrainAlloc->cIoXfersPending++;
5855	else if (RT_FAILURE(rc))
5856	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write updated grain table in '%s'"), pExtent->pszFullname);
5857	if (pExtent->pRGD)
5858	{
5859	/* Update backup grain table on disk. */
5860	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
5861	VMDK_SECTOR2BYTE(uRGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
5862	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
5863	vmdkAllocGrainComplete, pGrainAlloc);
5864	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
5865	pGrainAlloc->cIoXfersPending++;
5866	else if (RT_FAILURE(rc))
5867	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write updated backup grain table in '%s'"), pExtent->pszFullname);
5868	}
5869	LogFlowFunc(("leaving rc=%Rrc\n", rc));
5870	return rc;
5871	}
5872	/**
5873	* Internal - complete the grain allocation by updating disk grain table if required.
5874	*/
5875	static DECLCALLBACK(int) vmdkAllocGrainComplete(void pBackendData, PVDIOCTX pIoCtx, void pvUser, int rcReq)
5876	{
5877	RT_NOREF1(rcReq);
5878	int rc = VINF_SUCCESS;
5879	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
5880	PVMDKGRAINALLOCASYNC pGrainAlloc = (PVMDKGRAINALLOCASYNC)pvUser;
5881	LogFlowFunc(("pBackendData=%#p pIoCtx=%#p pvUser=%#p rcReq=%Rrc\n",
5882	pBackendData, pIoCtx, pvUser, rcReq));
5883	pGrainAlloc->cIoXfersPending--;
5884	if (!pGrainAlloc->cIoXfersPending && pGrainAlloc->fGTUpdateNeeded)
5885	rc = vmdkAllocGrainGTUpdate(pImage, pGrainAlloc->pExtent, pIoCtx, pGrainAlloc);
5886	if (!pGrainAlloc->cIoXfersPending)
5887	{
5888	/* Grain allocation completed. */
5889	RTMemFree(pGrainAlloc);
5890	}
5891	LogFlowFunc(("Leaving rc=%Rrc\n", rc));
5892	return rc;
5893	}
5894	/**
5895	* Internal. Allocates a new grain table (if necessary).
5896	*/
5897	static int vmdkAllocGrain(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, PVDIOCTX pIoCtx,
5898	uint64_t uSector, uint64_t cbWrite)
5899	{
5900	PVMDKGTCACHE pCache = pImage->pGTCache; NOREF(pCache);
5901	uint64_t uGDIndex, uGTSector, uRGTSector;
5902	uint64_t uFileOffset;
5903	PVMDKGRAINALLOCASYNC pGrainAlloc = NULL;
5904	int rc;
5905	LogFlowFunc(("pCache=%#p pExtent=%#p pIoCtx=%#p uSector=%llu cbWrite=%llu\n",
5906	pCache, pExtent, pIoCtx, uSector, cbWrite));
5907	pGrainAlloc = (PVMDKGRAINALLOCASYNC)RTMemAllocZ(sizeof(VMDKGRAINALLOCASYNC));
5908	if (!pGrainAlloc)
5909	return VERR_NO_MEMORY;
5910	pGrainAlloc->pExtent = pExtent;
5911	pGrainAlloc->uSector = uSector;
5912	uGDIndex = uSector / pExtent->cSectorsPerGDE;
5913	if (uGDIndex >= pExtent->cGDEntries)
5914	{
5915	RTMemFree(pGrainAlloc);
5916	return VERR_OUT_OF_RANGE;
5917	}
5918	uGTSector = pExtent->pGD[uGDIndex];
5919	if (pExtent->pRGD)
5920	uRGTSector = pExtent->pRGD[uGDIndex];
5921	else
5922	uRGTSector = 0; /*< avoid compiler warning /
5923	if (!uGTSector)
5924	{
5925	LogFlow(("Allocating new grain table\n"));
5926	/* There is no grain table referenced by this grain directory
5927	* entry. So there is absolutely no data in this area. Allocate
5928	* a new grain table and put the reference to it in the GDs. */
5929	uFileOffset = pExtent->uAppendPosition;
5930	if (!uFileOffset)
5931	{
5932	RTMemFree(pGrainAlloc);
5933	return VERR_INTERNAL_ERROR;
5934	}
5935	Assert(!(uFileOffset % 512));
5936	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
5937	uGTSector = VMDK_BYTE2SECTOR(uFileOffset);
5938	/* Normally the grain table is preallocated for hosted sparse extents
5939	* that support more than 32 bit sector numbers. So this shouldn't
5940	* ever happen on a valid extent. */
5941	if (uGTSector > UINT32_MAX)
5942	{
5943	RTMemFree(pGrainAlloc);
5944	return VERR_VD_VMDK_INVALID_HEADER;
5945	}
5946	/* Write grain table by writing the required number of grain table
5947	* cache chunks. Allocate memory dynamically here or we flood the
5948	* metadata cache with very small entries. */
5949	size_t cbGTDataTmp = pExtent->cGTEntries * sizeof(uint32_t);
5950	uint32_t paGTDataTmp = (uint32_t )RTMemTmpAllocZ(cbGTDataTmp);
5951	if (!paGTDataTmp)
5952	{
5953	RTMemFree(pGrainAlloc);
5954	return VERR_NO_MEMORY;
5955	}
5956	memset(paGTDataTmp, '\0', cbGTDataTmp);
5957	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
5958	VMDK_SECTOR2BYTE(uGTSector),
5959	paGTDataTmp, cbGTDataTmp, pIoCtx,
5960	vmdkAllocGrainComplete, pGrainAlloc);
5961	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
5962	pGrainAlloc->cIoXfersPending++;
5963	else if (RT_FAILURE(rc))
5964	{
5965	RTMemTmpFree(paGTDataTmp);
5966	RTMemFree(pGrainAlloc);
5967	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write grain table allocation in '%s'"), pExtent->pszFullname);
5968	}
5969	pExtent->uAppendPosition = RT_ALIGN_64( pExtent->uAppendPosition
5970	+ cbGTDataTmp, 512);
5971	if (pExtent->pRGD)
5972	{
5973	AssertReturn(!uRGTSector, VERR_VD_VMDK_INVALID_HEADER);
5974	uFileOffset = pExtent->uAppendPosition;
5975	if (!uFileOffset)
5976	return VERR_INTERNAL_ERROR;
5977	Assert(!(uFileOffset % 512));
5978	uRGTSector = VMDK_BYTE2SECTOR(uFileOffset);
5979	/* Normally the redundant grain table is preallocated for hosted
5980	* sparse extents that support more than 32 bit sector numbers. So
5981	* this shouldn't ever happen on a valid extent. */
5982	if (uRGTSector > UINT32_MAX)
5983	{
5984	RTMemTmpFree(paGTDataTmp);
5985	return VERR_VD_VMDK_INVALID_HEADER;
5986	}
5987	/* Write grain table by writing the required number of grain table
5988	* cache chunks. Allocate memory dynamically here or we flood the
5989	* metadata cache with very small entries. */
5990	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
5991	VMDK_SECTOR2BYTE(uRGTSector),
5992	paGTDataTmp, cbGTDataTmp, pIoCtx,
5993	vmdkAllocGrainComplete, pGrainAlloc);
5994	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
5995	pGrainAlloc->cIoXfersPending++;
5996	else if (RT_FAILURE(rc))
5997	{
5998	RTMemTmpFree(paGTDataTmp);
5999	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write backup grain table allocation in '%s'"), pExtent->pszFullname);
6000	}
6001	pExtent->uAppendPosition = pExtent->uAppendPosition + cbGTDataTmp;
6002	}
6003	RTMemTmpFree(paGTDataTmp);
6004	/* Update the grain directory on disk (doing it before writing the
6005	* grain table will result in a garbled extent if the operation is
6006	* aborted for some reason. Otherwise the worst that can happen is
6007	* some unused sectors in the extent. */
6008	uint32_t uGTSectorLE = RT_H2LE_U64(uGTSector);
6009	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6010	VMDK_SECTOR2BYTE(pExtent->uSectorGD) + uGDIndex * sizeof(uGTSectorLE),
6011	&uGTSectorLE, sizeof(uGTSectorLE), pIoCtx,
6012	vmdkAllocGrainComplete, pGrainAlloc);
6013	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6014	pGrainAlloc->cIoXfersPending++;
6015	else if (RT_FAILURE(rc))
6016	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write grain directory entry in '%s'"), pExtent->pszFullname);
6017	if (pExtent->pRGD)
6018	{
6019	uint32_t uRGTSectorLE = RT_H2LE_U64(uRGTSector);
6020	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6021	VMDK_SECTOR2BYTE(pExtent->uSectorRGD) + uGDIndex * sizeof(uGTSectorLE),
6022	&uRGTSectorLE, sizeof(uRGTSectorLE), pIoCtx,
6023	vmdkAllocGrainComplete, pGrainAlloc);
6024	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6025	pGrainAlloc->cIoXfersPending++;
6026	else if (RT_FAILURE(rc))
6027	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write backup grain directory entry in '%s'"), pExtent->pszFullname);
6028	}
6029	/* As the final step update the in-memory copy of the GDs. */
6030	pExtent->pGD[uGDIndex] = uGTSector;
6031	if (pExtent->pRGD)
6032	pExtent->pRGD[uGDIndex] = uRGTSector;
6033	}
6034	LogFlow(("uGTSector=%llu uRGTSector=%llu\n", uGTSector, uRGTSector));
6035	pGrainAlloc->uGTSector = uGTSector;
6036	pGrainAlloc->uRGTSector = uRGTSector;
6037	uFileOffset = pExtent->uAppendPosition;
6038	if (!uFileOffset)
6039	return VERR_INTERNAL_ERROR;
6040	Assert(!(uFileOffset % 512));
6041	pGrainAlloc->uGrainOffset = uFileOffset;
6042	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6043	{
6044	AssertMsgReturn(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
6045	("Accesses to stream optimized images must be synchronous\n"),
6046	VERR_INVALID_STATE);
6047	if (cbWrite != VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
6048	return vdIfError(pImage->pIfError, VERR_INTERNAL_ERROR, RT_SRC_POS, N_("VMDK: not enough data for a compressed data block in '%s'"), pExtent->pszFullname);
6049	/* Invalidate cache, just in case some code incorrectly allows mixing
6050	* of reads and writes. Normally shouldn't be needed. */
6051	pExtent->uGrainSectorAbs = 0;
6052	/* Write compressed data block and the markers. */
6053	uint32_t cbGrain = 0;
6054	size_t cbSeg = 0;
6055	RTSGSEG Segment;
6056	unsigned cSegments = 1;
6057	cbSeg = vdIfIoIntIoCtxSegArrayCreate(pImage->pIfIo, pIoCtx, &Segment,
6058	&cSegments, cbWrite);
6059	Assert(cbSeg == cbWrite);
6060	rc = vmdkFileDeflateSync(pImage, pExtent, uFileOffset,
6061	Segment.pvSeg, cbWrite, uSector, &cbGrain);
6062	if (RT_FAILURE(rc))
6063	{
6064	AssertRC(rc);
6065	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write allocated compressed data block in '%s'"), pExtent->pszFullname);
6066	}
6067	pExtent->uLastGrainAccess = uSector / pExtent->cSectorsPerGrain;
6068	pExtent->uAppendPosition += cbGrain;
6069	}
6070	else
6071	{
6072	/* Write the data. Always a full grain, or we're in big trouble. */
6073	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
6074	uFileOffset, pIoCtx, cbWrite,
6075	vmdkAllocGrainComplete, pGrainAlloc);
6076	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6077	pGrainAlloc->cIoXfersPending++;
6078	else if (RT_FAILURE(rc))
6079	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write allocated data block in '%s'"), pExtent->pszFullname);
6080	pExtent->uAppendPosition += cbWrite;
6081	}
6082	rc = vmdkAllocGrainGTUpdate(pImage, pExtent, pIoCtx, pGrainAlloc);
6083	if (!pGrainAlloc->cIoXfersPending)
6084	{
6085	/* Grain allocation completed. */
6086	RTMemFree(pGrainAlloc);
6087	}
6088	LogFlowFunc(("leaving rc=%Rrc\n", rc));
6089	return rc;
6090	}
6091	/**
6092	* Internal. Reads the contents by sequentially going over the compressed
6093	* grains (hoping that they are in sequence).
6094	*/
6095	static int vmdkStreamReadSequential(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
6096	uint64_t uSector, PVDIOCTX pIoCtx,
6097	uint64_t cbRead)
6098	{
6099	int rc;
6100	LogFlowFunc(("pImage=%#p pExtent=%#p uSector=%llu pIoCtx=%#p cbRead=%llu\n",
6101	pImage, pExtent, uSector, pIoCtx, cbRead));
6102	AssertMsgReturn(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
6103	("Async I/O not supported for sequential stream optimized images\n"),
6104	VERR_INVALID_STATE);
6105	/* Do not allow to go back. */
6106	uint32_t uGrain = uSector / pExtent->cSectorsPerGrain;
6107	if (uGrain < pExtent->uLastGrainAccess)
6108	return VERR_VD_VMDK_INVALID_STATE;
6109	pExtent->uLastGrainAccess = uGrain;
6110	/* After a previous error do not attempt to recover, as it would need
6111	* seeking (in the general case backwards which is forbidden). */
6112	if (!pExtent->uGrainSectorAbs)
6113	return VERR_VD_VMDK_INVALID_STATE;
6114	/* Check if we need to read something from the image or if what we have
6115	* in the buffer is good to fulfill the request. */
6116	if (!pExtent->cbGrainStreamRead \|\| uGrain > pExtent->uGrain)
6117	{
6118	uint32_t uGrainSectorAbs = pExtent->uGrainSectorAbs
6119	+ VMDK_BYTE2SECTOR(pExtent->cbGrainStreamRead);
6120	/* Get the marker from the next data block - and skip everything which
6121	* is not a compressed grain. If it's a compressed grain which is for
6122	* the requested sector (or after), read it. */
6123	VMDKMARKER Marker;
6124	do
6125	{
6126	RT_ZERO(Marker);
6127	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
6128	VMDK_SECTOR2BYTE(uGrainSectorAbs),
6129	&Marker, RT_UOFFSETOF(VMDKMARKER, uType));
6130	if (RT_FAILURE(rc))
6131	return rc;
6132	Marker.uSector = RT_LE2H_U64(Marker.uSector);
6133	Marker.cbSize = RT_LE2H_U32(Marker.cbSize);
6134	if (Marker.cbSize == 0)
6135	{
6136	/* A marker for something else than a compressed grain. */
6137	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
6138	VMDK_SECTOR2BYTE(uGrainSectorAbs)
6139	+ RT_UOFFSETOF(VMDKMARKER, uType),
6140	&Marker.uType, sizeof(Marker.uType));
6141	if (RT_FAILURE(rc))
6142	return rc;
6143	Marker.uType = RT_LE2H_U32(Marker.uType);
6144	switch (Marker.uType)
6145	{
6146	case VMDK_MARKER_EOS:
6147	uGrainSectorAbs++;
6148	/* Read (or mostly skip) to the end of file. Uses the
6149	* Marker (LBA sector) as it is unused anyway. This
6150	* makes sure that really everything is read in the
6151	* success case. If this read fails it means the image
6152	* is truncated, but this is harmless so ignore. */
6153	vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
6154	VMDK_SECTOR2BYTE(uGrainSectorAbs)
6155	+ 511,
6156	&Marker.uSector, 1);
6157	break;
6158	case VMDK_MARKER_GT:
6159	uGrainSectorAbs += 1 + VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
6160	break;
6161	case VMDK_MARKER_GD:
6162	uGrainSectorAbs += 1 + VMDK_BYTE2SECTOR(RT_ALIGN(pExtent->cGDEntries * sizeof(uint32_t), 512));
6163	break;
6164	case VMDK_MARKER_FOOTER:
6165	uGrainSectorAbs += 2;
6166	break;
6167	case VMDK_MARKER_UNSPECIFIED:
6168	/* Skip over the contents of the unspecified marker
6169	* type 4 which exists in some vSphere created files. */
6170	/** @todo figure out what the payload means. */
6171	uGrainSectorAbs += 1;
6172	break;
6173	default:
6174	AssertMsgFailed(("VMDK: corrupted marker, type=%#x\n", Marker.uType));
6175	pExtent->uGrainSectorAbs = 0;
6176	return VERR_VD_VMDK_INVALID_STATE;
6177	}
6178	pExtent->cbGrainStreamRead = 0;
6179	}
6180	else
6181	{
6182	/* A compressed grain marker. If it is at/after what we're
6183	* interested in read and decompress data. */
6184	if (uSector > Marker.uSector + pExtent->cSectorsPerGrain)
6185	{
6186	uGrainSectorAbs += VMDK_BYTE2SECTOR(RT_ALIGN(Marker.cbSize + RT_UOFFSETOF(VMDKMARKER, uType), 512));
6187	continue;
6188	}
6189	uint64_t uLBA = 0;
6190	uint32_t cbGrainStreamRead = 0;
6191	rc = vmdkFileInflateSync(pImage, pExtent,
6192	VMDK_SECTOR2BYTE(uGrainSectorAbs),
6193	pExtent->pvGrain,
6194	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
6195	&Marker, &uLBA, &cbGrainStreamRead);
6196	if (RT_FAILURE(rc))
6197	{
6198	pExtent->uGrainSectorAbs = 0;
6199	return rc;
6200	}
6201	if ( pExtent->uGrain
6202	&& uLBA / pExtent->cSectorsPerGrain <= pExtent->uGrain)
6203	{
6204	pExtent->uGrainSectorAbs = 0;
6205	return VERR_VD_VMDK_INVALID_STATE;
6206	}
6207	pExtent->uGrain = uLBA / pExtent->cSectorsPerGrain;
6208	pExtent->cbGrainStreamRead = cbGrainStreamRead;
6209	break;
6210	}
6211	} while (Marker.uType != VMDK_MARKER_EOS);
6212	pExtent->uGrainSectorAbs = uGrainSectorAbs;
6213	if (!pExtent->cbGrainStreamRead && Marker.uType == VMDK_MARKER_EOS)
6214	{
6215	pExtent->uGrain = UINT32_MAX;
6216	/* Must set a non-zero value for pExtent->cbGrainStreamRead or
6217	* the next read would try to get more data, and we're at EOF. */
6218	pExtent->cbGrainStreamRead = 1;
6219	}
6220	}
6221	if (pExtent->uGrain > uSector / pExtent->cSectorsPerGrain)
6222	{
6223	/* The next data block we have is not for this area, so just return
6224	* that there is no data. */
6225	LogFlowFunc(("returns VERR_VD_BLOCK_FREE\n"));
6226	return VERR_VD_BLOCK_FREE;
6227	}
6228	uint32_t uSectorInGrain = uSector % pExtent->cSectorsPerGrain;
6229	vdIfIoIntIoCtxCopyTo(pImage->pIfIo, pIoCtx,
6230	(uint8_t *)pExtent->pvGrain + VMDK_SECTOR2BYTE(uSectorInGrain),
6231	cbRead);
6232	LogFlowFunc(("returns VINF_SUCCESS\n"));
6233	return VINF_SUCCESS;
6234	}
6235	/**
6236	* Replaces a fragment of a string with the specified string.
6237	*
6238	* @returns Pointer to the allocated UTF-8 string.
6239	* @param pszWhere UTF-8 string to search in.
6240	* @param pszWhat UTF-8 string to search for.
6241	* @param pszByWhat UTF-8 string to replace the found string with.
6242	*
6243	* @note r=bird: This is only used by vmdkRenameWorker(). The first use is
6244	* for updating the base name in the descriptor, the second is for
6245	* generating new filenames for extents. This code borked when
6246	* RTPathAbs started correcting the driver letter case on windows,
6247	* when strstr failed because the pExtent->pszFullname was not
6248	* subjected to RTPathAbs but while pExtent->pszFullname was. I fixed
6249	* this by apply RTPathAbs to the places it wasn't applied.
6250	*
6251	* However, this highlights some undocumented ASSUMPTIONS as well as
6252	* terrible short commings of the approach.
6253	*
6254	* Given the right filename, it may also screw up the descriptor. Take
6255	* the descriptor text 'RW 2048 SPARSE "Test0.vmdk"' for instance,
6256	* we'll be asked to replace "Test0" with something, no problem. No,
6257	* imagine 'RW 2048 SPARSE "SPARSE.vmdk"', 'RW 2048 SPARSE "RW.vmdk"'
6258	* or 'RW 2048 SPARSE "2048.vmdk"', and the strstr approach falls on
6259	* its bum. The descriptor string must be parsed and reconstructed,
6260	* the lazy strstr approach doesn't cut it.
6261	*
6262	* I'm also curious as to what would be the correct escaping of '"' in
6263	* the file name and how that is supposed to be handled, because it
6264	* needs to be or such names must be rejected in several places (maybe
6265	* they are, I didn't check).
6266	*
6267	* When this function is used to replace the start of a path, I think
6268	* the assumption from the prep/setup code is that we kind of knows
6269	* what we're working on (I could be wrong). However, using strstr
6270	* instead of strncmp/RTStrNICmp makes no sense and isn't future proof.
6271	* Especially on unix systems, weird stuff could happen if someone
6272	* unwittingly tinkers with the prep/setup code. What should really be
6273	* done here is using a new RTPathStartEx function that (via flags)
6274	* allows matching partial final component and returns the length of
6275	* what it matched up (in case it skipped slashes and '.' components).
6276	*
6277	*/
6278	static char vmdkStrReplace(const char pszWhere, const char *pszWhat,
6279	const char *pszByWhat)
6280	{
6281	AssertPtr(pszWhere);
6282	AssertPtr(pszWhat);
6283	AssertPtr(pszByWhat);
6284	const char *pszFoundStr = strstr(pszWhere, pszWhat);
6285	if (!pszFoundStr)
6286	{
6287	LogFlowFunc(("Failed to find '%s' in '%s'!\n", pszWhat, pszWhere));
6288	return NULL;
6289	}
6290	size_t cbFinal = strlen(pszWhere) + 1 + strlen(pszByWhat) - strlen(pszWhat);
6291	char *pszNewStr = RTStrAlloc(cbFinal);
6292	if (pszNewStr)
6293	{
6294	char *pszTmp = pszNewStr;
6295	memcpy(pszTmp, pszWhere, pszFoundStr - pszWhere);
6296	pszTmp += pszFoundStr - pszWhere;
6297	memcpy(pszTmp, pszByWhat, strlen(pszByWhat));
6298	pszTmp += strlen(pszByWhat);
6299	strcpy(pszTmp, pszFoundStr + strlen(pszWhat));
6300	}
6301	return pszNewStr;
6302	}
6303	/** @copydoc VDIMAGEBACKEND::pfnProbe */
6304	static DECLCALLBACK(int) vmdkProbe(const char *pszFilename, PVDINTERFACE pVDIfsDisk,
6305	PVDINTERFACE pVDIfsImage, VDTYPE enmDesiredType, VDTYPE *penmType)
6306	{
6307	RT_NOREF(enmDesiredType);
6308	LogFlowFunc(("pszFilename=\"%s\" pVDIfsDisk=%#p pVDIfsImage=%#p penmType=%#p\n",
6309	pszFilename, pVDIfsDisk, pVDIfsImage, penmType));
6310	AssertReturn((VALID_PTR(pszFilename) && *pszFilename), VERR_INVALID_PARAMETER);
6311	int rc = VINF_SUCCESS;
6312	PVMDKIMAGE pImage = (PVMDKIMAGE)RTMemAllocZ(RT_UOFFSETOF(VMDKIMAGE, RegionList.aRegions[1]));
6313	if (RT_LIKELY(pImage))
6314	{
6315	pImage->pszFilename = pszFilename;
6316	pImage->pFile = NULL;
6317	pImage->pExtents = NULL;
6318	pImage->pFiles = NULL;
6319	pImage->pGTCache = NULL;
6320	pImage->pDescData = NULL;
6321	pImage->pVDIfsDisk = pVDIfsDisk;
6322	pImage->pVDIfsImage = pVDIfsImage;
6323	/** @todo speed up this test open (VD_OPEN_FLAGS_INFO) by skipping as
6324	* much as possible in vmdkOpenImage. */
6325	rc = vmdkOpenImage(pImage, VD_OPEN_FLAGS_INFO \| VD_OPEN_FLAGS_READONLY);
6326	vmdkFreeImage(pImage, false, false /fFlush/);
6327	RTMemFree(pImage);
6328	if (RT_SUCCESS(rc))
6329	*penmType = VDTYPE_HDD;
6330	}
6331	else
6332	rc = VERR_NO_MEMORY;
6333	LogFlowFunc(("returns %Rrc\n", rc));
6334	return rc;
6335	}
6336	/** @copydoc VDIMAGEBACKEND::pfnOpen */
6337	static DECLCALLBACK(int) vmdkOpen(const char *pszFilename, unsigned uOpenFlags,
6338	PVDINTERFACE pVDIfsDisk, PVDINTERFACE pVDIfsImage,
6339	VDTYPE enmType, void **ppBackendData)
6340	{
6341	RT_NOREF1(enmType); /*< @todo r=klaus make use of the type info. /
6342	LogFlowFunc(("pszFilename=\"%s\" uOpenFlags=%#x pVDIfsDisk=%#p pVDIfsImage=%#p enmType=%u ppBackendData=%#p\n",
6343	pszFilename, uOpenFlags, pVDIfsDisk, pVDIfsImage, enmType, ppBackendData));
6344	int rc;
6345	/* Check open flags. All valid flags are supported. */
6346	AssertReturn(!(uOpenFlags & ~VD_OPEN_FLAGS_MASK), VERR_INVALID_PARAMETER);
6347	AssertReturn((VALID_PTR(pszFilename) && *pszFilename), VERR_INVALID_PARAMETER);
6348	PVMDKIMAGE pImage = (PVMDKIMAGE)RTMemAllocZ(RT_UOFFSETOF(VMDKIMAGE, RegionList.aRegions[1]));
6349	if (RT_LIKELY(pImage))
6350	{
6351	pImage->pszFilename = pszFilename;
6352	pImage->pFile = NULL;
6353	pImage->pExtents = NULL;
6354	pImage->pFiles = NULL;
6355	pImage->pGTCache = NULL;
6356	pImage->pDescData = NULL;
6357	pImage->pVDIfsDisk = pVDIfsDisk;
6358	pImage->pVDIfsImage = pVDIfsImage;
6359	rc = vmdkOpenImage(pImage, uOpenFlags);
6360	if (RT_SUCCESS(rc))
6361	*ppBackendData = pImage;
6362	else
6363	RTMemFree(pImage);
6364	}
6365	else
6366	rc = VERR_NO_MEMORY;
6367	LogFlowFunc(("returns %Rrc (pBackendData=%#p)\n", rc, *ppBackendData));
6368	return rc;
6369	}
6370	/** @copydoc VDIMAGEBACKEND::pfnCreate */
6371	static DECLCALLBACK(int) vmdkCreate(const char *pszFilename, uint64_t cbSize,
6372	unsigned uImageFlags, const char *pszComment,
6373	PCVDGEOMETRY pPCHSGeometry, PCVDGEOMETRY pLCHSGeometry,
6374	PCRTUUID pUuid, unsigned uOpenFlags,
6375	unsigned uPercentStart, unsigned uPercentSpan,
6376	PVDINTERFACE pVDIfsDisk, PVDINTERFACE pVDIfsImage,
6377	PVDINTERFACE pVDIfsOperation, VDTYPE enmType,
6378	void **ppBackendData)
6379	{
6380	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",
6381	pszFilename, cbSize, uImageFlags, pszComment, pPCHSGeometry, pLCHSGeometry, pUuid, uOpenFlags, uPercentStart, uPercentSpan, pVDIfsDisk, pVDIfsImage, pVDIfsOperation, enmType, ppBackendData));
6382	int rc;
6383	/* Check the VD container type and image flags. */
6384	if ( enmType != VDTYPE_HDD
6385	\|\| (uImageFlags & ~VD_VMDK_IMAGE_FLAGS_MASK) != 0)
6386	return VERR_VD_INVALID_TYPE;
6387	/* Check size. Maximum 256TB-64K for sparse images, otherwise unlimited. */
6388	if ( !(uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
6389	&& ( !cbSize
6390	\|\| (!(uImageFlags & VD_IMAGE_FLAGS_FIXED) && cbSize >= _1T * 256 - _64K)))
6391	return VERR_VD_INVALID_SIZE;
6392	/* Check image flags for invalid combinations. */
6393	if ( (uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6394	&& (uImageFlags & ~(VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED \| VD_IMAGE_FLAGS_DIFF)))
6395	return VERR_INVALID_PARAMETER;
6396	/* Check open flags. All valid flags are supported. */
6397	AssertReturn(!(uOpenFlags & ~VD_OPEN_FLAGS_MASK), VERR_INVALID_PARAMETER);
6398	AssertReturn( VALID_PTR(pszFilename)
6399	&& *pszFilename
6400	&& VALID_PTR(pPCHSGeometry)
6401	&& VALID_PTR(pLCHSGeometry)
6402	&& !( uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX
6403	&& !(uImageFlags & VD_IMAGE_FLAGS_FIXED)),
6404	VERR_INVALID_PARAMETER);
6405	PVMDKIMAGE pImage = (PVMDKIMAGE)RTMemAllocZ(RT_UOFFSETOF(VMDKIMAGE, RegionList.aRegions[1]));
6406	if (RT_LIKELY(pImage))
6407	{
6408	PVDINTERFACEPROGRESS pIfProgress = VDIfProgressGet(pVDIfsOperation);
6409	pImage->pszFilename = pszFilename;
6410	pImage->pFile = NULL;
6411	pImage->pExtents = NULL;
6412	pImage->pFiles = NULL;
6413	pImage->pGTCache = NULL;
6414	pImage->pDescData = NULL;
6415	pImage->pVDIfsDisk = pVDIfsDisk;
6416	pImage->pVDIfsImage = pVDIfsImage;
6417	/* Descriptors for split images can be pretty large, especially if the
6418	* filename is long. So prepare for the worst, and allocate quite some
6419	* memory for the descriptor in this case. */
6420	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
6421	pImage->cbDescAlloc = VMDK_SECTOR2BYTE(200);
6422	else
6423	pImage->cbDescAlloc = VMDK_SECTOR2BYTE(20);
6424	pImage->pDescData = (char *)RTMemAllocZ(pImage->cbDescAlloc);
6425	if (RT_LIKELY(pImage->pDescData))
6426	{
6427	rc = vmdkCreateImage(pImage, cbSize, uImageFlags, pszComment,
6428	pPCHSGeometry, pLCHSGeometry, pUuid,
6429	pIfProgress, uPercentStart, uPercentSpan);
6430	if (RT_SUCCESS(rc))
6431	{
6432	/* So far the image is opened in read/write mode. Make sure the
6433	* image is opened in read-only mode if the caller requested that. */
6434	if (uOpenFlags & VD_OPEN_FLAGS_READONLY)
6435	{
6436	vmdkFreeImage(pImage, false, true /fFlush/);
6437	rc = vmdkOpenImage(pImage, uOpenFlags);
6438	}
6439	if (RT_SUCCESS(rc))
6440	*ppBackendData = pImage;
6441	}
6442	if (RT_FAILURE(rc))
6443	RTMemFree(pImage->pDescData);
6444	}
6445	else
6446	rc = VERR_NO_MEMORY;
6447	if (RT_FAILURE(rc))
6448	RTMemFree(pImage);
6449	}
6450	else
6451	rc = VERR_NO_MEMORY;
6452	LogFlowFunc(("returns %Rrc (pBackendData=%#p)\n", rc, *ppBackendData));
6453	return rc;
6454	}
6455	/**
6456	* Prepares the state for renaming a VMDK image, setting up the state and allocating
6457	* memory.
6458	*
6459	* @returns VBox status code.
6460	* @param pImage VMDK image instance.
6461	* @param pRenameState The state to initialize.
6462	* @param pszFilename The new filename.
6463	*/
6464	static int vmdkRenameStatePrepare(PVMDKIMAGE pImage, PVMDKRENAMESTATE pRenameState, const char *pszFilename)
6465	{
6466	AssertReturn(RTPathFilename(pszFilename) != NULL, VERR_INVALID_PARAMETER);
6467	int rc = VINF_SUCCESS;
6468	memset(&pRenameState->DescriptorCopy, 0, sizeof(pRenameState->DescriptorCopy));
6469	/*
6470	* Allocate an array to store both old and new names of renamed files
6471	* in case we have to roll back the changes. Arrays are initialized
6472	* with zeros. We actually save stuff when and if we change it.
6473	*/
6474	pRenameState->cExtents = pImage->cExtents;
6475	pRenameState->apszOldName = (char *)RTMemTmpAllocZ((pRenameState->cExtents + 1) sizeof(char *));
6476	pRenameState->apszNewName = (char *)RTMemTmpAllocZ((pRenameState->cExtents + 1) sizeof(char *));
6477	pRenameState->apszNewLines = (char *)RTMemTmpAllocZ(pRenameState->cExtents sizeof(char *));
6478	if ( pRenameState->apszOldName
6479	&& pRenameState->apszNewName
6480	&& pRenameState->apszNewLines)
6481	{
6482	/* Save the descriptor size and position. */
6483	if (pImage->pDescData)
6484	{
6485	/* Separate descriptor file. */
6486	pRenameState->fEmbeddedDesc = false;
6487	}
6488	else
6489	{
6490	/* Embedded descriptor file. */
6491	pRenameState->ExtentCopy = pImage->pExtents[0];
6492	pRenameState->fEmbeddedDesc = true;
6493	}
6494	/* Save the descriptor content. */
6495	pRenameState->DescriptorCopy.cLines = pImage->Descriptor.cLines;
6496	for (unsigned i = 0; i < pRenameState->DescriptorCopy.cLines; i++)
6497	{
6498	pRenameState->DescriptorCopy.aLines[i] = RTStrDup(pImage->Descriptor.aLines[i]);
6499	if (!pRenameState->DescriptorCopy.aLines[i])
6500	{
6501	rc = VERR_NO_MEMORY;
6502	break;
6503	}
6504	}
6505	if (RT_SUCCESS(rc))
6506	{
6507	/* Prepare both old and new base names used for string replacement. */
6508	pRenameState->pszNewBaseName = RTStrDup(RTPathFilename(pszFilename));
6509	AssertReturn(pRenameState->pszNewBaseName, VERR_NO_STR_MEMORY);
6510	RTPathStripSuffix(pRenameState->pszNewBaseName);
6511	pRenameState->pszOldBaseName = RTStrDup(RTPathFilename(pImage->pszFilename));
6512	AssertReturn(pRenameState->pszOldBaseName, VERR_NO_STR_MEMORY);
6513	RTPathStripSuffix(pRenameState->pszOldBaseName);
6514	/* Prepare both old and new full names used for string replacement.
6515	Note! Must abspath the stuff here, so the strstr weirdness later in
6516	the renaming process get a match against abspath'ed extent paths.
6517	See RTPathAbsDup call in vmdkDescriptorReadSparse(). */
6518	pRenameState->pszNewFullName = RTPathAbsDup(pszFilename);
6519	AssertReturn(pRenameState->pszNewFullName, VERR_NO_STR_MEMORY);
6520	RTPathStripSuffix(pRenameState->pszNewFullName);
6521	pRenameState->pszOldFullName = RTPathAbsDup(pImage->pszFilename);
6522	AssertReturn(pRenameState->pszOldFullName, VERR_NO_STR_MEMORY);
6523	RTPathStripSuffix(pRenameState->pszOldFullName);
6524	/* Save the old name for easy access to the old descriptor file. */
6525	pRenameState->pszOldDescName = RTStrDup(pImage->pszFilename);
6526	AssertReturn(pRenameState->pszOldDescName, VERR_NO_STR_MEMORY);
6527	/* Save old image name. */
6528	pRenameState->pszOldImageName = pImage->pszFilename;
6529	}
6530	}
6531	else
6532	rc = VERR_NO_TMP_MEMORY;
6533	return rc;
6534	}
6535	/**
6536	* Destroys the given rename state, freeing all allocated memory.
6537	*
6538	* @returns nothing.
6539	* @param pRenameState The rename state to destroy.
6540	*/
6541	static void vmdkRenameStateDestroy(PVMDKRENAMESTATE pRenameState)
6542	{
6543	for (unsigned i = 0; i < pRenameState->DescriptorCopy.cLines; i++)
6544	if (pRenameState->DescriptorCopy.aLines[i])
6545	RTStrFree(pRenameState->DescriptorCopy.aLines[i]);
6546	if (pRenameState->apszOldName)
6547	{
6548	for (unsigned i = 0; i <= pRenameState->cExtents; i++)
6549	if (pRenameState->apszOldName[i])
6550	RTStrFree(pRenameState->apszOldName[i]);
6551	RTMemTmpFree(pRenameState->apszOldName);
6552	}
6553	if (pRenameState->apszNewName)
6554	{
6555	for (unsigned i = 0; i <= pRenameState->cExtents; i++)
6556	if (pRenameState->apszNewName[i])
6557	RTStrFree(pRenameState->apszNewName[i]);
6558	RTMemTmpFree(pRenameState->apszNewName);
6559	}
6560	if (pRenameState->apszNewLines)
6561	{
6562	for (unsigned i = 0; i < pRenameState->cExtents; i++)
6563	if (pRenameState->apszNewLines[i])
6564	RTStrFree(pRenameState->apszNewLines[i]);
6565	RTMemTmpFree(pRenameState->apszNewLines);
6566	}
6567	if (pRenameState->pszOldDescName)
6568	RTStrFree(pRenameState->pszOldDescName);
6569	if (pRenameState->pszOldBaseName)
6570	RTStrFree(pRenameState->pszOldBaseName);
6571	if (pRenameState->pszNewBaseName)
6572	RTStrFree(pRenameState->pszNewBaseName);
6573	if (pRenameState->pszOldFullName)
6574	RTStrFree(pRenameState->pszOldFullName);
6575	if (pRenameState->pszNewFullName)
6576	RTStrFree(pRenameState->pszNewFullName);
6577	}
6578	/**
6579	* Rolls back the rename operation to the original state.
6580	*
6581	* @returns VBox status code.
6582	* @param pImage VMDK image instance.
6583	* @param pRenameState The rename state.
6584	*/
6585	static int vmdkRenameRollback(PVMDKIMAGE pImage, PVMDKRENAMESTATE pRenameState)
6586	{
6587	int rc = VINF_SUCCESS;
6588	if (!pRenameState->fImageFreed)
6589	{
6590	/*
6591	* Some extents may have been closed, close the rest. We will
6592	* re-open the whole thing later.
6593	*/
6594	vmdkFreeImage(pImage, false, true /fFlush/);
6595	}
6596	/* Rename files back. */
6597	for (unsigned i = 0; i <= pRenameState->cExtents; i++)
6598	{
6599	if (pRenameState->apszOldName[i])
6600	{
6601	rc = vdIfIoIntFileMove(pImage->pIfIo, pRenameState->apszNewName[i], pRenameState->apszOldName[i], 0);
6602	AssertRC(rc);
6603	}
6604	}
6605	/* Restore the old descriptor. */
6606	PVMDKFILE pFile;
6607	rc = vmdkFileOpen(pImage, &pFile, NULL, pRenameState->pszOldDescName,
6608	VDOpenFlagsToFileOpenFlags(VD_OPEN_FLAGS_NORMAL,
6609	false /* fCreate */));
6610	AssertRC(rc);
6611	if (pRenameState->fEmbeddedDesc)
6612	{
6613	pRenameState->ExtentCopy.pFile = pFile;
6614	pImage->pExtents = &pRenameState->ExtentCopy;
6615	}
6616	else
6617	{
6618	/* Shouldn't be null for separate descriptor.
6619	* There will be no access to the actual content.
6620	*/
6621	pImage->pDescData = pRenameState->pszOldDescName;
6622	pImage->pFile = pFile;
6623	}
6624	pImage->Descriptor = pRenameState->DescriptorCopy;
6625	vmdkWriteDescriptor(pImage, NULL);
6626	vmdkFileClose(pImage, &pFile, false);
6627	/* Get rid of the stuff we implanted. */
6628	pImage->pExtents = NULL;
6629	pImage->pFile = NULL;
6630	pImage->pDescData = NULL;
6631	/* Re-open the image back. */
6632	pImage->pszFilename = pRenameState->pszOldImageName;
6633	rc = vmdkOpenImage(pImage, pImage->uOpenFlags);
6634	return rc;
6635	}
6636	/**
6637	* Rename worker doing the real work.
6638	*
6639	* @returns VBox status code.
6640	* @param pImage VMDK image instance.
6641	* @param pRenameState The rename state.
6642	* @param pszFilename The new filename.
6643	*/
6644	static int vmdkRenameWorker(PVMDKIMAGE pImage, PVMDKRENAMESTATE pRenameState, const char *pszFilename)
6645	{
6646	int rc = VINF_SUCCESS;
6647	unsigned i, line;
6648	/* Update the descriptor with modified extent names. */
6649	for (i = 0, line = pImage->Descriptor.uFirstExtent;
6650	i < pRenameState->cExtents;
6651	i++, line = pImage->Descriptor.aNextLines[line])
6652	{
6653	/* Update the descriptor. */
6654	pRenameState->apszNewLines[i] = vmdkStrReplace(pImage->Descriptor.aLines[line],
6655	pRenameState->pszOldBaseName,
6656	pRenameState->pszNewBaseName);
6657	if (!pRenameState->apszNewLines[i])
6658	{
6659	rc = VERR_NO_MEMORY;
6660	break;
6661	}
6662	pImage->Descriptor.aLines[line] = pRenameState->apszNewLines[i];
6663	}
6664	if (RT_SUCCESS(rc))
6665	{
6666	/* Make sure the descriptor gets written back. */
6667	pImage->Descriptor.fDirty = true;
6668	/* Flush the descriptor now, in case it is embedded. */
6669	vmdkFlushImage(pImage, NULL);
6670	/* Close and rename/move extents. */
6671	for (i = 0; i < pRenameState->cExtents; i++)
6672	{
6673	PVMDKEXTENT pExtent = &pImage->pExtents[i];
6674	/* Compose new name for the extent. */
6675	pRenameState->apszNewName[i] = vmdkStrReplace(pExtent->pszFullname,
6676	pRenameState->pszOldFullName,
6677	pRenameState->pszNewFullName);
6678	if (!pRenameState->apszNewName[i])
6679	{
6680	rc = VERR_NO_MEMORY;
6681	break;
6682	}
6683	/* Close the extent file. */
6684	rc = vmdkFileClose(pImage, &pExtent->pFile, false);
6685	if (RT_FAILURE(rc))
6686	break;;
6687	/* Rename the extent file. */
6688	rc = vdIfIoIntFileMove(pImage->pIfIo, pExtent->pszFullname, pRenameState->apszNewName[i], 0);
6689	if (RT_FAILURE(rc))
6690	break;
6691	/* Remember the old name. */
6692	pRenameState->apszOldName[i] = RTStrDup(pExtent->pszFullname);
6693	}
6694	if (RT_SUCCESS(rc))
6695	{
6696	/* Release all old stuff. */
6697	rc = vmdkFreeImage(pImage, false, true /fFlush/);
6698	if (RT_SUCCESS(rc))
6699	{
6700	pRenameState->fImageFreed = true;
6701	/* Last elements of new/old name arrays are intended for
6702	* storing descriptor's names.
6703	*/
6704	pRenameState->apszNewName[pRenameState->cExtents] = RTStrDup(pszFilename);
6705	/* Rename the descriptor file if it's separate. */
6706	if (!pRenameState->fEmbeddedDesc)
6707	{
6708	rc = vdIfIoIntFileMove(pImage->pIfIo, pImage->pszFilename, pRenameState->apszNewName[pRenameState->cExtents], 0);
6709	if (RT_SUCCESS(rc))
6710	{
6711	/* Save old name only if we may need to change it back. */
6712	pRenameState->apszOldName[pRenameState->cExtents] = RTStrDup(pszFilename);
6713	}
6714	}
6715	/* Update pImage with the new information. */
6716	pImage->pszFilename = pszFilename;
6717	/* Open the new image. */
6718	rc = vmdkOpenImage(pImage, pImage->uOpenFlags);
6719	}
6720	}
6721	}
6722	return rc;
6723	}
6724	/** @copydoc VDIMAGEBACKEND::pfnRename */
6725	static DECLCALLBACK(int) vmdkRename(void pBackendData, const char pszFilename)
6726	{
6727	LogFlowFunc(("pBackendData=%#p pszFilename=%#p\n", pBackendData, pszFilename));
6728	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6729	VMDKRENAMESTATE RenameState;
6730	memset(&RenameState, 0, sizeof(RenameState));
6731	/* Check arguments. */
6732	AssertReturn(( pImage
6733	&& VALID_PTR(pszFilename)
6734	&& *pszFilename
6735	&& !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)), VERR_INVALID_PARAMETER);
6736	int rc = vmdkRenameStatePrepare(pImage, &RenameState, pszFilename);
6737	if (RT_SUCCESS(rc))
6738	{
6739	/* --- Up to this point we have not done any damage yet. --- */
6740	rc = vmdkRenameWorker(pImage, &RenameState, pszFilename);
6741	/* Roll back all changes in case of failure. */
6742	if (RT_FAILURE(rc))
6743	{
6744	int rrc = vmdkRenameRollback(pImage, &RenameState);
6745	AssertRC(rrc);
6746	}
6747	}
6748	vmdkRenameStateDestroy(&RenameState);
6749	LogFlowFunc(("returns %Rrc\n", rc));
6750	return rc;
6751	}
6752	/** @copydoc VDIMAGEBACKEND::pfnClose */
6753	static DECLCALLBACK(int) vmdkClose(void *pBackendData, bool fDelete)
6754	{
6755	LogFlowFunc(("pBackendData=%#p fDelete=%d\n", pBackendData, fDelete));
6756	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6757	int rc = vmdkFreeImage(pImage, fDelete, true /fFlush/);
6758	RTMemFree(pImage);
6759	LogFlowFunc(("returns %Rrc\n", rc));
6760	return rc;
6761	}
6762	/** @copydoc VDIMAGEBACKEND::pfnRead */
6763	static DECLCALLBACK(int) vmdkRead(void *pBackendData, uint64_t uOffset, size_t cbToRead,
6764	PVDIOCTX pIoCtx, size_t *pcbActuallyRead)
6765	{
6766	LogFlowFunc(("pBackendData=%#p uOffset=%llu pIoCtx=%#p cbToRead=%zu pcbActuallyRead=%#p\n",
6767	pBackendData, uOffset, pIoCtx, cbToRead, pcbActuallyRead));
6768	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6769	AssertPtr(pImage);
6770	Assert(uOffset % 512 == 0);
6771	Assert(cbToRead % 512 == 0);
6772	AssertReturn((VALID_PTR(pIoCtx) && cbToRead), VERR_INVALID_PARAMETER);
6773	AssertReturn(uOffset + cbToRead <= pImage->cbSize, VERR_INVALID_PARAMETER);
6774	/* Find the extent and check access permissions as defined in the extent descriptor. */
6775	PVMDKEXTENT pExtent;
6776	uint64_t uSectorExtentRel;
6777	int rc = vmdkFindExtent(pImage, VMDK_BYTE2SECTOR(uOffset),
6778	&pExtent, &uSectorExtentRel);
6779	if ( RT_SUCCESS(rc)
6780	&& pExtent->enmAccess != VMDKACCESS_NOACCESS)
6781	{
6782	/* Clip read range to remain in this extent. */
6783	cbToRead = RT_MIN(cbToRead, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
6784	/* Handle the read according to the current extent type. */
6785	switch (pExtent->enmType)
6786	{
6787	case VMDKETYPE_HOSTED_SPARSE:
6788	{
6789	uint64_t uSectorExtentAbs;
6790	rc = vmdkGetSector(pImage, pIoCtx, pExtent, uSectorExtentRel, &uSectorExtentAbs);
6791	if (RT_FAILURE(rc))
6792	break;
6793	/* Clip read range to at most the rest of the grain. */
6794	cbToRead = RT_MIN(cbToRead, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain - uSectorExtentRel % pExtent->cSectorsPerGrain));
6795	Assert(!(cbToRead % 512));
6796	if (uSectorExtentAbs == 0)
6797	{
6798	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6799	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
6800	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
6801	rc = VERR_VD_BLOCK_FREE;
6802	else
6803	rc = vmdkStreamReadSequential(pImage, pExtent,
6804	uSectorExtentRel,
6805	pIoCtx, cbToRead);
6806	}
6807	else
6808	{
6809	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6810	{
6811	AssertMsg(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
6812	("Async I/O is not supported for stream optimized VMDK's\n"));
6813	uint32_t uSectorInGrain = uSectorExtentRel % pExtent->cSectorsPerGrain;
6814	uSectorExtentAbs -= uSectorInGrain;
6815	if (pExtent->uGrainSectorAbs != uSectorExtentAbs)
6816	{
6817	uint64_t uLBA = 0; /* gcc maybe uninitialized */
6818	rc = vmdkFileInflateSync(pImage, pExtent,
6819	VMDK_SECTOR2BYTE(uSectorExtentAbs),
6820	pExtent->pvGrain,
6821	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
6822	NULL, &uLBA, NULL);
6823	if (RT_FAILURE(rc))
6824	{
6825	pExtent->uGrainSectorAbs = 0;
6826	break;
6827	}
6828	pExtent->uGrainSectorAbs = uSectorExtentAbs;
6829	pExtent->uGrain = uSectorExtentRel / pExtent->cSectorsPerGrain;
6830	Assert(uLBA == uSectorExtentRel);
6831	}
6832	vdIfIoIntIoCtxCopyTo(pImage->pIfIo, pIoCtx,
6833	(uint8_t *)pExtent->pvGrain
6834	+ VMDK_SECTOR2BYTE(uSectorInGrain),
6835	cbToRead);
6836	}
6837	else
6838	rc = vdIfIoIntFileReadUser(pImage->pIfIo, pExtent->pFile->pStorage,
6839	VMDK_SECTOR2BYTE(uSectorExtentAbs),
6840	pIoCtx, cbToRead);
6841	}
6842	break;
6843	}
6844	case VMDKETYPE_VMFS:
6845	case VMDKETYPE_FLAT:
6846	rc = vdIfIoIntFileReadUser(pImage->pIfIo, pExtent->pFile->pStorage,
6847	VMDK_SECTOR2BYTE(uSectorExtentRel),
6848	pIoCtx, cbToRead);
6849	break;
6850	case VMDKETYPE_ZERO:
6851	{
6852	size_t cbSet;
6853	cbSet = vdIfIoIntIoCtxSet(pImage->pIfIo, pIoCtx, 0, cbToRead);
6854	Assert(cbSet == cbToRead);
6855	break;
6856	}
6857	}
6858	if (pcbActuallyRead)
6859	*pcbActuallyRead = cbToRead;
6860	}
6861	else if (RT_SUCCESS(rc))
6862	rc = VERR_VD_VMDK_INVALID_STATE;
6863	LogFlowFunc(("returns %Rrc\n", rc));
6864	return rc;
6865	}
6866	/** @copydoc VDIMAGEBACKEND::pfnWrite */
6867	static DECLCALLBACK(int) vmdkWrite(void *pBackendData, uint64_t uOffset, size_t cbToWrite,
6868	PVDIOCTX pIoCtx, size_t pcbWriteProcess, size_t pcbPreRead,
6869	size_t *pcbPostRead, unsigned fWrite)
6870	{
6871	LogFlowFunc(("pBackendData=%#p uOffset=%llu pIoCtx=%#p cbToWrite=%zu pcbWriteProcess=%#p pcbPreRead=%#p pcbPostRead=%#p\n",
6872	pBackendData, uOffset, pIoCtx, cbToWrite, pcbWriteProcess, pcbPreRead, pcbPostRead));
6873	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6874	int rc;
6875	AssertPtr(pImage);
6876	Assert(uOffset % 512 == 0);
6877	Assert(cbToWrite % 512 == 0);
6878	AssertReturn((VALID_PTR(pIoCtx) && cbToWrite), VERR_INVALID_PARAMETER);
6879	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
6880	{
6881	PVMDKEXTENT pExtent;
6882	uint64_t uSectorExtentRel;
6883	uint64_t uSectorExtentAbs;
6884	/* No size check here, will do that later when the extent is located.
6885	* There are sparse images out there which according to the spec are
6886	* invalid, because the total size is not a multiple of the grain size.
6887	* Also for sparse images which are stitched together in odd ways (not at
6888	* grain boundaries, and with the nominal size not being a multiple of the
6889	* grain size), this would prevent writing to the last grain. */
6890	rc = vmdkFindExtent(pImage, VMDK_BYTE2SECTOR(uOffset),
6891	&pExtent, &uSectorExtentRel);
6892	if (RT_SUCCESS(rc))
6893	{
6894	if ( pExtent->enmAccess != VMDKACCESS_READWRITE
6895	&& ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6896	&& !pImage->pExtents[0].uAppendPosition
6897	&& pExtent->enmAccess != VMDKACCESS_READONLY))
6898	rc = VERR_VD_VMDK_INVALID_STATE;
6899	else
6900	{
6901	/* Handle the write according to the current extent type. */
6902	switch (pExtent->enmType)
6903	{
6904	case VMDKETYPE_HOSTED_SPARSE:
6905	rc = vmdkGetSector(pImage, pIoCtx, pExtent, uSectorExtentRel, &uSectorExtentAbs);
6906	if (RT_SUCCESS(rc))
6907	{
6908	if ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
6909	&& uSectorExtentRel < (uint64_t)pExtent->uLastGrainAccess * pExtent->cSectorsPerGrain)
6910	rc = VERR_VD_VMDK_INVALID_WRITE;
6911	else
6912	{
6913	/* Clip write range to at most the rest of the grain. */
6914	cbToWrite = RT_MIN(cbToWrite,
6915	VMDK_SECTOR2BYTE( pExtent->cSectorsPerGrain
6916	- uSectorExtentRel % pExtent->cSectorsPerGrain));
6917	if (uSectorExtentAbs == 0)
6918	{
6919	if (!(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
6920	{
6921	if (cbToWrite == VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
6922	{
6923	/* Full block write to a previously unallocated block.
6924	* Check if the caller wants to avoid the automatic alloc. */
6925	if (!(fWrite & VD_WRITE_NO_ALLOC))
6926	{
6927	/* Allocate GT and find out where to store the grain. */
6928	rc = vmdkAllocGrain(pImage, pExtent, pIoCtx,
6929	uSectorExtentRel, cbToWrite);
6930	}
6931	else
6932	rc = VERR_VD_BLOCK_FREE;
6933	*pcbPreRead = 0;
6934	*pcbPostRead = 0;
6935	}
6936	else
6937	{
6938	/* Clip write range to remain in this extent. */
6939	cbToWrite = RT_MIN(cbToWrite,
6940	VMDK_SECTOR2BYTE( pExtent->uSectorOffset
6941	+ pExtent->cNominalSectors - uSectorExtentRel));
6942	*pcbPreRead = VMDK_SECTOR2BYTE(uSectorExtentRel % pExtent->cSectorsPerGrain);
6943	pcbPostRead = VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain) - cbToWrite - pcbPreRead;
6944	rc = VERR_VD_BLOCK_FREE;
6945	}
6946	}
6947	else
6948	rc = vmdkStreamAllocGrain(pImage, pExtent, uSectorExtentRel,
6949	pIoCtx, cbToWrite);
6950	}
6951	else
6952	{
6953	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6954	{
6955	/* A partial write to a streamOptimized image is simply
6956	* invalid. It requires rewriting already compressed data
6957	* which is somewhere between expensive and impossible. */
6958	rc = VERR_VD_VMDK_INVALID_STATE;
6959	pExtent->uGrainSectorAbs = 0;
6960	AssertRC(rc);
6961	}
6962	else
6963	{
6964	Assert(!(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED));
6965	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
6966	VMDK_SECTOR2BYTE(uSectorExtentAbs),
6967	pIoCtx, cbToWrite, NULL, NULL);
6968	}
6969	}
6970	}
6971	}
6972	break;
6973	case VMDKETYPE_VMFS:
6974	case VMDKETYPE_FLAT:
6975	/* Clip write range to remain in this extent. */
6976	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
6977	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
6978	VMDK_SECTOR2BYTE(uSectorExtentRel),
6979	pIoCtx, cbToWrite, NULL, NULL);
6980	break;
6981	case VMDKETYPE_ZERO:
6982	/* Clip write range to remain in this extent. */
6983	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
6984	break;
6985	}
6986	}
6987	if (pcbWriteProcess)
6988	*pcbWriteProcess = cbToWrite;
6989	}
6990	}
6991	else
6992	rc = VERR_VD_IMAGE_READ_ONLY;
6993	LogFlowFunc(("returns %Rrc\n", rc));
6994	return rc;
6995	}
6996	/** @copydoc VDIMAGEBACKEND::pfnFlush */
6997	static DECLCALLBACK(int) vmdkFlush(void *pBackendData, PVDIOCTX pIoCtx)
6998	{
6999	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7000	return vmdkFlushImage(pImage, pIoCtx);
7001	}
7002	/** @copydoc VDIMAGEBACKEND::pfnGetVersion */
7003	static DECLCALLBACK(unsigned) vmdkGetVersion(void *pBackendData)
7004	{
7005	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
7006	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7007	AssertPtrReturn(pImage, 0);
7008	return VMDK_IMAGE_VERSION;
7009	}
7010	/** @copydoc VDIMAGEBACKEND::pfnGetFileSize */
7011	static DECLCALLBACK(uint64_t) vmdkGetFileSize(void *pBackendData)
7012	{
7013	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
7014	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7015	uint64_t cb = 0;
7016	AssertPtrReturn(pImage, 0);
7017	if (pImage->pFile != NULL)
7018	{
7019	uint64_t cbFile;
7020	int rc = vdIfIoIntFileGetSize(pImage->pIfIo, pImage->pFile->pStorage, &cbFile);
7021	if (RT_SUCCESS(rc))
7022	cb += cbFile;
7023	}
7024	for (unsigned i = 0; i < pImage->cExtents; i++)
7025	{
7026	if (pImage->pExtents[i].pFile != NULL)
7027	{
7028	uint64_t cbFile;
7029	int rc = vdIfIoIntFileGetSize(pImage->pIfIo, pImage->pExtents[i].pFile->pStorage, &cbFile);
7030	if (RT_SUCCESS(rc))
7031	cb += cbFile;
7032	}
7033	}
7034	LogFlowFunc(("returns %lld\n", cb));
7035	return cb;
7036	}
7037	/** @copydoc VDIMAGEBACKEND::pfnGetPCHSGeometry */
7038	static DECLCALLBACK(int) vmdkGetPCHSGeometry(void *pBackendData, PVDGEOMETRY pPCHSGeometry)
7039	{
7040	LogFlowFunc(("pBackendData=%#p pPCHSGeometry=%#p\n", pBackendData, pPCHSGeometry));
7041	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7042	int rc = VINF_SUCCESS;
7043	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7044	if (pImage->PCHSGeometry.cCylinders)
7045	*pPCHSGeometry = pImage->PCHSGeometry;
7046	else
7047	rc = VERR_VD_GEOMETRY_NOT_SET;
7048	LogFlowFunc(("returns %Rrc (PCHS=%u/%u/%u)\n", rc, pPCHSGeometry->cCylinders, pPCHSGeometry->cHeads, pPCHSGeometry->cSectors));
7049	return rc;
7050	}
7051	/** @copydoc VDIMAGEBACKEND::pfnSetPCHSGeometry */
7052	static DECLCALLBACK(int) vmdkSetPCHSGeometry(void *pBackendData, PCVDGEOMETRY pPCHSGeometry)
7053	{
7054	LogFlowFunc(("pBackendData=%#p pPCHSGeometry=%#p PCHS=%u/%u/%u\n",
7055	pBackendData, pPCHSGeometry, pPCHSGeometry->cCylinders, pPCHSGeometry->cHeads, pPCHSGeometry->cSectors));
7056	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7057	int rc = VINF_SUCCESS;
7058	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7059	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7060	{
7061	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7062	{
7063	rc = vmdkDescSetPCHSGeometry(pImage, pPCHSGeometry);
7064	if (RT_SUCCESS(rc))
7065	pImage->PCHSGeometry = *pPCHSGeometry;
7066	}
7067	else
7068	rc = VERR_NOT_SUPPORTED;
7069	}
7070	else
7071	rc = VERR_VD_IMAGE_READ_ONLY;
7072	LogFlowFunc(("returns %Rrc\n", rc));
7073	return rc;
7074	}
7075	/** @copydoc VDIMAGEBACKEND::pfnGetLCHSGeometry */
7076	static DECLCALLBACK(int) vmdkGetLCHSGeometry(void *pBackendData, PVDGEOMETRY pLCHSGeometry)
7077	{
7078	LogFlowFunc(("pBackendData=%#p pLCHSGeometry=%#p\n", pBackendData, pLCHSGeometry));
7079	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7080	int rc = VINF_SUCCESS;
7081	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7082	if (pImage->LCHSGeometry.cCylinders)
7083	*pLCHSGeometry = pImage->LCHSGeometry;
7084	else
7085	rc = VERR_VD_GEOMETRY_NOT_SET;
7086	LogFlowFunc(("returns %Rrc (LCHS=%u/%u/%u)\n", rc, pLCHSGeometry->cCylinders, pLCHSGeometry->cHeads, pLCHSGeometry->cSectors));
7087	return rc;
7088	}
7089	/** @copydoc VDIMAGEBACKEND::pfnSetLCHSGeometry */
7090	static DECLCALLBACK(int) vmdkSetLCHSGeometry(void *pBackendData, PCVDGEOMETRY pLCHSGeometry)
7091	{
7092	LogFlowFunc(("pBackendData=%#p pLCHSGeometry=%#p LCHS=%u/%u/%u\n",
7093	pBackendData, pLCHSGeometry, pLCHSGeometry->cCylinders, pLCHSGeometry->cHeads, pLCHSGeometry->cSectors));
7094	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7095	int rc = VINF_SUCCESS;
7096	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7097	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7098	{
7099	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7100	{
7101	rc = vmdkDescSetLCHSGeometry(pImage, pLCHSGeometry);
7102	if (RT_SUCCESS(rc))
7103	pImage->LCHSGeometry = *pLCHSGeometry;
7104	}
7105	else
7106	rc = VERR_NOT_SUPPORTED;
7107	}
7108	else
7109	rc = VERR_VD_IMAGE_READ_ONLY;
7110	LogFlowFunc(("returns %Rrc\n", rc));
7111	return rc;
7112	}
7113	/** @copydoc VDIMAGEBACKEND::pfnQueryRegions */
7114	static DECLCALLBACK(int) vmdkQueryRegions(void pBackendData, PCVDREGIONLIST ppRegionList)
7115	{
7116	LogFlowFunc(("pBackendData=%#p ppRegionList=%#p\n", pBackendData, ppRegionList));
7117	PVMDKIMAGE pThis = (PVMDKIMAGE)pBackendData;
7118	AssertPtrReturn(pThis, VERR_VD_NOT_OPENED);
7119	*ppRegionList = &pThis->RegionList;
7120	LogFlowFunc(("returns %Rrc\n", VINF_SUCCESS));
7121	return VINF_SUCCESS;
7122	}
7123	/** @copydoc VDIMAGEBACKEND::pfnRegionListRelease */
7124	static DECLCALLBACK(void) vmdkRegionListRelease(void *pBackendData, PCVDREGIONLIST pRegionList)
7125	{
7126	RT_NOREF1(pRegionList);
7127	LogFlowFunc(("pBackendData=%#p pRegionList=%#p\n", pBackendData, pRegionList));
7128	PVMDKIMAGE pThis = (PVMDKIMAGE)pBackendData;
7129	AssertPtr(pThis); RT_NOREF(pThis);
7130	/* Nothing to do here. */
7131	}
7132	/** @copydoc VDIMAGEBACKEND::pfnGetImageFlags */
7133	static DECLCALLBACK(unsigned) vmdkGetImageFlags(void *pBackendData)
7134	{
7135	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
7136	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7137	AssertPtrReturn(pImage, 0);
7138	LogFlowFunc(("returns %#x\n", pImage->uImageFlags));
7139	return pImage->uImageFlags;
7140	}
7141	/** @copydoc VDIMAGEBACKEND::pfnGetOpenFlags */
7142	static DECLCALLBACK(unsigned) vmdkGetOpenFlags(void *pBackendData)
7143	{
7144	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
7145	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7146	AssertPtrReturn(pImage, 0);
7147	LogFlowFunc(("returns %#x\n", pImage->uOpenFlags));
7148	return pImage->uOpenFlags;
7149	}
7150	/** @copydoc VDIMAGEBACKEND::pfnSetOpenFlags */
7151	static DECLCALLBACK(int) vmdkSetOpenFlags(void *pBackendData, unsigned uOpenFlags)
7152	{
7153	LogFlowFunc(("pBackendData=%#p uOpenFlags=%#x\n", pBackendData, uOpenFlags));
7154	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7155	int rc;
7156	/* Image must be opened and the new flags must be valid. */
7157	if (!pImage \|\| (uOpenFlags & ~( VD_OPEN_FLAGS_READONLY \| VD_OPEN_FLAGS_INFO
7158	\| VD_OPEN_FLAGS_ASYNC_IO \| VD_OPEN_FLAGS_SHAREABLE
7159	\| VD_OPEN_FLAGS_SEQUENTIAL \| VD_OPEN_FLAGS_SKIP_CONSISTENCY_CHECKS)))
7160	rc = VERR_INVALID_PARAMETER;
7161	else
7162	{
7163	/* StreamOptimized images need special treatment: reopen is prohibited. */
7164	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7165	{
7166	if (pImage->uOpenFlags == uOpenFlags)
7167	rc = VINF_SUCCESS;
7168	else
7169	rc = VERR_INVALID_PARAMETER;
7170	}
7171	else
7172	{
7173	/* Implement this operation via reopening the image. */
7174	vmdkFreeImage(pImage, false, true /fFlush/);
7175	rc = vmdkOpenImage(pImage, uOpenFlags);
7176	}
7177	}
7178	LogFlowFunc(("returns %Rrc\n", rc));
7179	return rc;
7180	}
7181	/** @copydoc VDIMAGEBACKEND::pfnGetComment */
7182	static DECLCALLBACK(int) vmdkGetComment(void pBackendData, char pszComment, size_t cbComment)
7183	{
7184	LogFlowFunc(("pBackendData=%#p pszComment=%#p cbComment=%zu\n", pBackendData, pszComment, cbComment));
7185	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7186	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7187	char *pszCommentEncoded = NULL;
7188	int rc = vmdkDescDDBGetStr(pImage, &pImage->Descriptor,
7189	"ddb.comment", &pszCommentEncoded);
7190	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
7191	{
7192	pszCommentEncoded = NULL;
7193	rc = VINF_SUCCESS;
7194	}
7195	if (RT_SUCCESS(rc))
7196	{
7197	if (pszComment && pszCommentEncoded)
7198	rc = vmdkDecodeString(pszCommentEncoded, pszComment, cbComment);
7199	else if (pszComment)
7200	*pszComment = '\0';
7201	if (pszCommentEncoded)
7202	RTMemTmpFree(pszCommentEncoded);
7203	}
7204	LogFlowFunc(("returns %Rrc comment='%s'\n", rc, pszComment));
7205	return rc;
7206	}
7207	/** @copydoc VDIMAGEBACKEND::pfnSetComment */
7208	static DECLCALLBACK(int) vmdkSetComment(void pBackendData, const char pszComment)
7209	{
7210	LogFlowFunc(("pBackendData=%#p pszComment=\"%s\"\n", pBackendData, pszComment));
7211	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7212	int rc;
7213	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7214	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7215	{
7216	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7217	rc = vmdkSetImageComment(pImage, pszComment);
7218	else
7219	rc = VERR_NOT_SUPPORTED;
7220	}
7221	else
7222	rc = VERR_VD_IMAGE_READ_ONLY;
7223	LogFlowFunc(("returns %Rrc\n", rc));
7224	return rc;
7225	}
7226	/** @copydoc VDIMAGEBACKEND::pfnGetUuid */
7227	static DECLCALLBACK(int) vmdkGetUuid(void *pBackendData, PRTUUID pUuid)
7228	{
7229	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
7230	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7231	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7232	*pUuid = pImage->ImageUuid;
7233	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
7234	return VINF_SUCCESS;
7235	}
7236	/** @copydoc VDIMAGEBACKEND::pfnSetUuid */
7237	static DECLCALLBACK(int) vmdkSetUuid(void *pBackendData, PCRTUUID pUuid)
7238	{
7239	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
7240	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7241	int rc = VINF_SUCCESS;
7242	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7243	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7244	{
7245	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7246	{
7247	pImage->ImageUuid = *pUuid;
7248	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
7249	VMDK_DDB_IMAGE_UUID, pUuid);
7250	if (RT_FAILURE(rc))
7251	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
7252	N_("VMDK: error storing image UUID in descriptor in '%s'"), pImage->pszFilename);
7253	}
7254	else
7255	rc = VERR_NOT_SUPPORTED;
7256	}
7257	else
7258	rc = VERR_VD_IMAGE_READ_ONLY;
7259	LogFlowFunc(("returns %Rrc\n", rc));
7260	return rc;
7261	}
7262	/** @copydoc VDIMAGEBACKEND::pfnGetModificationUuid */
7263	static DECLCALLBACK(int) vmdkGetModificationUuid(void *pBackendData, PRTUUID pUuid)
7264	{
7265	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
7266	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7267	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7268	*pUuid = pImage->ModificationUuid;
7269	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
7270	return VINF_SUCCESS;
7271	}
7272	/** @copydoc VDIMAGEBACKEND::pfnSetModificationUuid */
7273	static DECLCALLBACK(int) vmdkSetModificationUuid(void *pBackendData, PCRTUUID pUuid)
7274	{
7275	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
7276	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7277	int rc = VINF_SUCCESS;
7278	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7279	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7280	{
7281	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7282	{
7283	/* Only touch the modification uuid if it changed. */
7284	if (RTUuidCompare(&pImage->ModificationUuid, pUuid))
7285	{
7286	pImage->ModificationUuid = *pUuid;
7287	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
7288	VMDK_DDB_MODIFICATION_UUID, pUuid);
7289	if (RT_FAILURE(rc))
7290	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing modification UUID in descriptor in '%s'"), pImage->pszFilename);
7291	}
7292	}
7293	else
7294	rc = VERR_NOT_SUPPORTED;
7295	}
7296	else
7297	rc = VERR_VD_IMAGE_READ_ONLY;
7298	LogFlowFunc(("returns %Rrc\n", rc));
7299	return rc;
7300	}
7301	/** @copydoc VDIMAGEBACKEND::pfnGetParentUuid */
7302	static DECLCALLBACK(int) vmdkGetParentUuid(void *pBackendData, PRTUUID pUuid)
7303	{
7304	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
7305	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7306	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7307	*pUuid = pImage->ParentUuid;
7308	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
7309	return VINF_SUCCESS;
7310	}
7311	/** @copydoc VDIMAGEBACKEND::pfnSetParentUuid */
7312	static DECLCALLBACK(int) vmdkSetParentUuid(void *pBackendData, PCRTUUID pUuid)
7313	{
7314	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
7315	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7316	int rc = VINF_SUCCESS;
7317	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7318	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7319	{
7320	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7321	{
7322	pImage->ParentUuid = *pUuid;
7323	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
7324	VMDK_DDB_PARENT_UUID, pUuid);
7325	if (RT_FAILURE(rc))
7326	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
7327	N_("VMDK: error storing parent image UUID in descriptor in '%s'"), pImage->pszFilename);
7328	}
7329	else
7330	rc = VERR_NOT_SUPPORTED;
7331	}
7332	else
7333	rc = VERR_VD_IMAGE_READ_ONLY;
7334	LogFlowFunc(("returns %Rrc\n", rc));
7335	return rc;
7336	}
7337	/** @copydoc VDIMAGEBACKEND::pfnGetParentModificationUuid */
7338	static DECLCALLBACK(int) vmdkGetParentModificationUuid(void *pBackendData, PRTUUID pUuid)
7339	{
7340	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
7341	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7342	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7343	*pUuid = pImage->ParentModificationUuid;
7344	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
7345	return VINF_SUCCESS;
7346	}
7347	/** @copydoc VDIMAGEBACKEND::pfnSetParentModificationUuid */
7348	static DECLCALLBACK(int) vmdkSetParentModificationUuid(void *pBackendData, PCRTUUID pUuid)
7349	{
7350	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
7351	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7352	int rc = VINF_SUCCESS;
7353	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7354	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7355	{
7356	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7357	{
7358	pImage->ParentModificationUuid = *pUuid;
7359	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
7360	VMDK_DDB_PARENT_MODIFICATION_UUID, pUuid);
7361	if (RT_FAILURE(rc))
7362	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent image UUID in descriptor in '%s'"), pImage->pszFilename);
7363	}
7364	else
7365	rc = VERR_NOT_SUPPORTED;
7366	}
7367	else
7368	rc = VERR_VD_IMAGE_READ_ONLY;
7369	LogFlowFunc(("returns %Rrc\n", rc));
7370	return rc;
7371	}
7372	/** @copydoc VDIMAGEBACKEND::pfnDump */
7373	static DECLCALLBACK(void) vmdkDump(void *pBackendData)
7374	{
7375	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7376	AssertPtrReturnVoid(pImage);
7377	vdIfErrorMessage(pImage->pIfError, "Header: Geometry PCHS=%u/%u/%u LCHS=%u/%u/%u cbSector=%llu\n",
7378	pImage->PCHSGeometry.cCylinders, pImage->PCHSGeometry.cHeads, pImage->PCHSGeometry.cSectors,
7379	pImage->LCHSGeometry.cCylinders, pImage->LCHSGeometry.cHeads, pImage->LCHSGeometry.cSectors,
7380	VMDK_BYTE2SECTOR(pImage->cbSize));
7381	vdIfErrorMessage(pImage->pIfError, "Header: uuidCreation={%RTuuid}\n", &pImage->ImageUuid);
7382	vdIfErrorMessage(pImage->pIfError, "Header: uuidModification={%RTuuid}\n", &pImage->ModificationUuid);
7383	vdIfErrorMessage(pImage->pIfError, "Header: uuidParent={%RTuuid}\n", &pImage->ParentUuid);
7384	vdIfErrorMessage(pImage->pIfError, "Header: uuidParentModification={%RTuuid}\n", &pImage->ParentModificationUuid);
7385	}
7386	const VDIMAGEBACKEND g_VmdkBackend =
7387	{
7388	/* u32Version */
7389	VD_IMGBACKEND_VERSION,
7390	/* pszBackendName */
7391	"VMDK",
7392	/* uBackendCaps */
7393	VD_CAP_UUID \| VD_CAP_CREATE_FIXED \| VD_CAP_CREATE_DYNAMIC
7394	\| VD_CAP_CREATE_SPLIT_2G \| VD_CAP_DIFF \| VD_CAP_FILE \| VD_CAP_ASYNC
7395	\| VD_CAP_VFS \| VD_CAP_PREFERRED,
7396	/* paFileExtensions */
7397	s_aVmdkFileExtensions,
7398	/* paConfigInfo */
7399	s_aVmdkConfigInfo,
7400	/* pfnProbe */
7401	vmdkProbe,
7402	/* pfnOpen */
7403	vmdkOpen,
7404	/* pfnCreate */
7405	vmdkCreate,
7406	/* pfnRename */
7407	vmdkRename,
7408	/* pfnClose */
7409	vmdkClose,
7410	/* pfnRead */
7411	vmdkRead,
7412	/* pfnWrite */
7413	vmdkWrite,
7414	/* pfnFlush */
7415	vmdkFlush,
7416	/* pfnDiscard */
7417	NULL,
7418	/* pfnGetVersion */
7419	vmdkGetVersion,
7420	/* pfnGetFileSize */
7421	vmdkGetFileSize,
7422	/* pfnGetPCHSGeometry */
7423	vmdkGetPCHSGeometry,
7424	/* pfnSetPCHSGeometry */
7425	vmdkSetPCHSGeometry,
7426	/* pfnGetLCHSGeometry */
7427	vmdkGetLCHSGeometry,
7428	/* pfnSetLCHSGeometry */
7429	vmdkSetLCHSGeometry,
7430	/* pfnQueryRegions */
7431	vmdkQueryRegions,
7432	/* pfnRegionListRelease */
7433	vmdkRegionListRelease,
7434	/* pfnGetImageFlags */
7435	vmdkGetImageFlags,
7436	/* pfnGetOpenFlags */
7437	vmdkGetOpenFlags,
7438	/* pfnSetOpenFlags */
7439	vmdkSetOpenFlags,
7440	/* pfnGetComment */
7441	vmdkGetComment,
7442	/* pfnSetComment */
7443	vmdkSetComment,
7444	/* pfnGetUuid */
7445	vmdkGetUuid,
7446	/* pfnSetUuid */
7447	vmdkSetUuid,
7448	/* pfnGetModificationUuid */
7449	vmdkGetModificationUuid,
7450	/* pfnSetModificationUuid */
7451	vmdkSetModificationUuid,
7452	/* pfnGetParentUuid */
7453	vmdkGetParentUuid,
7454	/* pfnSetParentUuid */
7455	vmdkSetParentUuid,
7456	/* pfnGetParentModificationUuid */
7457	vmdkGetParentModificationUuid,
7458	/* pfnSetParentModificationUuid */
7459	vmdkSetParentModificationUuid,
7460	/* pfnDump */
7461	vmdkDump,
7462	/* pfnGetTimestamp */
7463	NULL,
7464	/* pfnGetParentTimestamp */
7465	NULL,
7466	/* pfnSetParentTimestamp */
7467	NULL,
7468	/* pfnGetParentFilename */
7469	NULL,
7470	/* pfnSetParentFilename */
7471	NULL,
7472	/* pfnComposeLocation */
7473	genericFileComposeLocation,
7474	/* pfnComposeName */
7475	genericFileComposeName,
7476	/* pfnCompact */
7477	NULL,
7478	/* pfnResize */
7479	NULL,
7480	/* pfnRepair */
7481	NULL,
7482	/* pfnTraverseMetadata */
7483	NULL,
7484	/* u32VersionEnd */
7485	VD_IMGBACKEND_VERSION
7486	};

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

source: vbox/trunk/src/VBox/Storage/VMDK.cpp@ 89825

Download in other formats: