VMDK.cpp@ 103131

Last change on this file since 103131 was 99963, checked in by vboxsync, 17 months ago
VBox/types.h,pdmnetifs.h: Replaced PDMDATASEG with RTSGSEG since the two structures are identical. This should make it easy to create a RTSGBUF for a PDMSCATTERGATHER buffer (just limit the operations to cbUsed bytes).
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 379.2 KB

Line
1	/* $Id: VMDK.cpp 99963 2023-05-24 23:27:54Z vboxsync $ */
2	/** @file
3	* VMDK disk image, core code.
4	*/
5
6	/*
7	* Copyright (C) 2006-2023 Oracle and/or its affiliates.
8	*
9	* This file is part of VirtualBox base platform packages, as
10	* available from https://www.virtualbox.org.
11	*
12	* This program is free software; you can redistribute it and/or
13	* modify it under the terms of the GNU General Public License
14	* as published by the Free Software Foundation, in version 3 of the
15	* License.
16	*
17	* This program is distributed in the hope that it will be useful, but
18	* WITHOUT ANY WARRANTY; without even the implied warranty of
19	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20	* General Public License for more details.
21	*
22	* You should have received a copy of the GNU General Public License
23	* along with this program; if not, see <https://www.gnu.org/licenses>.
24	*
25	* SPDX-License-Identifier: GPL-3.0-only
26	*/
27
28
29	/*********************************************************************************************************************************
30	* Header Files *
31	*********************************************************************************************************************************/
32	#define LOG_GROUP LOG_GROUP_VD_VMDK
33	#include <VBox/log.h> /* before VBox/vd-ifs.h */
34	#include <VBox/vd-plugin.h>
35	#include <VBox/err.h>
36
37	#include <iprt/assert.h>
38	#include <iprt/alloc.h>
39	#include <iprt/base64.h>
40	#include <iprt/ctype.h>
41	#include <iprt/crc.h>
42	#include <iprt/dvm.h>
43	#include <iprt/uuid.h>
44	#include <iprt/path.h>
45	#include <iprt/rand.h>
46	#include <iprt/sg.h>
47	#include <iprt/sort.h>
48	#include <iprt/string.h>
49	#include <iprt/zip.h>
50	#include <iprt/asm.h>
51	#ifdef RT_OS_WINDOWS
52	# include <iprt/utf16.h>
53	# include <iprt/uni.h>
54	# include <iprt/uni.h>
55	# include <iprt/nt/nt-and-windows.h>
56	# include <winioctl.h>
57	#endif
58	#ifdef RT_OS_LINUX
59	# include <errno.h>
60	# include <sys/stat.h>
61	# include <iprt/dir.h>
62	# include <iprt/symlink.h>
63	# include <iprt/linux/sysfs.h>
64	#endif
65	#ifdef RT_OS_FREEBSD
66	#include <libgeom.h>
67	#include <sys/stat.h>
68	#include <stdlib.h>
69	#endif
70	#ifdef RT_OS_SOLARIS
71	#include <sys/dkio.h>
72	#include <sys/vtoc.h>
73	#include <sys/efi_partition.h>
74	#include <unistd.h>
75	#include <errno.h>
76	#endif
77	#ifdef RT_OS_DARWIN
78	# include <sys/stat.h>
79	# include <sys/disk.h>
80	# include <errno.h>
81	/* The following structure and IOCTLs are defined in znu bsd/sys/disk.h but
82	inside KERNEL ifdefs and thus stripped from the SDK edition of the header.
83	While we could try include the header from the Kernel.framework, it's a lot
84	easier to just add the structure and 4 defines here. */
85	typedef struct
86	{
87	uint64_t offset;
88	uint64_t length;
89	uint8_t reserved0128[12];
90	dev_t dev;
91	} dk_physical_extent_t;
92	# define DKIOCGETBASE _IOR( 'd', 73, uint64_t)
93	# define DKIOCLOCKPHYSICALEXTENTS _IO( 'd', 81)
94	# define DKIOCGETPHYSICALEXTENT _IOWR('d', 82, dk_physical_extent_t)
95	# define DKIOCUNLOCKPHYSICALEXTENTS _IO( 'd', 83)
96	#endif /* RT_OS_DARWIN */
97
98	#include "VDBackends.h"
99
100
101	/*********************************************************************************************************************************
102	* Constants And Macros, Structures and Typedefs *
103	*********************************************************************************************************************************/
104
105	/** Maximum encoded string size (including NUL) we allow for VMDK images.
106	* Deliberately not set high to avoid running out of descriptor space. */
107	#define VMDK_ENCODED_COMMENT_MAX 1024
108
109	/** VMDK descriptor DDB entry for PCHS cylinders. */
110	#define VMDK_DDB_GEO_PCHS_CYLINDERS "ddb.geometry.cylinders"
111
112	/** VMDK descriptor DDB entry for PCHS heads. */
113	#define VMDK_DDB_GEO_PCHS_HEADS "ddb.geometry.heads"
114
115	/** VMDK descriptor DDB entry for PCHS sectors. */
116	#define VMDK_DDB_GEO_PCHS_SECTORS "ddb.geometry.sectors"
117
118	/** VMDK descriptor DDB entry for LCHS cylinders. */
119	#define VMDK_DDB_GEO_LCHS_CYLINDERS "ddb.geometry.biosCylinders"
120
121	/** VMDK descriptor DDB entry for LCHS heads. */
122	#define VMDK_DDB_GEO_LCHS_HEADS "ddb.geometry.biosHeads"
123
124	/** VMDK descriptor DDB entry for LCHS sectors. */
125	#define VMDK_DDB_GEO_LCHS_SECTORS "ddb.geometry.biosSectors"
126
127	/** VMDK descriptor DDB entry for image UUID. */
128	#define VMDK_DDB_IMAGE_UUID "ddb.uuid.image"
129
130	/** VMDK descriptor DDB entry for image modification UUID. */
131	#define VMDK_DDB_MODIFICATION_UUID "ddb.uuid.modification"
132
133	/** VMDK descriptor DDB entry for parent image UUID. */
134	#define VMDK_DDB_PARENT_UUID "ddb.uuid.parent"
135
136	/** VMDK descriptor DDB entry for parent image modification UUID. */
137	#define VMDK_DDB_PARENT_MODIFICATION_UUID "ddb.uuid.parentmodification"
138
139	/** No compression for streamOptimized files. */
140	#define VMDK_COMPRESSION_NONE 0
141
142	/** Deflate compression for streamOptimized files. */
143	#define VMDK_COMPRESSION_DEFLATE 1
144
145	/** Marker that the actual GD value is stored in the footer. */
146	#define VMDK_GD_AT_END 0xffffffffffffffffULL
147
148	/** Marker for end-of-stream in streamOptimized images. */
149	#define VMDK_MARKER_EOS 0
150
151	/** Marker for grain table block in streamOptimized images. */
152	#define VMDK_MARKER_GT 1
153
154	/** Marker for grain directory block in streamOptimized images. */
155	#define VMDK_MARKER_GD 2
156
157	/** Marker for footer in streamOptimized images. */
158	#define VMDK_MARKER_FOOTER 3
159
160	/** Marker for unknown purpose in streamOptimized images.
161	* Shows up in very recent images created by vSphere, but only sporadically.
162	* They "forgot" to document that one in the VMDK specification. */
163	#define VMDK_MARKER_UNSPECIFIED 4
164
165	/** Dummy marker for "don't check the marker value". */
166	#define VMDK_MARKER_IGNORE 0xffffffffU
167
168	/**
169	* Magic number for hosted images created by VMware Workstation 4, VMware
170	* Workstation 5, VMware Server or VMware Player. Not necessarily sparse.
171	*/
172	#define VMDK_SPARSE_MAGICNUMBER 0x564d444b /* 'V' 'M' 'D' 'K' */
173
174	/** VMDK sector size in bytes. */
175	#define VMDK_SECTOR_SIZE 512
176	/** Max string buffer size for uint64_t with null term */
177	#define UINT64_MAX_BUFF_SIZE 21
178	/** Grain directory entry size in bytes */
179	#define VMDK_GRAIN_DIR_ENTRY_SIZE 4
180	/** Grain table size in bytes */
181	#define VMDK_GRAIN_TABLE_SIZE 2048
182
183	/**
184	* VMDK hosted binary extent header. The "Sparse" is a total misnomer, as
185	* this header is also used for monolithic flat images.
186	*/
187	#pragma pack(1)
188	typedef struct SparseExtentHeader
189	{
190	uint32_t magicNumber;
191	uint32_t version;
192	uint32_t flags;
193	uint64_t capacity;
194	uint64_t grainSize;
195	uint64_t descriptorOffset;
196	uint64_t descriptorSize;
197	uint32_t numGTEsPerGT;
198	uint64_t rgdOffset;
199	uint64_t gdOffset;
200	uint64_t overHead;
201	bool uncleanShutdown;
202	char singleEndLineChar;
203	char nonEndLineChar;
204	char doubleEndLineChar1;
205	char doubleEndLineChar2;
206	uint16_t compressAlgorithm;
207	uint8_t pad[433];
208	} SparseExtentHeader;
209	#pragma pack()
210
211	/** The maximum allowed descriptor size in the extent header in sectors. */
212	#define VMDK_SPARSE_DESCRIPTOR_SIZE_MAX UINT64_C(20480) /* 10MB */
213
214	/** VMDK capacity for a single chunk when 2G splitting is turned on. Should be
215	* divisible by the default grain size (64K) */
216	#define VMDK_2G_SPLIT_SIZE (2047 * 1024 * 1024)
217
218	/** VMDK streamOptimized file format marker. The type field may or may not
219	* be actually valid, but there's always data to read there. */
220	#pragma pack(1)
221	typedef struct VMDKMARKER
222	{
223	uint64_t uSector;
224	uint32_t cbSize;
225	uint32_t uType;
226	} VMDKMARKER, *PVMDKMARKER;
227	#pragma pack()
228
229
230	/** Convert sector number/size to byte offset/size. */
231	#define VMDK_SECTOR2BYTE(u) ((uint64_t)(u) << 9)
232
233	/** Convert byte offset/size to sector number/size. */
234	#define VMDK_BYTE2SECTOR(u) ((u) >> 9)
235
236	/**
237	* VMDK extent type.
238	*/
239	typedef enum VMDKETYPE
240	{
241	/** Hosted sparse extent. */
242	VMDKETYPE_HOSTED_SPARSE = 1,
243	/** Flat extent. */
244	VMDKETYPE_FLAT,
245	/** Zero extent. */
246	VMDKETYPE_ZERO,
247	/** VMFS extent, used by ESX. */
248	VMDKETYPE_VMFS
249	} VMDKETYPE, *PVMDKETYPE;
250
251	/**
252	* VMDK access type for a extent.
253	*/
254	typedef enum VMDKACCESS
255	{
256	/** No access allowed. */
257	VMDKACCESS_NOACCESS = 0,
258	/** Read-only access. */
259	VMDKACCESS_READONLY,
260	/** Read-write access. */
261	VMDKACCESS_READWRITE
262	} VMDKACCESS, *PVMDKACCESS;
263
264	/** Forward declaration for PVMDKIMAGE. */
265	typedef struct VMDKIMAGE *PVMDKIMAGE;
266
267	/**
268	* Extents files entry. Used for opening a particular file only once.
269	*/
270	typedef struct VMDKFILE
271	{
272	/** Pointer to file path. Local copy. */
273	const char *pszFilename;
274	/** Pointer to base name. Local copy. */
275	const char *pszBasename;
276	/** File open flags for consistency checking. */
277	unsigned fOpen;
278	/** Handle for sync/async file abstraction.*/
279	PVDIOSTORAGE pStorage;
280	/** Reference counter. */
281	unsigned uReferences;
282	/** Flag whether the file should be deleted on last close. */
283	bool fDelete;
284	/** Pointer to the image we belong to (for debugging purposes). */
285	PVMDKIMAGE pImage;
286	/** Pointer to next file descriptor. */
287	struct VMDKFILE *pNext;
288	/** Pointer to the previous file descriptor. */
289	struct VMDKFILE *pPrev;
290	} VMDKFILE, *PVMDKFILE;
291
292	/**
293	* VMDK extent data structure.
294	*/
295	typedef struct VMDKEXTENT
296	{
297	/** File handle. */
298	PVMDKFILE pFile;
299	/** Base name of the image extent. */
300	const char *pszBasename;
301	/** Full name of the image extent. */
302	const char *pszFullname;
303	/** Number of sectors in this extent. */
304	uint64_t cSectors;
305	/** Number of sectors per block (grain in VMDK speak). */
306	uint64_t cSectorsPerGrain;
307	/** Starting sector number of descriptor. */
308	uint64_t uDescriptorSector;
309	/** Size of descriptor in sectors. */
310	uint64_t cDescriptorSectors;
311	/** Starting sector number of grain directory. */
312	uint64_t uSectorGD;
313	/** Starting sector number of redundant grain directory. */
314	uint64_t uSectorRGD;
315	/** Total number of metadata sectors. */
316	uint64_t cOverheadSectors;
317	/** Nominal size (i.e. as described by the descriptor) of this extent. */
318	uint64_t cNominalSectors;
319	/** Sector offset (i.e. as described by the descriptor) of this extent. */
320	uint64_t uSectorOffset;
321	/** Number of entries in a grain table. */
322	uint32_t cGTEntries;
323	/** Number of sectors reachable via a grain directory entry. */
324	uint32_t cSectorsPerGDE;
325	/** Number of entries in the grain directory. */
326	uint32_t cGDEntries;
327	/** Pointer to the next free sector. Legacy information. Do not use. */
328	uint32_t uFreeSector;
329	/** Number of this extent in the list of images. */
330	uint32_t uExtent;
331	/** Pointer to the descriptor (NULL if no descriptor in this extent). */
332	char *pDescData;
333	/** Pointer to the grain directory. */
334	uint32_t *pGD;
335	/** Pointer to the redundant grain directory. */
336	uint32_t *pRGD;
337	/** VMDK version of this extent. 1=1.0/1.1 */
338	uint32_t uVersion;
339	/** Type of this extent. */
340	VMDKETYPE enmType;
341	/** Access to this extent. */
342	VMDKACCESS enmAccess;
343	/** Flag whether this extent is marked as unclean. */
344	bool fUncleanShutdown;
345	/** Flag whether the metadata in the extent header needs to be updated. */
346	bool fMetaDirty;
347	/** Flag whether there is a footer in this extent. */
348	bool fFooter;
349	/** Compression type for this extent. */
350	uint16_t uCompression;
351	/** Append position for writing new grain. Only for sparse extents. */
352	uint64_t uAppendPosition;
353	/** Last grain which was accessed. Only for streamOptimized extents. */
354	uint32_t uLastGrainAccess;
355	/** Starting sector corresponding to the grain buffer. */
356	uint32_t uGrainSectorAbs;
357	/** Grain number corresponding to the grain buffer. */
358	uint32_t uGrain;
359	/** Actual size of the compressed data, only valid for reading. */
360	uint32_t cbGrainStreamRead;
361	/** Size of compressed grain buffer for streamOptimized extents. */
362	size_t cbCompGrain;
363	/** Compressed grain buffer for streamOptimized extents, with marker. */
364	void *pvCompGrain;
365	/** Decompressed grain buffer for streamOptimized extents. */
366	void *pvGrain;
367	/** Reference to the image in which this extent is used. Do not use this
368	* on a regular basis to avoid passing pImage references to functions
369	* explicitly. */
370	struct VMDKIMAGE *pImage;
371	} VMDKEXTENT, *PVMDKEXTENT;
372
373	/**
374	* Grain table cache size. Allocated per image.
375	*/
376	#define VMDK_GT_CACHE_SIZE 256
377
378	/**
379	* Grain table block size. Smaller than an actual grain table block to allow
380	* more grain table blocks to be cached without having to allocate excessive
381	* amounts of memory for the cache.
382	*/
383	#define VMDK_GT_CACHELINE_SIZE 128
384
385
386	/**
387	* Maximum number of lines in a descriptor file. Not worth the effort of
388	* making it variable. Descriptor files are generally very short (~20 lines),
389	* with the exception of sparse files split in 2G chunks, which need for the
390	* maximum size (almost 2T) exactly 1025 lines for the disk database.
391	*/
392	#define VMDK_DESCRIPTOR_LINES_MAX 1100U
393
394	/**
395	* Parsed descriptor information. Allows easy access and update of the
396	* descriptor (whether separate file or not). Free form text files suck.
397	*/
398	typedef struct VMDKDESCRIPTOR
399	{
400	/** Line number of first entry of the disk descriptor. */
401	unsigned uFirstDesc;
402	/** Line number of first entry in the extent description. */
403	unsigned uFirstExtent;
404	/** Line number of first disk database entry. */
405	unsigned uFirstDDB;
406	/** Total number of lines. */
407	unsigned cLines;
408	/** Total amount of memory available for the descriptor. */
409	size_t cbDescAlloc;
410	/** Set if descriptor has been changed and not yet written to disk. */
411	bool fDirty;
412	/** Array of pointers to the data in the descriptor. */
413	char *aLines[VMDK_DESCRIPTOR_LINES_MAX];
414	/** Array of line indices pointing to the next non-comment line. */
415	unsigned aNextLines[VMDK_DESCRIPTOR_LINES_MAX];
416	} VMDKDESCRIPTOR, *PVMDKDESCRIPTOR;
417
418
419	/**
420	* Cache entry for translating extent/sector to a sector number in that
421	* extent.
422	*/
423	typedef struct VMDKGTCACHEENTRY
424	{
425	/** Extent number for which this entry is valid. */
426	uint32_t uExtent;
427	/** GT data block number. */
428	uint64_t uGTBlock;
429	/** Data part of the cache entry. */
430	uint32_t aGTData[VMDK_GT_CACHELINE_SIZE];
431	} VMDKGTCACHEENTRY, *PVMDKGTCACHEENTRY;
432
433	/**
434	* Cache data structure for blocks of grain table entries. For now this is a
435	* fixed size direct mapping cache, but this should be adapted to the size of
436	* the sparse image and maybe converted to a set-associative cache. The
437	* implementation below implements a write-through cache with write allocate.
438	*/
439	typedef struct VMDKGTCACHE
440	{
441	/** Cache entries. */
442	VMDKGTCACHEENTRY aGTCache[VMDK_GT_CACHE_SIZE];
443	/** Number of cache entries (currently unused). */
444	unsigned cEntries;
445	} VMDKGTCACHE, *PVMDKGTCACHE;
446
447	/**
448	* Complete VMDK image data structure. Mainly a collection of extents and a few
449	* extra global data fields.
450	*/
451	typedef struct VMDKIMAGE
452	{
453	/** Image name. */
454	const char *pszFilename;
455	/** Descriptor file if applicable. */
456	PVMDKFILE pFile;
457
458	/** Pointer to the per-disk VD interface list. */
459	PVDINTERFACE pVDIfsDisk;
460	/** Pointer to the per-image VD interface list. */
461	PVDINTERFACE pVDIfsImage;
462
463	/** Error interface. */
464	PVDINTERFACEERROR pIfError;
465	/** I/O interface. */
466	PVDINTERFACEIOINT pIfIo;
467
468
469	/** Pointer to the image extents. */
470	PVMDKEXTENT pExtents;
471	/** Number of image extents. */
472	unsigned cExtents;
473	/** Pointer to the files list, for opening a file referenced multiple
474	* times only once (happens mainly with raw partition access). */
475	PVMDKFILE pFiles;
476
477	/**
478	* Pointer to an array of segment entries for async I/O.
479	* This is an optimization because the task number to submit is not known
480	* and allocating/freeing an array in the read/write functions every time
481	* is too expensive.
482	*/
483	PRTSGSEG paSegments;
484	/** Entries available in the segments array. */
485	unsigned cSegments;
486
487	/** Open flags passed by VBoxHD layer. */
488	unsigned uOpenFlags;
489	/** Image flags defined during creation or determined during open. */
490	unsigned uImageFlags;
491	/** Total size of the image. */
492	uint64_t cbSize;
493	/** Physical geometry of this image. */
494	VDGEOMETRY PCHSGeometry;
495	/** Logical geometry of this image. */
496	VDGEOMETRY LCHSGeometry;
497	/** Image UUID. */
498	RTUUID ImageUuid;
499	/** Image modification UUID. */
500	RTUUID ModificationUuid;
501	/** Parent image UUID. */
502	RTUUID ParentUuid;
503	/** Parent image modification UUID. */
504	RTUUID ParentModificationUuid;
505
506	/** Pointer to grain table cache, if this image contains sparse extents. */
507	PVMDKGTCACHE pGTCache;
508	/** Pointer to the descriptor (NULL if no separate descriptor file). */
509	char *pDescData;
510	/** Allocation size of the descriptor file. */
511	size_t cbDescAlloc;
512	/** Parsed descriptor file content. */
513	VMDKDESCRIPTOR Descriptor;
514	/** The static region list. */
515	VDREGIONLIST RegionList;
516	} VMDKIMAGE;
517
518
519	/** State for the input/output callout of the inflate reader/deflate writer. */
520	typedef struct VMDKCOMPRESSIO
521	{
522	/* Image this operation relates to. */
523	PVMDKIMAGE pImage;
524	/* Current read position. */
525	ssize_t iOffset;
526	/* Size of the compressed grain buffer (available data). */
527	size_t cbCompGrain;
528	/* Pointer to the compressed grain buffer. */
529	void *pvCompGrain;
530	} VMDKCOMPRESSIO;
531
532
533	/** Tracks async grain allocation. */
534	typedef struct VMDKGRAINALLOCASYNC
535	{
536	/** Flag whether the allocation failed. */
537	bool fIoErr;
538	/** Current number of transfers pending.
539	* If reached 0 and there is an error the old state is restored. */
540	unsigned cIoXfersPending;
541	/** Sector number */
542	uint64_t uSector;
543	/** Flag whether the grain table needs to be updated. */
544	bool fGTUpdateNeeded;
545	/** Extent the allocation happens. */
546	PVMDKEXTENT pExtent;
547	/** Position of the new grain, required for the grain table update. */
548	uint64_t uGrainOffset;
549	/** Grain table sector. */
550	uint64_t uGTSector;
551	/** Backup grain table sector. */
552	uint64_t uRGTSector;
553	} VMDKGRAINALLOCASYNC, *PVMDKGRAINALLOCASYNC;
554
555	/**
556	* State information for vmdkRename() and helpers.
557	*/
558	typedef struct VMDKRENAMESTATE
559	{
560	/** Array of old filenames. */
561	char **apszOldName;
562	/** Array of new filenames. */
563	char **apszNewName;
564	/** Array of new lines in the extent descriptor. */
565	char **apszNewLines;
566	/** Name of the old descriptor file if not a sparse image. */
567	char *pszOldDescName;
568	/** Flag whether we called vmdkFreeImage(). */
569	bool fImageFreed;
570	/** Flag whther the descriptor is embedded in the image (sparse) or
571	* in a separate file. */
572	bool fEmbeddedDesc;
573	/** Number of extents in the image. */
574	unsigned cExtents;
575	/** New base filename. */
576	char *pszNewBaseName;
577	/** The old base filename. */
578	char *pszOldBaseName;
579	/** New full filename. */
580	char *pszNewFullName;
581	/** Old full filename. */
582	char *pszOldFullName;
583	/** The old image name. */
584	const char *pszOldImageName;
585	/** Copy of the original VMDK descriptor. */
586	VMDKDESCRIPTOR DescriptorCopy;
587	/** Copy of the extent state for sparse images. */
588	VMDKEXTENT ExtentCopy;
589	} VMDKRENAMESTATE;
590	/** Pointer to a VMDK rename state. */
591	typedef VMDKRENAMESTATE *PVMDKRENAMESTATE;
592
593
594	/*********************************************************************************************************************************
595	* Static Variables *
596	*********************************************************************************************************************************/
597
598	/** NULL-terminated array of supported file extensions. */
599	static const VDFILEEXTENSION s_aVmdkFileExtensions[] =
600	{
601	{"vmdk", VDTYPE_HDD},
602	{NULL, VDTYPE_INVALID}
603	};
604
605	/** NULL-terminated array of configuration option. */
606	static const VDCONFIGINFO s_aVmdkConfigInfo[] =
607	{
608	/* Options for VMDK raw disks */
609	{ "RawDrive", NULL, VDCFGVALUETYPE_STRING, 0 },
610	{ "Partitions", NULL, VDCFGVALUETYPE_STRING, 0 },
611	{ "BootSector", NULL, VDCFGVALUETYPE_BYTES, 0 },
612	{ "Relative", NULL, VDCFGVALUETYPE_INTEGER, 0 },
613
614	/* End of options list */
615	{ NULL, NULL, VDCFGVALUETYPE_INTEGER, 0 }
616	};
617
618
619	/*********************************************************************************************************************************
620	* Internal Functions *
621	*********************************************************************************************************************************/
622
623	static void vmdkFreeStreamBuffers(PVMDKEXTENT pExtent);
624	static int vmdkFreeExtentData(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
625	bool fDelete);
626
627	static int vmdkCreateExtents(PVMDKIMAGE pImage, unsigned cExtents);
628	static int vmdkFlushImage(PVMDKIMAGE pImage, PVDIOCTX pIoCtx);
629	static int vmdkSetImageComment(PVMDKIMAGE pImage, const char *pszComment);
630	static int vmdkFreeImage(PVMDKIMAGE pImage, bool fDelete, bool fFlush);
631
632	static DECLCALLBACK(int) vmdkAllocGrainComplete(void *pBackendData, PVDIOCTX pIoCtx,
633	void *pvUser, int rcReq);
634
635	/**
636	* Internal: open a file (using a file descriptor cache to ensure each file
637	* is only opened once - anything else can cause locking problems).
638	*/
639	static int vmdkFileOpen(PVMDKIMAGE pImage, PVMDKFILE *ppVmdkFile,
640	const char pszBasename, const char pszFilename, uint32_t fOpen)
641	{
642	int rc = VINF_SUCCESS;
643	PVMDKFILE pVmdkFile;
644
645	for (pVmdkFile = pImage->pFiles;
646	pVmdkFile != NULL;
647	pVmdkFile = pVmdkFile->pNext)
648	{
649	if (!strcmp(pszFilename, pVmdkFile->pszFilename))
650	{
651	Assert(fOpen == pVmdkFile->fOpen);
652	pVmdkFile->uReferences++;
653
654	*ppVmdkFile = pVmdkFile;
655
656	return rc;
657	}
658	}
659
660	/* If we get here, there's no matching entry in the cache. */
661	pVmdkFile = (PVMDKFILE)RTMemAllocZ(sizeof(VMDKFILE));
662	if (!pVmdkFile)
663	{
664	*ppVmdkFile = NULL;
665	return VERR_NO_MEMORY;
666	}
667
668	pVmdkFile->pszFilename = RTStrDup(pszFilename);
669	if (!pVmdkFile->pszFilename)
670	{
671	RTMemFree(pVmdkFile);
672	*ppVmdkFile = NULL;
673	return VERR_NO_MEMORY;
674	}
675
676	if (pszBasename)
677	{
678	pVmdkFile->pszBasename = RTStrDup(pszBasename);
679	if (!pVmdkFile->pszBasename)
680	{
681	RTStrFree((char )(void )pVmdkFile->pszFilename);
682	RTMemFree(pVmdkFile);
683	*ppVmdkFile = NULL;
684	return VERR_NO_MEMORY;
685	}
686	}
687
688	pVmdkFile->fOpen = fOpen;
689
690	rc = vdIfIoIntFileOpen(pImage->pIfIo, pszFilename, fOpen,
691	&pVmdkFile->pStorage);
692	if (RT_SUCCESS(rc))
693	{
694	pVmdkFile->uReferences = 1;
695	pVmdkFile->pImage = pImage;
696	pVmdkFile->pNext = pImage->pFiles;
697	if (pImage->pFiles)
698	pImage->pFiles->pPrev = pVmdkFile;
699	pImage->pFiles = pVmdkFile;
700	*ppVmdkFile = pVmdkFile;
701	}
702	else
703	{
704	RTStrFree((char )(void )pVmdkFile->pszFilename);
705	RTMemFree(pVmdkFile);
706	*ppVmdkFile = NULL;
707	}
708
709	return rc;
710	}
711
712	/**
713	* Internal: close a file, updating the file descriptor cache.
714	*/
715	static int vmdkFileClose(PVMDKIMAGE pImage, PVMDKFILE *ppVmdkFile, bool fDelete)
716	{
717	int rc = VINF_SUCCESS;
718	PVMDKFILE pVmdkFile = *ppVmdkFile;
719
720	AssertPtr(pVmdkFile);
721
722	pVmdkFile->fDelete \|= fDelete;
723	Assert(pVmdkFile->uReferences);
724	pVmdkFile->uReferences--;
725	if (pVmdkFile->uReferences == 0)
726	{
727	PVMDKFILE pPrev;
728	PVMDKFILE pNext;
729
730	/* Unchain the element from the list. */
731	pPrev = pVmdkFile->pPrev;
732	pNext = pVmdkFile->pNext;
733
734	if (pNext)
735	pNext->pPrev = pPrev;
736	if (pPrev)
737	pPrev->pNext = pNext;
738	else
739	pImage->pFiles = pNext;
740
741	rc = vdIfIoIntFileClose(pImage->pIfIo, pVmdkFile->pStorage);
742
743	bool fFileDel = pVmdkFile->fDelete;
744	if ( pVmdkFile->pszBasename
745	&& fFileDel)
746	{
747	const char *pszSuffix = RTPathSuffix(pVmdkFile->pszBasename);
748	if ( RTPathHasPath(pVmdkFile->pszBasename)
749	\|\| !pszSuffix
750	\|\| ( strcmp(pszSuffix, ".vmdk")
751	&& strcmp(pszSuffix, ".bin")
752	&& strcmp(pszSuffix, ".img")))
753	fFileDel = false;
754	}
755
756	if (fFileDel)
757	{
758	int rc2 = vdIfIoIntFileDelete(pImage->pIfIo, pVmdkFile->pszFilename);
759	if (RT_SUCCESS(rc))
760	rc = rc2;
761	}
762	else if (pVmdkFile->fDelete)
763	LogRel(("VMDK: Denying deletion of %s\n", pVmdkFile->pszBasename));
764	RTStrFree((char )(void )pVmdkFile->pszFilename);
765	if (pVmdkFile->pszBasename)
766	RTStrFree((char )(void )pVmdkFile->pszBasename);
767	RTMemFree(pVmdkFile);
768	}
769
770	*ppVmdkFile = NULL;
771	return rc;
772	}
773
774	/#define VMDK_USE_BLOCK_DECOMP_API - test and enable /
775	#ifndef VMDK_USE_BLOCK_DECOMP_API
776	static DECLCALLBACK(int) vmdkFileInflateHelper(void pvUser, void pvBuf, size_t cbBuf, size_t *pcbBuf)
777	{
778	VMDKCOMPRESSIO pInflateState = (VMDKCOMPRESSIO )pvUser;
779	size_t cbInjected = 0;
780
781	Assert(cbBuf);
782	if (pInflateState->iOffset < 0)
783	{
784	(uint8_t )pvBuf = RTZIPTYPE_ZLIB;
785	pvBuf = (uint8_t *)pvBuf + 1;
786	cbBuf--;
787	cbInjected = 1;
788	pInflateState->iOffset = RT_UOFFSETOF(VMDKMARKER, uType);
789	}
790	if (!cbBuf)
791	{
792	if (pcbBuf)
793	*pcbBuf = cbInjected;
794	return VINF_SUCCESS;
795	}
796	cbBuf = RT_MIN(cbBuf, pInflateState->cbCompGrain - pInflateState->iOffset);
797	memcpy(pvBuf,
798	(uint8_t *)pInflateState->pvCompGrain + pInflateState->iOffset,
799	cbBuf);
800	pInflateState->iOffset += cbBuf;
801	Assert(pcbBuf);
802	*pcbBuf = cbBuf + cbInjected;
803	return VINF_SUCCESS;
804	}
805	#endif
806
807	/**
808	* Internal: read from a file and inflate the compressed data,
809	* distinguishing between async and normal operation
810	*/
811	DECLINLINE(int) vmdkFileInflateSync(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
812	uint64_t uOffset, void *pvBuf,
813	size_t cbToRead, const void *pcvMarker,
814	uint64_t puLBA, uint32_t pcbMarkerData)
815	{
816	int rc;
817	#ifndef VMDK_USE_BLOCK_DECOMP_API
818	PRTZIPDECOMP pZip = NULL;
819	#endif
820	VMDKMARKER pMarker = (VMDKMARKER )pExtent->pvCompGrain;
821	size_t cbCompSize, cbActuallyRead;
822
823	if (!pcvMarker)
824	{
825	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
826	uOffset, pMarker, RT_UOFFSETOF(VMDKMARKER, uType));
827	if (RT_FAILURE(rc))
828	return rc;
829	}
830	else
831	{
832	memcpy(pMarker, pcvMarker, RT_UOFFSETOF(VMDKMARKER, uType));
833	/* pcvMarker endianness has already been partially transformed, fix it */
834	pMarker->uSector = RT_H2LE_U64(pMarker->uSector);
835	pMarker->cbSize = RT_H2LE_U32(pMarker->cbSize);
836	}
837
838	cbCompSize = RT_LE2H_U32(pMarker->cbSize);
839	if (cbCompSize == 0)
840	{
841	AssertMsgFailed(("VMDK: corrupted marker\n"));
842	return VERR_VD_VMDK_INVALID_FORMAT;
843	}
844
845	/* Sanity check - the expansion ratio should be much less than 2. */
846	Assert(cbCompSize < 2 * cbToRead);
847	if (cbCompSize >= 2 * cbToRead)
848	return VERR_VD_VMDK_INVALID_FORMAT;
849
850	/* Compressed grain marker. Data follows immediately. */
851	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
852	uOffset + RT_UOFFSETOF(VMDKMARKER, uType),
853	(uint8_t *)pExtent->pvCompGrain
854	+ RT_UOFFSETOF(VMDKMARKER, uType),
855	RT_ALIGN_Z( cbCompSize
856	+ RT_UOFFSETOF(VMDKMARKER, uType),
857	512)
858	- RT_UOFFSETOF(VMDKMARKER, uType));
859
860	if (puLBA)
861	*puLBA = RT_LE2H_U64(pMarker->uSector);
862	if (pcbMarkerData)
863	*pcbMarkerData = RT_ALIGN( cbCompSize
864	+ RT_UOFFSETOF(VMDKMARKER, uType),
865	512);
866
867	#ifdef VMDK_USE_BLOCK_DECOMP_API
868	rc = RTZipBlockDecompress(RTZIPTYPE_ZLIB, 0 /fFlags/,
869	pExtent->pvCompGrain, cbCompSize + RT_UOFFSETOF(VMDKMARKER, uType), NULL,
870	pvBuf, cbToRead, &cbActuallyRead);
871	#else
872	VMDKCOMPRESSIO InflateState;
873	InflateState.pImage = pImage;
874	InflateState.iOffset = -1;
875	InflateState.cbCompGrain = cbCompSize + RT_UOFFSETOF(VMDKMARKER, uType);
876	InflateState.pvCompGrain = pExtent->pvCompGrain;
877
878	rc = RTZipDecompCreate(&pZip, &InflateState, vmdkFileInflateHelper);
879	if (RT_FAILURE(rc))
880	return rc;
881	rc = RTZipDecompress(pZip, pvBuf, cbToRead, &cbActuallyRead);
882	RTZipDecompDestroy(pZip);
883	#endif /* !VMDK_USE_BLOCK_DECOMP_API */
884	if (RT_FAILURE(rc))
885	{
886	if (rc == VERR_ZIP_CORRUPTED)
887	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: Compressed image is corrupted '%s'"), pExtent->pszFullname);
888	return rc;
889	}
890	if (cbActuallyRead != cbToRead)
891	rc = VERR_VD_VMDK_INVALID_FORMAT;
892	return rc;
893	}
894
895	static DECLCALLBACK(int) vmdkFileDeflateHelper(void pvUser, const void pvBuf, size_t cbBuf)
896	{
897	VMDKCOMPRESSIO pDeflateState = (VMDKCOMPRESSIO )pvUser;
898
899	Assert(cbBuf);
900	if (pDeflateState->iOffset < 0)
901	{
902	pvBuf = (const uint8_t *)pvBuf + 1;
903	cbBuf--;
904	pDeflateState->iOffset = RT_UOFFSETOF(VMDKMARKER, uType);
905	}
906	if (!cbBuf)
907	return VINF_SUCCESS;
908	if (pDeflateState->iOffset + cbBuf > pDeflateState->cbCompGrain)
909	return VERR_BUFFER_OVERFLOW;
910	memcpy((uint8_t *)pDeflateState->pvCompGrain + pDeflateState->iOffset,
911	pvBuf, cbBuf);
912	pDeflateState->iOffset += cbBuf;
913	return VINF_SUCCESS;
914	}
915
916	/**
917	* Internal: deflate the uncompressed data and write to a file,
918	* distinguishing between async and normal operation
919	*/
920	DECLINLINE(int) vmdkFileDeflateSync(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
921	uint64_t uOffset, const void *pvBuf,
922	size_t cbToWrite, uint64_t uLBA,
923	uint32_t *pcbMarkerData)
924	{
925	int rc;
926	PRTZIPCOMP pZip = NULL;
927	VMDKCOMPRESSIO DeflateState;
928
929	DeflateState.pImage = pImage;
930	DeflateState.iOffset = -1;
931	DeflateState.cbCompGrain = pExtent->cbCompGrain;
932	DeflateState.pvCompGrain = pExtent->pvCompGrain;
933
934	rc = RTZipCompCreate(&pZip, &DeflateState, vmdkFileDeflateHelper,
935	RTZIPTYPE_ZLIB, RTZIPLEVEL_DEFAULT);
936	if (RT_FAILURE(rc))
937	return rc;
938	rc = RTZipCompress(pZip, pvBuf, cbToWrite);
939	if (RT_SUCCESS(rc))
940	rc = RTZipCompFinish(pZip);
941	RTZipCompDestroy(pZip);
942	if (RT_SUCCESS(rc))
943	{
944	Assert( DeflateState.iOffset > 0
945	&& (size_t)DeflateState.iOffset <= DeflateState.cbCompGrain);
946
947	/* pad with zeroes to get to a full sector size */
948	uint32_t uSize = DeflateState.iOffset;
949	if (uSize % 512)
950	{
951	uint32_t uSizeAlign = RT_ALIGN(uSize, 512);
952	memset((uint8_t *)pExtent->pvCompGrain + uSize, '\0',
953	uSizeAlign - uSize);
954	uSize = uSizeAlign;
955	}
956
957	if (pcbMarkerData)
958	*pcbMarkerData = uSize;
959
960	/* Compressed grain marker. Data follows immediately. */
961	VMDKMARKER pMarker = (VMDKMARKER )pExtent->pvCompGrain;
962	pMarker->uSector = RT_H2LE_U64(uLBA);
963	pMarker->cbSize = RT_H2LE_U32( DeflateState.iOffset
964	- RT_UOFFSETOF(VMDKMARKER, uType));
965	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
966	uOffset, pMarker, uSize);
967	if (RT_FAILURE(rc))
968	return rc;
969	}
970	return rc;
971	}
972
973
974	/**
975	* Internal: check if all files are closed, prevent leaking resources.
976	*/
977	static int vmdkFileCheckAllClose(PVMDKIMAGE pImage)
978	{
979	int rc = VINF_SUCCESS, rc2;
980	PVMDKFILE pVmdkFile;
981
982	Assert(pImage->pFiles == NULL);
983	for (pVmdkFile = pImage->pFiles;
984	pVmdkFile != NULL;
985	pVmdkFile = pVmdkFile->pNext)
986	{
987	LogRel(("VMDK: leaking reference to file \"%s\"\n",
988	pVmdkFile->pszFilename));
989	pImage->pFiles = pVmdkFile->pNext;
990
991	rc2 = vmdkFileClose(pImage, &pVmdkFile, pVmdkFile->fDelete);
992
993	if (RT_SUCCESS(rc))
994	rc = rc2;
995	}
996	return rc;
997	}
998
999	/**
1000	* Internal: truncate a string (at a UTF8 code point boundary) and encode the
1001	* critical non-ASCII characters.
1002	*/
1003	static char vmdkEncodeString(const char psz)
1004	{
1005	char szEnc[VMDK_ENCODED_COMMENT_MAX + 3];
1006	char *pszDst = szEnc;
1007
1008	AssertPtr(psz);
1009
1010	for (; *psz; psz = RTStrNextCp(psz))
1011	{
1012	char *pszDstPrev = pszDst;
1013	RTUNICP Cp = RTStrGetCp(psz);
1014	if (Cp == '\\')
1015	{
1016	pszDst = RTStrPutCp(pszDst, Cp);
1017	pszDst = RTStrPutCp(pszDst, Cp);
1018	}
1019	else if (Cp == '\n')
1020	{
1021	pszDst = RTStrPutCp(pszDst, '\\');
1022	pszDst = RTStrPutCp(pszDst, 'n');
1023	}
1024	else if (Cp == '\r')
1025	{
1026	pszDst = RTStrPutCp(pszDst, '\\');
1027	pszDst = RTStrPutCp(pszDst, 'r');
1028	}
1029	else
1030	pszDst = RTStrPutCp(pszDst, Cp);
1031	if (pszDst - szEnc >= VMDK_ENCODED_COMMENT_MAX - 1)
1032	{
1033	pszDst = pszDstPrev;
1034	break;
1035	}
1036	}
1037	*pszDst = '\0';
1038	return RTStrDup(szEnc);
1039	}
1040
1041	/**
1042	* Internal: decode a string and store it into the specified string.
1043	*/
1044	static int vmdkDecodeString(const char pszEncoded, char psz, size_t cb)
1045	{
1046	int rc = VINF_SUCCESS;
1047	char szBuf[4];
1048
1049	if (!cb)
1050	return VERR_BUFFER_OVERFLOW;
1051
1052	AssertPtr(psz);
1053
1054	for (; *pszEncoded; pszEncoded = RTStrNextCp(pszEncoded))
1055	{
1056	char *pszDst = szBuf;
1057	RTUNICP Cp = RTStrGetCp(pszEncoded);
1058	if (Cp == '\\')
1059	{
1060	pszEncoded = RTStrNextCp(pszEncoded);
1061	RTUNICP CpQ = RTStrGetCp(pszEncoded);
1062	if (CpQ == 'n')
1063	RTStrPutCp(pszDst, '\n');
1064	else if (CpQ == 'r')
1065	RTStrPutCp(pszDst, '\r');
1066	else if (CpQ == '\0')
1067	{
1068	rc = VERR_VD_VMDK_INVALID_HEADER;
1069	break;
1070	}
1071	else
1072	RTStrPutCp(pszDst, CpQ);
1073	}
1074	else
1075	pszDst = RTStrPutCp(pszDst, Cp);
1076
1077	/* Need to leave space for terminating NUL. */
1078	if ((size_t)(pszDst - szBuf) + 1 >= cb)
1079	{
1080	rc = VERR_BUFFER_OVERFLOW;
1081	break;
1082	}
1083	memcpy(psz, szBuf, pszDst - szBuf);
1084	psz += pszDst - szBuf;
1085	}
1086	*psz = '\0';
1087	return rc;
1088	}
1089
1090	/**
1091	* Internal: free all buffers associated with grain directories.
1092	*/
1093	static void vmdkFreeGrainDirectory(PVMDKEXTENT pExtent)
1094	{
1095	if (pExtent->pGD)
1096	{
1097	RTMemFree(pExtent->pGD);
1098	pExtent->pGD = NULL;
1099	}
1100	if (pExtent->pRGD)
1101	{
1102	RTMemFree(pExtent->pRGD);
1103	pExtent->pRGD = NULL;
1104	}
1105	}
1106
1107	/**
1108	* Internal: allocate the compressed/uncompressed buffers for streamOptimized
1109	* images.
1110	*/
1111	static int vmdkAllocStreamBuffers(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
1112	{
1113	int rc = VINF_SUCCESS;
1114
1115	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
1116	{
1117	/* streamOptimized extents need a compressed grain buffer, which must
1118	* be big enough to hold uncompressible data (which needs ~8 bytes
1119	* more than the uncompressed data), the marker and padding. */
1120	pExtent->cbCompGrain = RT_ALIGN_Z( VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain)
1121	+ 8 + sizeof(VMDKMARKER), 512);
1122	pExtent->pvCompGrain = RTMemAlloc(pExtent->cbCompGrain);
1123	if (RT_LIKELY(pExtent->pvCompGrain))
1124	{
1125	/* streamOptimized extents need a decompressed grain buffer. */
1126	pExtent->pvGrain = RTMemAlloc(VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
1127	if (!pExtent->pvGrain)
1128	rc = VERR_NO_MEMORY;
1129	}
1130	else
1131	rc = VERR_NO_MEMORY;
1132	}
1133
1134	if (RT_FAILURE(rc))
1135	vmdkFreeStreamBuffers(pExtent);
1136	return rc;
1137	}
1138
1139	/**
1140	* Internal: allocate all buffers associated with grain directories.
1141	*/
1142	static int vmdkAllocGrainDirectory(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
1143	{
1144	RT_NOREF1(pImage);
1145	int rc = VINF_SUCCESS;
1146	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
1147
1148	pExtent->pGD = (uint32_t *)RTMemAllocZ(cbGD);
1149	if (RT_LIKELY(pExtent->pGD))
1150	{
1151	if (pExtent->uSectorRGD)
1152	{
1153	pExtent->pRGD = (uint32_t *)RTMemAllocZ(cbGD);
1154	if (RT_UNLIKELY(!pExtent->pRGD))
1155	rc = VERR_NO_MEMORY;
1156	}
1157	}
1158	else
1159	rc = VERR_NO_MEMORY;
1160
1161	if (RT_FAILURE(rc))
1162	vmdkFreeGrainDirectory(pExtent);
1163	return rc;
1164	}
1165
1166	/**
1167	* Converts the grain directory from little to host endianess.
1168	*
1169	* @param pGD The grain directory.
1170	* @param cGDEntries Number of entries in the grain directory to convert.
1171	*/
1172	DECLINLINE(void) vmdkGrainDirectoryConvToHost(uint32_t *pGD, uint32_t cGDEntries)
1173	{
1174	uint32_t *pGDTmp = pGD;
1175
1176	for (uint32_t i = 0; i < cGDEntries; i++, pGDTmp++)
1177	pGDTmp = RT_LE2H_U32(pGDTmp);
1178	}
1179
1180	/**
1181	* Read the grain directory and allocated grain tables verifying them against
1182	* their back up copies if available.
1183	*
1184	* @returns VBox status code.
1185	* @param pImage Image instance data.
1186	* @param pExtent The VMDK extent.
1187	*/
1188	static int vmdkReadGrainDirectory(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
1189	{
1190	int rc = VINF_SUCCESS;
1191	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
1192
1193	AssertReturn(( pExtent->enmType == VMDKETYPE_HOSTED_SPARSE
1194	&& pExtent->uSectorGD != VMDK_GD_AT_END
1195	&& pExtent->uSectorRGD != VMDK_GD_AT_END), VERR_INTERNAL_ERROR);
1196
1197	rc = vmdkAllocGrainDirectory(pImage, pExtent);
1198	if (RT_SUCCESS(rc))
1199	{
1200	/* The VMDK 1.1 spec seems to talk about compressed grain directories,
1201	* but in reality they are not compressed. */
1202	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1203	VMDK_SECTOR2BYTE(pExtent->uSectorGD),
1204	pExtent->pGD, cbGD);
1205	if (RT_SUCCESS(rc))
1206	{
1207	vmdkGrainDirectoryConvToHost(pExtent->pGD, pExtent->cGDEntries);
1208
1209	if ( pExtent->uSectorRGD
1210	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_SKIP_CONSISTENCY_CHECKS))
1211	{
1212	/* The VMDK 1.1 spec seems to talk about compressed grain directories,
1213	* but in reality they are not compressed. */
1214	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1215	VMDK_SECTOR2BYTE(pExtent->uSectorRGD),
1216	pExtent->pRGD, cbGD);
1217	if (RT_SUCCESS(rc))
1218	{
1219	vmdkGrainDirectoryConvToHost(pExtent->pRGD, pExtent->cGDEntries);
1220
1221	/* Check grain table and redundant grain table for consistency. */
1222	size_t cbGT = pExtent->cGTEntries * sizeof(uint32_t);
1223	size_t cbGTBuffers = cbGT; /* Start with space for one GT. */
1224	size_t cbGTBuffersMax = _1M;
1225
1226	uint32_t pTmpGT1 = (uint32_t )RTMemAlloc(cbGTBuffers);
1227	uint32_t pTmpGT2 = (uint32_t )RTMemAlloc(cbGTBuffers);
1228
1229	if ( !pTmpGT1
1230	\|\| !pTmpGT2)
1231	rc = VERR_NO_MEMORY;
1232
1233	size_t i = 0;
1234	uint32_t *pGDTmp = pExtent->pGD;
1235	uint32_t *pRGDTmp = pExtent->pRGD;
1236
1237	/* Loop through all entries. */
1238	while (i < pExtent->cGDEntries)
1239	{
1240	uint32_t uGTStart = *pGDTmp;
1241	uint32_t uRGTStart = *pRGDTmp;
1242	size_t cbGTRead = cbGT;
1243
1244	/* If no grain table is allocated skip the entry. */
1245	if (pGDTmp == 0 && pRGDTmp == 0)
1246	{
1247	i++;
1248	continue;
1249	}
1250
1251	if (pGDTmp == 0 \|\| pRGDTmp == 0 \|\| pGDTmp == pRGDTmp)
1252	{
1253	/* Just one grain directory entry refers to a not yet allocated
1254	* grain table or both grain directory copies refer to the same
1255	* grain table. Not allowed. */
1256	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1257	N_("VMDK: inconsistent references to grain directory in '%s'"), pExtent->pszFullname);
1258	break;
1259	}
1260
1261	i++;
1262	pGDTmp++;
1263	pRGDTmp++;
1264
1265	/*
1266	* Read a few tables at once if adjacent to decrease the number
1267	* of I/O requests. Read at maximum 1MB at once.
1268	*/
1269	while ( i < pExtent->cGDEntries
1270	&& cbGTRead < cbGTBuffersMax)
1271	{
1272	/* If no grain table is allocated skip the entry. */
1273	if (pGDTmp == 0 && pRGDTmp == 0)
1274	{
1275	i++;
1276	continue;
1277	}
1278
1279	if (pGDTmp == 0 \|\| pRGDTmp == 0 \|\| pGDTmp == pRGDTmp)
1280	{
1281	/* Just one grain directory entry refers to a not yet allocated
1282	* grain table or both grain directory copies refer to the same
1283	* grain table. Not allowed. */
1284	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1285	N_("VMDK: inconsistent references to grain directory in '%s'"), pExtent->pszFullname);
1286	break;
1287	}
1288
1289	/* Check that the start offsets are adjacent.*/
1290	if ( VMDK_SECTOR2BYTE(uGTStart) + cbGTRead != VMDK_SECTOR2BYTE(*pGDTmp)
1291	\|\| VMDK_SECTOR2BYTE(uRGTStart) + cbGTRead != VMDK_SECTOR2BYTE(*pRGDTmp))
1292	break;
1293
1294	i++;
1295	pGDTmp++;
1296	pRGDTmp++;
1297	cbGTRead += cbGT;
1298	}
1299
1300	/* Increase buffers if required. */
1301	if ( RT_SUCCESS(rc)
1302	&& cbGTBuffers < cbGTRead)
1303	{
1304	uint32_t *pTmp;
1305	pTmp = (uint32_t *)RTMemRealloc(pTmpGT1, cbGTRead);
1306	if (pTmp)
1307	{
1308	pTmpGT1 = pTmp;
1309	pTmp = (uint32_t *)RTMemRealloc(pTmpGT2, cbGTRead);
1310	if (pTmp)
1311	pTmpGT2 = pTmp;
1312	else
1313	rc = VERR_NO_MEMORY;
1314	}
1315	else
1316	rc = VERR_NO_MEMORY;
1317
1318	if (rc == VERR_NO_MEMORY)
1319	{
1320	/* Reset to the old values. */
1321	rc = VINF_SUCCESS;
1322	i -= cbGTRead / cbGT;
1323	cbGTRead = cbGT;
1324
1325	/* Don't try to increase the buffer again in the next run. */
1326	cbGTBuffersMax = cbGTBuffers;
1327	}
1328	}
1329
1330	if (RT_SUCCESS(rc))
1331	{
1332	/* The VMDK 1.1 spec seems to talk about compressed grain tables,
1333	* but in reality they are not compressed. */
1334	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1335	VMDK_SECTOR2BYTE(uGTStart),
1336	pTmpGT1, cbGTRead);
1337	if (RT_FAILURE(rc))
1338	{
1339	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1340	N_("VMDK: error reading grain table in '%s'"), pExtent->pszFullname);
1341	break;
1342	}
1343	/* The VMDK 1.1 spec seems to talk about compressed grain tables,
1344	* but in reality they are not compressed. */
1345	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1346	VMDK_SECTOR2BYTE(uRGTStart),
1347	pTmpGT2, cbGTRead);
1348	if (RT_FAILURE(rc))
1349	{
1350	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1351	N_("VMDK: error reading backup grain table in '%s'"), pExtent->pszFullname);
1352	break;
1353	}
1354	if (memcmp(pTmpGT1, pTmpGT2, cbGTRead))
1355	{
1356	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1357	N_("VMDK: inconsistency between grain table and backup grain table in '%s'"), pExtent->pszFullname);
1358	break;
1359	}
1360	}
1361	} /* while (i < pExtent->cGDEntries) */
1362
1363	/** @todo figure out what to do for unclean VMDKs. */
1364	if (pTmpGT1)
1365	RTMemFree(pTmpGT1);
1366	if (pTmpGT2)
1367	RTMemFree(pTmpGT2);
1368	}
1369	else
1370	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1371	N_("VMDK: could not read redundant grain directory in '%s'"), pExtent->pszFullname);
1372	}
1373	}
1374	else
1375	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1376	N_("VMDK: could not read grain directory in '%s': %Rrc"), pExtent->pszFullname, rc);
1377	}
1378
1379	if (RT_FAILURE(rc))
1380	vmdkFreeGrainDirectory(pExtent);
1381	return rc;
1382	}
1383
1384	/**
1385	* Creates a new grain directory for the given extent at the given start sector.
1386	*
1387	* @returns VBox status code.
1388	* @param pImage Image instance data.
1389	* @param pExtent The VMDK extent.
1390	* @param uStartSector Where the grain directory should be stored in the image.
1391	* @param fPreAlloc Flag whether to pre allocate the grain tables at this point.
1392	*/
1393	static int vmdkCreateGrainDirectory(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
1394	uint64_t uStartSector, bool fPreAlloc)
1395	{
1396	int rc = VINF_SUCCESS;
1397	unsigned i;
1398	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
1399	size_t cbGDRounded = RT_ALIGN_64(cbGD, 512);
1400	size_t cbGTRounded;
1401	uint64_t cbOverhead;
1402
1403	if (fPreAlloc)
1404	{
1405	cbGTRounded = RT_ALIGN_64(pExtent->cGDEntries * pExtent->cGTEntries * sizeof(uint32_t), 512);
1406	cbOverhead = VMDK_SECTOR2BYTE(uStartSector) + cbGDRounded + cbGTRounded;
1407	}
1408	else
1409	{
1410	/* Use a dummy start sector for layout computation. */
1411	if (uStartSector == VMDK_GD_AT_END)
1412	uStartSector = 1;
1413	cbGTRounded = 0;
1414	cbOverhead = VMDK_SECTOR2BYTE(uStartSector) + cbGDRounded;
1415	}
1416
1417	/* For streamOptimized extents there is only one grain directory,
1418	* and for all others take redundant grain directory into account. */
1419	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
1420	{
1421	cbOverhead = RT_ALIGN_64(cbOverhead,
1422	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
1423	}
1424	else
1425	{
1426	cbOverhead += cbGDRounded + cbGTRounded;
1427	cbOverhead = RT_ALIGN_64(cbOverhead,
1428	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
1429	rc = vdIfIoIntFileSetSize(pImage->pIfIo, pExtent->pFile->pStorage, cbOverhead);
1430	}
1431
1432	if (RT_SUCCESS(rc))
1433	{
1434	pExtent->uAppendPosition = cbOverhead;
1435	pExtent->cOverheadSectors = VMDK_BYTE2SECTOR(cbOverhead);
1436
1437	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
1438	{
1439	pExtent->uSectorRGD = 0;
1440	pExtent->uSectorGD = uStartSector;
1441	}
1442	else
1443	{
1444	pExtent->uSectorRGD = uStartSector;
1445	pExtent->uSectorGD = uStartSector + VMDK_BYTE2SECTOR(cbGDRounded + cbGTRounded);
1446	}
1447
1448	rc = vmdkAllocStreamBuffers(pImage, pExtent);
1449	if (RT_SUCCESS(rc))
1450	{
1451	rc = vmdkAllocGrainDirectory(pImage, pExtent);
1452	if ( RT_SUCCESS(rc)
1453	&& fPreAlloc)
1454	{
1455	uint32_t uGTSectorLE;
1456	uint64_t uOffsetSectors;
1457
1458	if (pExtent->pRGD)
1459	{
1460	uOffsetSectors = pExtent->uSectorRGD + VMDK_BYTE2SECTOR(cbGDRounded);
1461	for (i = 0; i < pExtent->cGDEntries; i++)
1462	{
1463	pExtent->pRGD[i] = uOffsetSectors;
1464	uGTSectorLE = RT_H2LE_U64(uOffsetSectors);
1465	/* Write the redundant grain directory entry to disk. */
1466	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
1467	VMDK_SECTOR2BYTE(pExtent->uSectorRGD) + i * sizeof(uGTSectorLE),
1468	&uGTSectorLE, sizeof(uGTSectorLE));
1469	if (RT_FAILURE(rc))
1470	{
1471	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write new redundant grain directory entry in '%s'"), pExtent->pszFullname);
1472	break;
1473	}
1474	uOffsetSectors += VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
1475	}
1476	}
1477
1478	if (RT_SUCCESS(rc))
1479	{
1480	uOffsetSectors = pExtent->uSectorGD + VMDK_BYTE2SECTOR(cbGDRounded);
1481	for (i = 0; i < pExtent->cGDEntries; i++)
1482	{
1483	pExtent->pGD[i] = uOffsetSectors;
1484	uGTSectorLE = RT_H2LE_U64(uOffsetSectors);
1485	/* Write the grain directory entry to disk. */
1486	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
1487	VMDK_SECTOR2BYTE(pExtent->uSectorGD) + i * sizeof(uGTSectorLE),
1488	&uGTSectorLE, sizeof(uGTSectorLE));
1489	if (RT_FAILURE(rc))
1490	{
1491	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write new grain directory entry in '%s'"), pExtent->pszFullname);
1492	break;
1493	}
1494	uOffsetSectors += VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
1495	}
1496	}
1497	}
1498	}
1499	}
1500
1501	if (RT_FAILURE(rc))
1502	vmdkFreeGrainDirectory(pExtent);
1503	return rc;
1504	}
1505
1506	/**
1507	* Unquotes the given string returning the result in a separate buffer.
1508	*
1509	* @returns VBox status code.
1510	* @param pImage The VMDK image state.
1511	* @param pszStr The string to unquote.
1512	* @param ppszUnquoted Where to store the return value, use RTMemTmpFree to
1513	* free.
1514	* @param ppszNext Where to store the pointer to any character following
1515	* the quoted value, optional.
1516	*/
1517	static int vmdkStringUnquote(PVMDKIMAGE pImage, const char *pszStr,
1518	char ppszUnquoted, char ppszNext)
1519	{
1520	const char *pszStart = pszStr;
1521	char *pszQ;
1522	char *pszUnquoted;
1523
1524	/* Skip over whitespace. */
1525	while (pszStr == ' ' \|\| pszStr == '\t')
1526	pszStr++;
1527
1528	if (*pszStr != '"')
1529	{
1530	pszQ = (char *)pszStr;
1531	while (pszQ && pszQ != ' ' && *pszQ != '\t')
1532	pszQ++;
1533	}
1534	else
1535	{
1536	pszStr++;
1537	pszQ = (char *)strchr(pszStr, '"');
1538	if (pszQ == NULL)
1539	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrectly quoted value in descriptor in '%s' (raw value %s)"),
1540	pImage->pszFilename, pszStart);
1541	}
1542
1543	pszUnquoted = (char *)RTMemTmpAlloc(pszQ - pszStr + 1);
1544	if (!pszUnquoted)
1545	return VERR_NO_MEMORY;
1546	memcpy(pszUnquoted, pszStr, pszQ - pszStr);
1547	pszUnquoted[pszQ - pszStr] = '\0';
1548	*ppszUnquoted = pszUnquoted;
1549	if (ppszNext)
1550	*ppszNext = pszQ + 1;
1551	return VINF_SUCCESS;
1552	}
1553
1554	static int vmdkDescInitStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1555	const char *pszLine)
1556	{
1557	char *pEnd = pDescriptor->aLines[pDescriptor->cLines];
1558	ssize_t cbDiff = strlen(pszLine) + 1;
1559
1560	if ( pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1
1561	&& pEnd - pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff)
1562	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1563
1564	memcpy(pEnd, pszLine, cbDiff);
1565	pDescriptor->cLines++;
1566	pDescriptor->aLines[pDescriptor->cLines] = pEnd + cbDiff;
1567	pDescriptor->fDirty = true;
1568
1569	return VINF_SUCCESS;
1570	}
1571
1572	static bool vmdkDescGetStr(PVMDKDESCRIPTOR pDescriptor, unsigned uStart,
1573	const char pszKey, const char *ppszValue)
1574	{
1575	size_t cbKey = strlen(pszKey);
1576	const char *pszValue;
1577
1578	while (uStart != 0)
1579	{
1580	if (!strncmp(pDescriptor->aLines[uStart], pszKey, cbKey))
1581	{
1582	/* Key matches, check for a '=' (preceded by whitespace). */
1583	pszValue = pDescriptor->aLines[uStart] + cbKey;
1584	while (pszValue == ' ' \|\| pszValue == '\t')
1585	pszValue++;
1586	if (*pszValue == '=')
1587	{
1588	*ppszValue = pszValue + 1;
1589	break;
1590	}
1591	}
1592	uStart = pDescriptor->aNextLines[uStart];
1593	}
1594	return !!uStart;
1595	}
1596
1597	static int vmdkDescSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1598	unsigned uStart,
1599	const char pszKey, const char pszValue)
1600	{
1601	char pszTmp = NULL; / (MSC naturally cannot figure this isn't used uninitialized) */
1602	size_t cbKey = strlen(pszKey);
1603	unsigned uLast = 0;
1604
1605	while (uStart != 0)
1606	{
1607	if (!strncmp(pDescriptor->aLines[uStart], pszKey, cbKey))
1608	{
1609	/* Key matches, check for a '=' (preceded by whitespace). */
1610	pszTmp = pDescriptor->aLines[uStart] + cbKey;
1611	while (pszTmp == ' ' \|\| pszTmp == '\t')
1612	pszTmp++;
1613	if (*pszTmp == '=')
1614	{
1615	pszTmp++;
1616	/** @todo r=bird: Doesn't skipping trailing blanks here just cause unecessary
1617	* bloat and potentially out of space error? */
1618	while (pszTmp == ' ' \|\| pszTmp == '\t')
1619	pszTmp++;
1620	break;
1621	}
1622	}
1623	if (!pDescriptor->aNextLines[uStart])
1624	uLast = uStart;
1625	uStart = pDescriptor->aNextLines[uStart];
1626	}
1627	if (uStart)
1628	{
1629	if (pszValue)
1630	{
1631	/* Key already exists, replace existing value. */
1632	size_t cbOldVal = strlen(pszTmp);
1633	size_t cbNewVal = strlen(pszValue);
1634	ssize_t cbDiff = cbNewVal - cbOldVal;
1635	/* Check for buffer overflow. */
1636	if ( pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[0]
1637	> (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff)
1638	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1639
1640	memmove(pszTmp + cbNewVal, pszTmp + cbOldVal,
1641	pDescriptor->aLines[pDescriptor->cLines] - pszTmp - cbOldVal);
1642	memcpy(pszTmp, pszValue, cbNewVal + 1);
1643	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1644	pDescriptor->aLines[i] += cbDiff;
1645	}
1646	else
1647	{
1648	memmove(pDescriptor->aLines[uStart], pDescriptor->aLines[uStart+1],
1649	pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[uStart+1] + 1);
1650	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1651	{
1652	pDescriptor->aLines[i-1] = pDescriptor->aLines[i];
1653	if (pDescriptor->aNextLines[i])
1654	pDescriptor->aNextLines[i-1] = pDescriptor->aNextLines[i] - 1;
1655	else
1656	pDescriptor->aNextLines[i-1] = 0;
1657	}
1658	pDescriptor->cLines--;
1659	/* Adjust starting line numbers of following descriptor sections. */
1660	if (uStart < pDescriptor->uFirstExtent)
1661	pDescriptor->uFirstExtent--;
1662	if (uStart < pDescriptor->uFirstDDB)
1663	pDescriptor->uFirstDDB--;
1664	}
1665	}
1666	else
1667	{
1668	/* Key doesn't exist, append after the last entry in this category. */
1669	if (!pszValue)
1670	{
1671	/* Key doesn't exist, and it should be removed. Simply a no-op. */
1672	return VINF_SUCCESS;
1673	}
1674	cbKey = strlen(pszKey);
1675	size_t cbValue = strlen(pszValue);
1676	ssize_t cbDiff = cbKey + 1 + cbValue + 1;
1677	/* Check for buffer overflow. */
1678	if ( (pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1)
1679	\|\| ( pDescriptor->aLines[pDescriptor->cLines]
1680	- pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff))
1681	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1682	for (unsigned i = pDescriptor->cLines + 1; i > uLast + 1; i--)
1683	{
1684	pDescriptor->aLines[i] = pDescriptor->aLines[i - 1];
1685	if (pDescriptor->aNextLines[i - 1])
1686	pDescriptor->aNextLines[i] = pDescriptor->aNextLines[i - 1] + 1;
1687	else
1688	pDescriptor->aNextLines[i] = 0;
1689	}
1690	uStart = uLast + 1;
1691	pDescriptor->aNextLines[uLast] = uStart;
1692	pDescriptor->aNextLines[uStart] = 0;
1693	pDescriptor->cLines++;
1694	pszTmp = pDescriptor->aLines[uStart];
1695	memmove(pszTmp + cbDiff, pszTmp,
1696	pDescriptor->aLines[pDescriptor->cLines] - pszTmp);
1697	memcpy(pDescriptor->aLines[uStart], pszKey, cbKey);
1698	pDescriptor->aLines[uStart][cbKey] = '=';
1699	memcpy(pDescriptor->aLines[uStart] + cbKey + 1, pszValue, cbValue + 1);
1700	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1701	pDescriptor->aLines[i] += cbDiff;
1702
1703	/* Adjust starting line numbers of following descriptor sections. */
1704	if (uStart <= pDescriptor->uFirstExtent)
1705	pDescriptor->uFirstExtent++;
1706	if (uStart <= pDescriptor->uFirstDDB)
1707	pDescriptor->uFirstDDB++;
1708	}
1709	pDescriptor->fDirty = true;
1710	return VINF_SUCCESS;
1711	}
1712
1713	static int vmdkDescBaseGetU32(PVMDKDESCRIPTOR pDescriptor, const char *pszKey,
1714	uint32_t *puValue)
1715	{
1716	const char *pszValue;
1717
1718	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDesc, pszKey,
1719	&pszValue))
1720	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1721	return RTStrToUInt32Ex(pszValue, NULL, 10, puValue);
1722	}
1723
1724	/**
1725	* Returns the value of the given key as a string allocating the necessary memory.
1726	*
1727	* @returns VBox status code.
1728	* @retval VERR_VD_VMDK_VALUE_NOT_FOUND if the value could not be found.
1729	* @param pImage The VMDK image state.
1730	* @param pDescriptor The descriptor to fetch the value from.
1731	* @param pszKey The key to get the value from.
1732	* @param ppszValue Where to store the return value, use RTMemTmpFree to
1733	* free.
1734	*/
1735	static int vmdkDescBaseGetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1736	const char pszKey, char *ppszValue)
1737	{
1738	const char *pszValue;
1739	char *pszValueUnquoted;
1740
1741	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDesc, pszKey,
1742	&pszValue))
1743	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1744	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1745	if (RT_FAILURE(rc))
1746	return rc;
1747	*ppszValue = pszValueUnquoted;
1748	return rc;
1749	}
1750
1751	static int vmdkDescBaseSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1752	const char pszKey, const char pszValue)
1753	{
1754	char *pszValueQuoted;
1755
1756	RTStrAPrintf(&pszValueQuoted, "\"%s\"", pszValue);
1757	if (!pszValueQuoted)
1758	return VERR_NO_STR_MEMORY;
1759	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc, pszKey,
1760	pszValueQuoted);
1761	RTStrFree(pszValueQuoted);
1762	return rc;
1763	}
1764
1765	static void vmdkDescExtRemoveDummy(PVMDKIMAGE pImage,
1766	PVMDKDESCRIPTOR pDescriptor)
1767	{
1768	RT_NOREF1(pImage);
1769	unsigned uEntry = pDescriptor->uFirstExtent;
1770	ssize_t cbDiff;
1771
1772	if (!uEntry)
1773	return;
1774
1775	cbDiff = strlen(pDescriptor->aLines[uEntry]) + 1;
1776	/* Move everything including \0 in the entry marking the end of buffer. */
1777	memmove(pDescriptor->aLines[uEntry], pDescriptor->aLines[uEntry + 1],
1778	pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[uEntry + 1] + 1);
1779	for (unsigned i = uEntry + 1; i <= pDescriptor->cLines; i++)
1780	{
1781	pDescriptor->aLines[i - 1] = pDescriptor->aLines[i] - cbDiff;
1782	if (pDescriptor->aNextLines[i])
1783	pDescriptor->aNextLines[i - 1] = pDescriptor->aNextLines[i] - 1;
1784	else
1785	pDescriptor->aNextLines[i - 1] = 0;
1786	}
1787	pDescriptor->cLines--;
1788	if (pDescriptor->uFirstDDB)
1789	pDescriptor->uFirstDDB--;
1790
1791	return;
1792	}
1793
1794	static void vmdkDescExtRemoveByLine(PVMDKIMAGE pImage,
1795	PVMDKDESCRIPTOR pDescriptor, unsigned uLine)
1796	{
1797	RT_NOREF1(pImage);
1798	unsigned uEntry = uLine;
1799	ssize_t cbDiff;
1800	if (!uEntry)
1801	return;
1802	cbDiff = strlen(pDescriptor->aLines[uEntry]) + 1;
1803	/* Move everything including \0 in the entry marking the end of buffer. */
1804	memmove(pDescriptor->aLines[uEntry], pDescriptor->aLines[uEntry + 1],
1805	pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[uEntry + 1] + 1);
1806	for (unsigned i = uEntry; i <= pDescriptor->cLines; i++)
1807	{
1808	if (i != uEntry)
1809	pDescriptor->aLines[i - 1] = pDescriptor->aLines[i] - cbDiff;
1810	if (pDescriptor->aNextLines[i])
1811	pDescriptor->aNextLines[i - 1] = pDescriptor->aNextLines[i] - 1;
1812	else
1813	pDescriptor->aNextLines[i - 1] = 0;
1814	}
1815	pDescriptor->cLines--;
1816	if (pDescriptor->uFirstDDB)
1817	pDescriptor->uFirstDDB--;
1818	return;
1819	}
1820
1821	static int vmdkDescExtInsert(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1822	VMDKACCESS enmAccess, uint64_t cNominalSectors,
1823	VMDKETYPE enmType, const char *pszBasename,
1824	uint64_t uSectorOffset)
1825	{
1826	static const char *apszAccess[] = { "NOACCESS", "RDONLY", "RW" };
1827	static const char *apszType[] = { "", "SPARSE", "FLAT", "ZERO", "VMFS" };
1828	char *pszTmp;
1829	unsigned uStart = pDescriptor->uFirstExtent, uLast = 0;
1830	char szExt[1024];
1831	ssize_t cbDiff;
1832
1833	Assert((unsigned)enmAccess < RT_ELEMENTS(apszAccess));
1834	Assert((unsigned)enmType < RT_ELEMENTS(apszType));
1835
1836	/* Find last entry in extent description. */
1837	while (uStart)
1838	{
1839	if (!pDescriptor->aNextLines[uStart])
1840	uLast = uStart;
1841	uStart = pDescriptor->aNextLines[uStart];
1842	}
1843
1844	if (enmType == VMDKETYPE_ZERO)
1845	{
1846	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s ", apszAccess[enmAccess],
1847	cNominalSectors, apszType[enmType]);
1848	}
1849	else if (enmType == VMDKETYPE_FLAT)
1850	{
1851	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s \"%s\" %llu",
1852	apszAccess[enmAccess], cNominalSectors,
1853	apszType[enmType], pszBasename, uSectorOffset);
1854	}
1855	else
1856	{
1857	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s \"%s\"",
1858	apszAccess[enmAccess], cNominalSectors,
1859	apszType[enmType], pszBasename);
1860	}
1861	cbDiff = strlen(szExt) + 1;
1862
1863	/* Check for buffer overflow. */
1864	if ( (pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1)
1865	\|\| ( pDescriptor->aLines[pDescriptor->cLines]
1866	- pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff))
1867	{
1868	if ((pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
1869	&& !(pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1))
1870	{
1871	pImage->cbDescAlloc *= 2;
1872	pDescriptor->cbDescAlloc *= 2;
1873	}
1874	else
1875	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1876	}
1877
1878	for (unsigned i = pDescriptor->cLines + 1; i > uLast + 1; i--)
1879	{
1880	pDescriptor->aLines[i] = pDescriptor->aLines[i - 1];
1881	if (pDescriptor->aNextLines[i - 1])
1882	pDescriptor->aNextLines[i] = pDescriptor->aNextLines[i - 1] + 1;
1883	else
1884	pDescriptor->aNextLines[i] = 0;
1885	}
1886	uStart = uLast + 1;
1887	pDescriptor->aNextLines[uLast] = uStart;
1888	pDescriptor->aNextLines[uStart] = 0;
1889	pDescriptor->cLines++;
1890	pszTmp = pDescriptor->aLines[uStart];
1891	memmove(pszTmp + cbDiff, pszTmp,
1892	pDescriptor->aLines[pDescriptor->cLines] - pszTmp);
1893	memcpy(pDescriptor->aLines[uStart], szExt, cbDiff);
1894	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1895	pDescriptor->aLines[i] += cbDiff;
1896
1897	/* Adjust starting line numbers of following descriptor sections. */
1898	if (uStart <= pDescriptor->uFirstDDB)
1899	pDescriptor->uFirstDDB++;
1900
1901	pDescriptor->fDirty = true;
1902	return VINF_SUCCESS;
1903	}
1904
1905	/**
1906	* Returns the value of the given key from the DDB as a string allocating
1907	* the necessary memory.
1908	*
1909	* @returns VBox status code.
1910	* @retval VERR_VD_VMDK_VALUE_NOT_FOUND if the value could not be found.
1911	* @param pImage The VMDK image state.
1912	* @param pDescriptor The descriptor to fetch the value from.
1913	* @param pszKey The key to get the value from.
1914	* @param ppszValue Where to store the return value, use RTMemTmpFree to
1915	* free.
1916	*/
1917	static int vmdkDescDDBGetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1918	const char pszKey, char *ppszValue)
1919	{
1920	const char *pszValue;
1921	char *pszValueUnquoted;
1922
1923	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1924	&pszValue))
1925	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1926	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1927	if (RT_FAILURE(rc))
1928	return rc;
1929	*ppszValue = pszValueUnquoted;
1930	return rc;
1931	}
1932
1933	static int vmdkDescDDBGetU32(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1934	const char pszKey, uint32_t puValue)
1935	{
1936	const char *pszValue;
1937	char *pszValueUnquoted;
1938
1939	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1940	&pszValue))
1941	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1942	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1943	if (RT_FAILURE(rc))
1944	return rc;
1945	rc = RTStrToUInt32Ex(pszValueUnquoted, NULL, 10, puValue);
1946	RTMemTmpFree(pszValueUnquoted);
1947	return rc;
1948	}
1949
1950	static int vmdkDescDDBGetUuid(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1951	const char *pszKey, PRTUUID pUuid)
1952	{
1953	const char *pszValue;
1954	char *pszValueUnquoted;
1955
1956	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1957	&pszValue))
1958	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1959	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1960	if (RT_FAILURE(rc))
1961	return rc;
1962	rc = RTUuidFromStr(pUuid, pszValueUnquoted);
1963	RTMemTmpFree(pszValueUnquoted);
1964	return rc;
1965	}
1966
1967	static int vmdkDescDDBSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1968	const char pszKey, const char pszVal)
1969	{
1970	int rc;
1971	char *pszValQuoted;
1972
1973	if (pszVal)
1974	{
1975	RTStrAPrintf(&pszValQuoted, "\"%s\"", pszVal);
1976	if (!pszValQuoted)
1977	return VERR_NO_STR_MEMORY;
1978	}
1979	else
1980	pszValQuoted = NULL;
1981	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1982	pszValQuoted);
1983	if (pszValQuoted)
1984	RTStrFree(pszValQuoted);
1985	return rc;
1986	}
1987
1988	static int vmdkDescDDBSetUuid(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1989	const char *pszKey, PCRTUUID pUuid)
1990	{
1991	char *pszUuid;
1992
1993	RTStrAPrintf(&pszUuid, "\"%RTuuid\"", pUuid);
1994	if (!pszUuid)
1995	return VERR_NO_STR_MEMORY;
1996	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1997	pszUuid);
1998	RTStrFree(pszUuid);
1999	return rc;
2000	}
2001
2002	static int vmdkDescDDBSetU32(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
2003	const char *pszKey, uint32_t uValue)
2004	{
2005	char *pszValue;
2006
2007	RTStrAPrintf(&pszValue, "\"%d\"", uValue);
2008	if (!pszValue)
2009	return VERR_NO_STR_MEMORY;
2010	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
2011	pszValue);
2012	RTStrFree(pszValue);
2013	return rc;
2014	}
2015
2016	/**
2017	* Splits the descriptor data into individual lines checking for correct line
2018	* endings and descriptor size.
2019	*
2020	* @returns VBox status code.
2021	* @param pImage The image instance.
2022	* @param pDesc The descriptor.
2023	* @param pszTmp The raw descriptor data from the image.
2024	*/
2025	static int vmdkDescSplitLines(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDesc, char *pszTmp)
2026	{
2027	unsigned cLine = 0;
2028	int rc = VINF_SUCCESS;
2029
2030	while ( RT_SUCCESS(rc)
2031	&& *pszTmp != '\0')
2032	{
2033	pDesc->aLines[cLine++] = pszTmp;
2034	if (cLine >= VMDK_DESCRIPTOR_LINES_MAX)
2035	{
2036	vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
2037	rc = VERR_VD_VMDK_INVALID_HEADER;
2038	break;
2039	}
2040
2041	while (pszTmp != '\0' && pszTmp != '\n')
2042	{
2043	if (*pszTmp == '\r')
2044	{
2045	if (*(pszTmp + 1) != '\n')
2046	{
2047	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: unsupported end of line in descriptor in '%s'"), pImage->pszFilename);
2048	break;
2049	}
2050	else
2051	{
2052	/* Get rid of CR character. */
2053	*pszTmp = '\0';
2054	}
2055	}
2056	pszTmp++;
2057	}
2058
2059	if (RT_FAILURE(rc))
2060	break;
2061
2062	/* Get rid of LF character. */
2063	if (*pszTmp == '\n')
2064	{
2065	*pszTmp = '\0';
2066	pszTmp++;
2067	}
2068	}
2069
2070	if (RT_SUCCESS(rc))
2071	{
2072	pDesc->cLines = cLine;
2073	/* Pointer right after the end of the used part of the buffer. */
2074	pDesc->aLines[cLine] = pszTmp;
2075	}
2076
2077	return rc;
2078	}
2079
2080	static int vmdkPreprocessDescriptor(PVMDKIMAGE pImage, char *pDescData,
2081	size_t cbDescData, PVMDKDESCRIPTOR pDescriptor)
2082	{
2083	pDescriptor->cbDescAlloc = cbDescData;
2084	int rc = vmdkDescSplitLines(pImage, pDescriptor, pDescData);
2085	if (RT_SUCCESS(rc))
2086	{
2087	if ( strcmp(pDescriptor->aLines[0], "# Disk DescriptorFile")
2088	&& strcmp(pDescriptor->aLines[0], "# Disk Descriptor File")
2089	&& strcmp(pDescriptor->aLines[0], "#Disk Descriptor File")
2090	&& strcmp(pDescriptor->aLines[0], "#Disk DescriptorFile"))
2091	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2092	N_("VMDK: descriptor does not start as expected in '%s'"), pImage->pszFilename);
2093	else
2094	{
2095	unsigned uLastNonEmptyLine = 0;
2096
2097	/* Initialize those, because we need to be able to reopen an image. */
2098	pDescriptor->uFirstDesc = 0;
2099	pDescriptor->uFirstExtent = 0;
2100	pDescriptor->uFirstDDB = 0;
2101	for (unsigned i = 0; i < pDescriptor->cLines; i++)
2102	{
2103	if (pDescriptor->aLines[i] != '#' && pDescriptor->aLines[i] != '\0')
2104	{
2105	if ( !strncmp(pDescriptor->aLines[i], "RW", 2)
2106	\|\| !strncmp(pDescriptor->aLines[i], "RDONLY", 6)
2107	\|\| !strncmp(pDescriptor->aLines[i], "NOACCESS", 8) )
2108	{
2109	/* An extent descriptor. */
2110	if (!pDescriptor->uFirstDesc \|\| pDescriptor->uFirstDDB)
2111	{
2112	/* Incorrect ordering of entries. */
2113	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2114	N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
2115	break;
2116	}
2117	if (!pDescriptor->uFirstExtent)
2118	{
2119	pDescriptor->uFirstExtent = i;
2120	uLastNonEmptyLine = 0;
2121	}
2122	}
2123	else if (!strncmp(pDescriptor->aLines[i], "ddb.", 4))
2124	{
2125	/* A disk database entry. */
2126	if (!pDescriptor->uFirstDesc \|\| !pDescriptor->uFirstExtent)
2127	{
2128	/* Incorrect ordering of entries. */
2129	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2130	N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
2131	break;
2132	}
2133	if (!pDescriptor->uFirstDDB)
2134	{
2135	pDescriptor->uFirstDDB = i;
2136	uLastNonEmptyLine = 0;
2137	}
2138	}
2139	else
2140	{
2141	/* A normal entry. */
2142	if (pDescriptor->uFirstExtent \|\| pDescriptor->uFirstDDB)
2143	{
2144	/* Incorrect ordering of entries. */
2145	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2146	N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
2147	break;
2148	}
2149	if (!pDescriptor->uFirstDesc)
2150	{
2151	pDescriptor->uFirstDesc = i;
2152	uLastNonEmptyLine = 0;
2153	}
2154	}
2155	if (uLastNonEmptyLine)
2156	pDescriptor->aNextLines[uLastNonEmptyLine] = i;
2157	uLastNonEmptyLine = i;
2158	}
2159	}
2160	}
2161	}
2162
2163	return rc;
2164	}
2165
2166	static int vmdkDescSetPCHSGeometry(PVMDKIMAGE pImage,
2167	PCVDGEOMETRY pPCHSGeometry)
2168	{
2169	int rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2170	VMDK_DDB_GEO_PCHS_CYLINDERS,
2171	pPCHSGeometry->cCylinders);
2172	if (RT_FAILURE(rc))
2173	return rc;
2174	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2175	VMDK_DDB_GEO_PCHS_HEADS,
2176	pPCHSGeometry->cHeads);
2177	if (RT_FAILURE(rc))
2178	return rc;
2179	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2180	VMDK_DDB_GEO_PCHS_SECTORS,
2181	pPCHSGeometry->cSectors);
2182	return rc;
2183	}
2184
2185	static int vmdkDescSetLCHSGeometry(PVMDKIMAGE pImage,
2186	PCVDGEOMETRY pLCHSGeometry)
2187	{
2188	int rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2189	VMDK_DDB_GEO_LCHS_CYLINDERS,
2190	pLCHSGeometry->cCylinders);
2191	if (RT_FAILURE(rc))
2192	return rc;
2193	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2194	VMDK_DDB_GEO_LCHS_HEADS,
2195
2196	pLCHSGeometry->cHeads);
2197	if (RT_FAILURE(rc))
2198	return rc;
2199	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2200	VMDK_DDB_GEO_LCHS_SECTORS,
2201	pLCHSGeometry->cSectors);
2202	return rc;
2203	}
2204
2205	static int vmdkCreateDescriptor(PVMDKIMAGE pImage, char *pDescData,
2206	size_t cbDescData, PVMDKDESCRIPTOR pDescriptor)
2207	{
2208	pDescriptor->uFirstDesc = 0;
2209	pDescriptor->uFirstExtent = 0;
2210	pDescriptor->uFirstDDB = 0;
2211	pDescriptor->cLines = 0;
2212	pDescriptor->cbDescAlloc = cbDescData;
2213	pDescriptor->fDirty = false;
2214	pDescriptor->aLines[pDescriptor->cLines] = pDescData;
2215	memset(pDescriptor->aNextLines, '\0', sizeof(pDescriptor->aNextLines));
2216
2217	int rc = vmdkDescInitStr(pImage, pDescriptor, "# Disk DescriptorFile");
2218	if (RT_SUCCESS(rc))
2219	rc = vmdkDescInitStr(pImage, pDescriptor, "version=1");
2220	if (RT_SUCCESS(rc))
2221	{
2222	pDescriptor->uFirstDesc = pDescriptor->cLines - 1;
2223	rc = vmdkDescInitStr(pImage, pDescriptor, "");
2224	}
2225	if (RT_SUCCESS(rc))
2226	rc = vmdkDescInitStr(pImage, pDescriptor, "# Extent description");
2227	if (RT_SUCCESS(rc))
2228	rc = vmdkDescInitStr(pImage, pDescriptor, "NOACCESS 0 ZERO ");
2229	if (RT_SUCCESS(rc))
2230	{
2231	pDescriptor->uFirstExtent = pDescriptor->cLines - 1;
2232	rc = vmdkDescInitStr(pImage, pDescriptor, "");
2233	}
2234	if (RT_SUCCESS(rc))
2235	{
2236	/* The trailing space is created by VMware, too. */
2237	rc = vmdkDescInitStr(pImage, pDescriptor, "# The disk Data Base ");
2238	}
2239	if (RT_SUCCESS(rc))
2240	rc = vmdkDescInitStr(pImage, pDescriptor, "#DDB");
2241	if (RT_SUCCESS(rc))
2242	rc = vmdkDescInitStr(pImage, pDescriptor, "");
2243	if (RT_SUCCESS(rc))
2244	rc = vmdkDescInitStr(pImage, pDescriptor, "ddb.virtualHWVersion = \"4\"");
2245	if (RT_SUCCESS(rc))
2246	{
2247	pDescriptor->uFirstDDB = pDescriptor->cLines - 1;
2248
2249	/* Now that the framework is in place, use the normal functions to insert
2250	* the remaining keys. */
2251	char szBuf[9];
2252	RTStrPrintf(szBuf, sizeof(szBuf), "%08x", RTRandU32());
2253	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc,
2254	"CID", szBuf);
2255	}
2256	if (RT_SUCCESS(rc))
2257	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc,
2258	"parentCID", "ffffffff");
2259	if (RT_SUCCESS(rc))
2260	rc = vmdkDescDDBSetStr(pImage, pDescriptor, "ddb.adapterType", "ide");
2261
2262	return rc;
2263	}
2264
2265	static int vmdkParseDescriptor(PVMDKIMAGE pImage, char *pDescData, size_t cbDescData)
2266	{
2267	int rc;
2268	unsigned cExtents;
2269	unsigned uLine;
2270	unsigned i;
2271
2272	rc = vmdkPreprocessDescriptor(pImage, pDescData, cbDescData,
2273	&pImage->Descriptor);
2274	if (RT_FAILURE(rc))
2275	return rc;
2276
2277	/* Check version, must be 1. */
2278	uint32_t uVersion;
2279	rc = vmdkDescBaseGetU32(&pImage->Descriptor, "version", &uVersion);
2280	if (RT_FAILURE(rc))
2281	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error finding key 'version' in descriptor in '%s'"), pImage->pszFilename);
2282	if (uVersion != 1)
2283	return vdIfError(pImage->pIfError, VERR_VD_VMDK_UNSUPPORTED_VERSION, RT_SRC_POS, N_("VMDK: unsupported format version in descriptor in '%s'"), pImage->pszFilename);
2284
2285	/* Get image creation type and determine image flags. */
2286	char pszCreateType = NULL; / initialized to make gcc shut up */
2287	rc = vmdkDescBaseGetStr(pImage, &pImage->Descriptor, "createType",
2288	&pszCreateType);
2289	if (RT_FAILURE(rc))
2290	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot get image type from descriptor in '%s'"), pImage->pszFilename);
2291	if ( !strcmp(pszCreateType, "twoGbMaxExtentSparse")
2292	\|\| !strcmp(pszCreateType, "twoGbMaxExtentFlat"))
2293	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_SPLIT_2G;
2294	else if ( !strcmp(pszCreateType, "partitionedDevice")
2295	\|\| !strcmp(pszCreateType, "fullDevice"))
2296	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_RAWDISK;
2297	else if (!strcmp(pszCreateType, "streamOptimized"))
2298	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED;
2299	else if (!strcmp(pszCreateType, "vmfs"))
2300	pImage->uImageFlags \|= VD_IMAGE_FLAGS_FIXED \| VD_VMDK_IMAGE_FLAGS_ESX;
2301	RTMemTmpFree(pszCreateType);
2302
2303	/* Count the number of extent config entries. */
2304	for (uLine = pImage->Descriptor.uFirstExtent, cExtents = 0;
2305	uLine != 0;
2306	uLine = pImage->Descriptor.aNextLines[uLine], cExtents++)
2307	/* nothing */;
2308
2309	if (!pImage->pDescData && cExtents != 1)
2310	{
2311	/* Monolithic image, must have only one extent (already opened). */
2312	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);
2313	}
2314
2315	if (pImage->pDescData)
2316	{
2317	/* Non-monolithic image, extents need to be allocated. */
2318	rc = vmdkCreateExtents(pImage, cExtents);
2319	if (RT_FAILURE(rc))
2320	return rc;
2321	}
2322
2323	for (i = 0, uLine = pImage->Descriptor.uFirstExtent;
2324	i < cExtents; i++, uLine = pImage->Descriptor.aNextLines[uLine])
2325	{
2326	char *pszLine = pImage->Descriptor.aLines[uLine];
2327
2328	/* Access type of the extent. */
2329	if (!strncmp(pszLine, "RW", 2))
2330	{
2331	pImage->pExtents[i].enmAccess = VMDKACCESS_READWRITE;
2332	pszLine += 2;
2333	}
2334	else if (!strncmp(pszLine, "RDONLY", 6))
2335	{
2336	pImage->pExtents[i].enmAccess = VMDKACCESS_READONLY;
2337	pszLine += 6;
2338	}
2339	else if (!strncmp(pszLine, "NOACCESS", 8))
2340	{
2341	pImage->pExtents[i].enmAccess = VMDKACCESS_NOACCESS;
2342	pszLine += 8;
2343	}
2344	else
2345	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2346	if (*pszLine++ != ' ')
2347	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2348
2349	/* Nominal size of the extent. */
2350	rc = RTStrToUInt64Ex(pszLine, &pszLine, 10,
2351	&pImage->pExtents[i].cNominalSectors);
2352	if (RT_FAILURE(rc))
2353	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2354	if (*pszLine++ != ' ')
2355	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2356
2357	/* Type of the extent. */
2358	if (!strncmp(pszLine, "SPARSE", 6))
2359	{
2360	pImage->pExtents[i].enmType = VMDKETYPE_HOSTED_SPARSE;
2361	pszLine += 6;
2362	}
2363	else if (!strncmp(pszLine, "FLAT", 4))
2364	{
2365	pImage->pExtents[i].enmType = VMDKETYPE_FLAT;
2366	pszLine += 4;
2367	}
2368	else if (!strncmp(pszLine, "ZERO", 4))
2369	{
2370	pImage->pExtents[i].enmType = VMDKETYPE_ZERO;
2371	pszLine += 4;
2372	}
2373	else if (!strncmp(pszLine, "VMFS", 4))
2374	{
2375	pImage->pExtents[i].enmType = VMDKETYPE_VMFS;
2376	pszLine += 4;
2377	}
2378	else
2379	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2380
2381	if (pImage->pExtents[i].enmType == VMDKETYPE_ZERO)
2382	{
2383	/* This one has no basename or offset. */
2384	if (*pszLine == ' ')
2385	pszLine++;
2386	if (*pszLine != '\0')
2387	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2388	pImage->pExtents[i].pszBasename = NULL;
2389	}
2390	else
2391	{
2392	/* All other extent types have basename and optional offset. */
2393	if (*pszLine++ != ' ')
2394	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2395
2396	/* Basename of the image. Surrounded by quotes. */
2397	char *pszBasename;
2398	rc = vmdkStringUnquote(pImage, pszLine, &pszBasename, &pszLine);
2399	if (RT_FAILURE(rc))
2400	return rc;
2401	pImage->pExtents[i].pszBasename = pszBasename;
2402	if (*pszLine == ' ')
2403	{
2404	pszLine++;
2405	if (*pszLine != '\0')
2406	{
2407	/* Optional offset in extent specified. */
2408	rc = RTStrToUInt64Ex(pszLine, &pszLine, 10,
2409	&pImage->pExtents[i].uSectorOffset);
2410	if (RT_FAILURE(rc))
2411	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2412	}
2413	}
2414
2415	if (*pszLine != '\0')
2416	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2417	}
2418	}
2419
2420	/* Determine PCHS geometry (autogenerate if necessary). */
2421	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2422	VMDK_DDB_GEO_PCHS_CYLINDERS,
2423	&pImage->PCHSGeometry.cCylinders);
2424	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2425	pImage->PCHSGeometry.cCylinders = 0;
2426	else if (RT_FAILURE(rc))
2427	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2428	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2429	VMDK_DDB_GEO_PCHS_HEADS,
2430	&pImage->PCHSGeometry.cHeads);
2431	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2432	pImage->PCHSGeometry.cHeads = 0;
2433	else if (RT_FAILURE(rc))
2434	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2435	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2436	VMDK_DDB_GEO_PCHS_SECTORS,
2437	&pImage->PCHSGeometry.cSectors);
2438	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2439	pImage->PCHSGeometry.cSectors = 0;
2440	else if (RT_FAILURE(rc))
2441	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2442	if ( pImage->PCHSGeometry.cCylinders == 0
2443	\|\| pImage->PCHSGeometry.cHeads == 0
2444	\|\| pImage->PCHSGeometry.cHeads > 16
2445	\|\| pImage->PCHSGeometry.cSectors == 0
2446	\|\| pImage->PCHSGeometry.cSectors > 63)
2447	{
2448	/* Mark PCHS geometry as not yet valid (can't do the calculation here
2449	* as the total image size isn't known yet). */
2450	pImage->PCHSGeometry.cCylinders = 0;
2451	pImage->PCHSGeometry.cHeads = 16;
2452	pImage->PCHSGeometry.cSectors = 63;
2453	}
2454
2455	/* Determine LCHS geometry (set to 0 if not specified). */
2456	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2457	VMDK_DDB_GEO_LCHS_CYLINDERS,
2458	&pImage->LCHSGeometry.cCylinders);
2459	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2460	pImage->LCHSGeometry.cCylinders = 0;
2461	else if (RT_FAILURE(rc))
2462	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2463	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2464	VMDK_DDB_GEO_LCHS_HEADS,
2465	&pImage->LCHSGeometry.cHeads);
2466	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2467	pImage->LCHSGeometry.cHeads = 0;
2468	else if (RT_FAILURE(rc))
2469	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2470	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2471	VMDK_DDB_GEO_LCHS_SECTORS,
2472	&pImage->LCHSGeometry.cSectors);
2473	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2474	pImage->LCHSGeometry.cSectors = 0;
2475	else if (RT_FAILURE(rc))
2476	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2477	if ( pImage->LCHSGeometry.cCylinders == 0
2478	\|\| pImage->LCHSGeometry.cHeads == 0
2479	\|\| pImage->LCHSGeometry.cSectors == 0)
2480	{
2481	pImage->LCHSGeometry.cCylinders = 0;
2482	pImage->LCHSGeometry.cHeads = 0;
2483	pImage->LCHSGeometry.cSectors = 0;
2484	}
2485
2486	/* Get image UUID. */
2487	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor, VMDK_DDB_IMAGE_UUID,
2488	&pImage->ImageUuid);
2489	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2490	{
2491	/* Image without UUID. Probably created by VMware and not yet used
2492	* by VirtualBox. Can only be added for images opened in read/write
2493	* mode, so don't bother producing a sensible UUID otherwise. */
2494	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2495	RTUuidClear(&pImage->ImageUuid);
2496	else
2497	{
2498	rc = RTUuidCreate(&pImage->ImageUuid);
2499	if (RT_FAILURE(rc))
2500	return rc;
2501	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2502	VMDK_DDB_IMAGE_UUID, &pImage->ImageUuid);
2503	if (RT_FAILURE(rc))
2504	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image UUID in descriptor in '%s'"), pImage->pszFilename);
2505	}
2506	}
2507	else if (RT_FAILURE(rc))
2508	return rc;
2509
2510	/* Get image modification UUID. */
2511	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor,
2512	VMDK_DDB_MODIFICATION_UUID,
2513	&pImage->ModificationUuid);
2514	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2515	{
2516	/* Image without UUID. Probably created by VMware and not yet used
2517	* by VirtualBox. Can only be added for images opened in read/write
2518	* mode, so don't bother producing a sensible UUID otherwise. */
2519	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2520	RTUuidClear(&pImage->ModificationUuid);
2521	else
2522	{
2523	rc = RTUuidCreate(&pImage->ModificationUuid);
2524	if (RT_FAILURE(rc))
2525	return rc;
2526	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2527	VMDK_DDB_MODIFICATION_UUID,
2528	&pImage->ModificationUuid);
2529	if (RT_FAILURE(rc))
2530	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image modification UUID in descriptor in '%s'"), pImage->pszFilename);
2531	}
2532	}
2533	else if (RT_FAILURE(rc))
2534	return rc;
2535
2536	/* Get UUID of parent image. */
2537	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor, VMDK_DDB_PARENT_UUID,
2538	&pImage->ParentUuid);
2539	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2540	{
2541	/* Image without UUID. Probably created by VMware and not yet used
2542	* by VirtualBox. Can only be added for images opened in read/write
2543	* mode, so don't bother producing a sensible UUID otherwise. */
2544	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2545	RTUuidClear(&pImage->ParentUuid);
2546	else
2547	{
2548	rc = RTUuidClear(&pImage->ParentUuid);
2549	if (RT_FAILURE(rc))
2550	return rc;
2551	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2552	VMDK_DDB_PARENT_UUID, &pImage->ParentUuid);
2553	if (RT_FAILURE(rc))
2554	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent UUID in descriptor in '%s'"), pImage->pszFilename);
2555	}
2556	}
2557	else if (RT_FAILURE(rc))
2558	return rc;
2559
2560	/* Get parent image modification UUID. */
2561	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor,
2562	VMDK_DDB_PARENT_MODIFICATION_UUID,
2563	&pImage->ParentModificationUuid);
2564	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2565	{
2566	/* Image without UUID. Probably created by VMware and not yet used
2567	* by VirtualBox. Can only be added for images opened in read/write
2568	* mode, so don't bother producing a sensible UUID otherwise. */
2569	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2570	RTUuidClear(&pImage->ParentModificationUuid);
2571	else
2572	{
2573	RTUuidClear(&pImage->ParentModificationUuid);
2574	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2575	VMDK_DDB_PARENT_MODIFICATION_UUID,
2576	&pImage->ParentModificationUuid);
2577	if (RT_FAILURE(rc))
2578	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent modification UUID in descriptor in '%s'"), pImage->pszFilename);
2579	}
2580	}
2581	else if (RT_FAILURE(rc))
2582	return rc;
2583
2584	return VINF_SUCCESS;
2585	}
2586
2587	/**
2588	* Internal : Prepares the descriptor to write to the image.
2589	*/
2590	static int vmdkDescriptorPrepare(PVMDKIMAGE pImage, uint64_t cbLimit,
2591	void *ppvData, size_t pcbData)
2592	{
2593	int rc = VINF_SUCCESS;
2594
2595	/*
2596	* Allocate temporary descriptor buffer.
2597	* In case there is no limit allocate a default
2598	* and increase if required.
2599	*/
2600	size_t cbDescriptor = cbLimit ? cbLimit : 4 * _1K;
2601	char pszDescriptor = (char )RTMemAllocZ(cbDescriptor);
2602	size_t offDescriptor = 0;
2603
2604	if (!pszDescriptor)
2605	return VERR_NO_MEMORY;
2606
2607	for (unsigned i = 0; i < pImage->Descriptor.cLines; i++)
2608	{
2609	const char *psz = pImage->Descriptor.aLines[i];
2610	size_t cb = strlen(psz);
2611
2612	/*
2613	* Increase the descriptor if there is no limit and
2614	* there is not enough room left for this line.
2615	*/
2616	if (offDescriptor + cb + 1 > cbDescriptor)
2617	{
2618	if (cbLimit)
2619	{
2620	rc = vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too long in '%s'"), pImage->pszFilename);
2621	break;
2622	}
2623	else
2624	{
2625	char *pszDescriptorNew = NULL;
2626	LogFlow(("Increasing descriptor cache\n"));
2627
2628	pszDescriptorNew = (char )RTMemRealloc(pszDescriptor, cbDescriptor + cb + 4 _1K);
2629	if (!pszDescriptorNew)
2630	{
2631	rc = VERR_NO_MEMORY;
2632	break;
2633	}
2634	pszDescriptor = pszDescriptorNew;
2635	cbDescriptor += cb + 4 * _1K;
2636	}
2637	}
2638
2639	if (cb > 0)
2640	{
2641	memcpy(pszDescriptor + offDescriptor, psz, cb);
2642	offDescriptor += cb;
2643	}
2644
2645	memcpy(pszDescriptor + offDescriptor, "\n", 1);
2646	offDescriptor++;
2647	}
2648
2649	if (RT_SUCCESS(rc))
2650	{
2651	*ppvData = pszDescriptor;
2652	*pcbData = offDescriptor;
2653	}
2654	else if (pszDescriptor)
2655	RTMemFree(pszDescriptor);
2656
2657	return rc;
2658	}
2659
2660	/**
2661	* Internal: write/update the descriptor part of the image.
2662	*/
2663	static int vmdkWriteDescriptor(PVMDKIMAGE pImage, PVDIOCTX pIoCtx)
2664	{
2665	int rc = VINF_SUCCESS;
2666	uint64_t cbLimit;
2667	uint64_t uOffset;
2668	PVMDKFILE pDescFile;
2669	void *pvDescriptor = NULL;
2670	size_t cbDescriptor;
2671
2672	if (pImage->pDescData)
2673	{
2674	/* Separate descriptor file. */
2675	uOffset = 0;
2676	cbLimit = 0;
2677	pDescFile = pImage->pFile;
2678	}
2679	else
2680	{
2681	/* Embedded descriptor file. */
2682	uOffset = VMDK_SECTOR2BYTE(pImage->pExtents[0].uDescriptorSector);
2683	cbLimit = VMDK_SECTOR2BYTE(pImage->pExtents[0].cDescriptorSectors);
2684	pDescFile = pImage->pExtents[0].pFile;
2685	}
2686	/* Bail out if there is no file to write to. */
2687	if (pDescFile == NULL)
2688	return VERR_INVALID_PARAMETER;
2689
2690	rc = vmdkDescriptorPrepare(pImage, cbLimit, &pvDescriptor, &cbDescriptor);
2691	if (RT_SUCCESS(rc))
2692	{
2693	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pDescFile->pStorage,
2694	uOffset, pvDescriptor,
2695	cbLimit ? cbLimit : cbDescriptor,
2696	pIoCtx, NULL, NULL);
2697	if ( RT_FAILURE(rc)
2698	&& rc != VERR_VD_ASYNC_IO_IN_PROGRESS)
2699	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error writing descriptor in '%s'"), pImage->pszFilename);
2700	}
2701
2702	if (RT_SUCCESS(rc) && !cbLimit)
2703	{
2704	rc = vdIfIoIntFileSetSize(pImage->pIfIo, pDescFile->pStorage, cbDescriptor);
2705	if (RT_FAILURE(rc))
2706	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error truncating descriptor in '%s'"), pImage->pszFilename);
2707	}
2708
2709	if (RT_SUCCESS(rc))
2710	pImage->Descriptor.fDirty = false;
2711
2712	if (pvDescriptor)
2713	RTMemFree(pvDescriptor);
2714	return rc;
2715
2716	}
2717
2718	/**
2719	* Internal: validate the consistency check values in a binary header.
2720	*/
2721	static int vmdkValidateHeader(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, const SparseExtentHeader *pHeader)
2722	{
2723	int rc = VINF_SUCCESS;
2724	if (RT_LE2H_U32(pHeader->magicNumber) != VMDK_SPARSE_MAGICNUMBER)
2725	{
2726	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrect magic in sparse extent header in '%s'"), pExtent->pszFullname);
2727	return rc;
2728	}
2729	if (RT_LE2H_U32(pHeader->version) != 1 && RT_LE2H_U32(pHeader->version) != 3)
2730	{
2731	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);
2732	return rc;
2733	}
2734	if ( (RT_LE2H_U32(pHeader->flags) & 1)
2735	&& ( pHeader->singleEndLineChar != '\n'
2736	\|\| pHeader->nonEndLineChar != ' '
2737	\|\| pHeader->doubleEndLineChar1 != '\r'
2738	\|\| pHeader->doubleEndLineChar2 != '\n') )
2739	{
2740	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: corrupted by CR/LF translation in '%s'"), pExtent->pszFullname);
2741	return rc;
2742	}
2743	if (RT_LE2H_U64(pHeader->descriptorSize) > VMDK_SPARSE_DESCRIPTOR_SIZE_MAX)
2744	{
2745	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor size out of bounds (%llu vs %llu) '%s'"),
2746	pExtent->pszFullname, RT_LE2H_U64(pHeader->descriptorSize), VMDK_SPARSE_DESCRIPTOR_SIZE_MAX);
2747	return rc;
2748	}
2749	return rc;
2750	}
2751
2752	/**
2753	* Internal: read metadata belonging to an extent with binary header, i.e.
2754	* as found in monolithic files.
2755	*/
2756	static int vmdkReadBinaryMetaExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
2757	bool fMagicAlreadyRead)
2758	{
2759	SparseExtentHeader Header;
2760	int rc;
2761
2762	if (!fMagicAlreadyRead)
2763	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage, 0,
2764	&Header, sizeof(Header));
2765	else
2766	{
2767	Header.magicNumber = RT_H2LE_U32(VMDK_SPARSE_MAGICNUMBER);
2768	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
2769	RT_UOFFSETOF(SparseExtentHeader, version),
2770	&Header.version,
2771	sizeof(Header)
2772	- RT_UOFFSETOF(SparseExtentHeader, version));
2773	}
2774
2775	if (RT_SUCCESS(rc))
2776	{
2777	rc = vmdkValidateHeader(pImage, pExtent, &Header);
2778	if (RT_SUCCESS(rc))
2779	{
2780	uint64_t cbFile = 0;
2781
2782	if ( (RT_LE2H_U32(Header.flags) & RT_BIT(17))
2783	&& RT_LE2H_U64(Header.gdOffset) == VMDK_GD_AT_END)
2784	pExtent->fFooter = true;
2785
2786	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2787	\|\| ( pExtent->fFooter
2788	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2789	{
2790	rc = vdIfIoIntFileGetSize(pImage->pIfIo, pExtent->pFile->pStorage, &cbFile);
2791	if (RT_FAILURE(rc))
2792	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot get size of '%s'"), pExtent->pszFullname);
2793	}
2794
2795	if (RT_SUCCESS(rc))
2796	{
2797	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
2798	pExtent->uAppendPosition = RT_ALIGN_64(cbFile, 512);
2799
2800	if ( pExtent->fFooter
2801	&& ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2802	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2803	{
2804	/* Read the footer, which comes before the end-of-stream marker. */
2805	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
2806	cbFile - 2*512, &Header,
2807	sizeof(Header));
2808	if (RT_FAILURE(rc))
2809	{
2810	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading extent footer in '%s'"), pExtent->pszFullname);
2811	rc = VERR_VD_VMDK_INVALID_HEADER;
2812	}
2813
2814	if (RT_SUCCESS(rc))
2815	rc = vmdkValidateHeader(pImage, pExtent, &Header);
2816	/* Prohibit any writes to this extent. */
2817	pExtent->uAppendPosition = 0;
2818	}
2819
2820	if (RT_SUCCESS(rc))
2821	{
2822	pExtent->uVersion = RT_LE2H_U32(Header.version);
2823	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE; /* Just dummy value, changed later. */
2824	pExtent->cSectors = RT_LE2H_U64(Header.capacity);
2825	pExtent->cSectorsPerGrain = RT_LE2H_U64(Header.grainSize);
2826	pExtent->uDescriptorSector = RT_LE2H_U64(Header.descriptorOffset);
2827	pExtent->cDescriptorSectors = RT_LE2H_U64(Header.descriptorSize);
2828	pExtent->cGTEntries = RT_LE2H_U32(Header.numGTEsPerGT);
2829	pExtent->cOverheadSectors = RT_LE2H_U64(Header.overHead);
2830	pExtent->fUncleanShutdown = !!Header.uncleanShutdown;
2831	pExtent->uCompression = RT_LE2H_U16(Header.compressAlgorithm);
2832	if (RT_LE2H_U32(Header.flags) & RT_BIT(1))
2833	{
2834	pExtent->uSectorRGD = RT_LE2H_U64(Header.rgdOffset);
2835	pExtent->uSectorGD = RT_LE2H_U64(Header.gdOffset);
2836	}
2837	else
2838	{
2839	pExtent->uSectorGD = RT_LE2H_U64(Header.gdOffset);
2840	pExtent->uSectorRGD = 0;
2841	}
2842
2843	if (pExtent->uDescriptorSector && !pExtent->cDescriptorSectors)
2844	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2845	N_("VMDK: inconsistent embedded descriptor config in '%s'"), pExtent->pszFullname);
2846
2847	if ( RT_SUCCESS(rc)
2848	&& ( pExtent->uSectorGD == VMDK_GD_AT_END
2849	\|\| pExtent->uSectorRGD == VMDK_GD_AT_END)
2850	&& ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2851	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2852	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2853	N_("VMDK: cannot resolve grain directory offset in '%s'"), pExtent->pszFullname);
2854
2855	if (RT_SUCCESS(rc))
2856	{
2857	uint64_t cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
2858	if (!cSectorsPerGDE \|\| cSectorsPerGDE > UINT32_MAX)
2859	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2860	N_("VMDK: incorrect grain directory size in '%s'"), pExtent->pszFullname);
2861	else
2862	{
2863	pExtent->cSectorsPerGDE = cSectorsPerGDE;
2864	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
2865
2866	/* Fix up the number of descriptor sectors, as some flat images have
2867	* really just one, and this causes failures when inserting the UUID
2868	* values and other extra information. */
2869	if (pExtent->cDescriptorSectors != 0 && pExtent->cDescriptorSectors < 4)
2870	{
2871	/* Do it the easy way - just fix it for flat images which have no
2872	* other complicated metadata which needs space too. */
2873	if ( pExtent->uDescriptorSector + 4 < pExtent->cOverheadSectors
2874	&& pExtent->cGTEntries * pExtent->cGDEntries == 0)
2875	pExtent->cDescriptorSectors = 4;
2876	}
2877	}
2878	}
2879	}
2880	}
2881	}
2882	}
2883	else
2884	{
2885	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading extent header in '%s'"), pExtent->pszFullname);
2886	rc = VERR_VD_VMDK_INVALID_HEADER;
2887	}
2888
2889	if (RT_FAILURE(rc))
2890	vmdkFreeExtentData(pImage, pExtent, false);
2891
2892	return rc;
2893	}
2894
2895	/**
2896	* Internal: read additional metadata belonging to an extent. For those
2897	* extents which have no additional metadata just verify the information.
2898	*/
2899	static int vmdkReadMetaExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
2900	{
2901	int rc = VINF_SUCCESS;
2902
2903	/* disabled the check as there are too many truncated vmdk images out there */
2904	#ifdef VBOX_WITH_VMDK_STRICT_SIZE_CHECK
2905	uint64_t cbExtentSize;
2906	/* The image must be a multiple of a sector in size and contain the data
2907	* area (flat images only). If not, it means the image is at least
2908	* truncated, or even seriously garbled. */
2909	rc = vdIfIoIntFileGetSize(pImage->pIfIo, pExtent->pFile->pStorage, &cbExtentSize);
2910	if (RT_FAILURE(rc))
2911	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting size in '%s'"), pExtent->pszFullname);
2912	else if ( cbExtentSize != RT_ALIGN_64(cbExtentSize, 512)
2913	&& (pExtent->enmType != VMDKETYPE_FLAT \|\| pExtent->cNominalSectors + pExtent->uSectorOffset > VMDK_BYTE2SECTOR(cbExtentSize)))
2914	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2915	N_("VMDK: file size is not a multiple of 512 in '%s', file is truncated or otherwise garbled"), pExtent->pszFullname);
2916	#endif /* VBOX_WITH_VMDK_STRICT_SIZE_CHECK */
2917	if ( RT_SUCCESS(rc)
2918	&& pExtent->enmType == VMDKETYPE_HOSTED_SPARSE)
2919	{
2920	/* The spec says that this must be a power of two and greater than 8,
2921	* but probably they meant not less than 8. */
2922	if ( (pExtent->cSectorsPerGrain & (pExtent->cSectorsPerGrain - 1))
2923	\|\| pExtent->cSectorsPerGrain < 8)
2924	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2925	N_("VMDK: invalid extent grain size %u in '%s'"), pExtent->cSectorsPerGrain, pExtent->pszFullname);
2926	else
2927	{
2928	/* This code requires that a grain table must hold a power of two multiple
2929	* of the number of entries per GT cache entry. */
2930	if ( (pExtent->cGTEntries & (pExtent->cGTEntries - 1))
2931	\|\| pExtent->cGTEntries < VMDK_GT_CACHELINE_SIZE)
2932	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2933	N_("VMDK: grain table cache size problem in '%s'"), pExtent->pszFullname);
2934	else
2935	{
2936	rc = vmdkAllocStreamBuffers(pImage, pExtent);
2937	if (RT_SUCCESS(rc))
2938	{
2939	/* Prohibit any writes to this streamOptimized extent. */
2940	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2941	pExtent->uAppendPosition = 0;
2942
2943	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2944	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2945	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
2946	rc = vmdkReadGrainDirectory(pImage, pExtent);
2947	else
2948	{
2949	pExtent->uGrainSectorAbs = pExtent->cOverheadSectors;
2950	pExtent->cbGrainStreamRead = 0;
2951	}
2952	}
2953	}
2954	}
2955	}
2956
2957	if (RT_FAILURE(rc))
2958	vmdkFreeExtentData(pImage, pExtent, false);
2959
2960	return rc;
2961	}
2962
2963	/**
2964	* Internal: write/update the metadata for a sparse extent.
2965	*/
2966	static int vmdkWriteMetaSparseExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
2967	uint64_t uOffset, PVDIOCTX pIoCtx)
2968	{
2969	SparseExtentHeader Header;
2970
2971	memset(&Header, '\0', sizeof(Header));
2972	Header.magicNumber = RT_H2LE_U32(VMDK_SPARSE_MAGICNUMBER);
2973	Header.version = RT_H2LE_U32(pExtent->uVersion);
2974	Header.flags = RT_H2LE_U32(RT_BIT(0));
2975	if (pExtent->pRGD)
2976	Header.flags \|= RT_H2LE_U32(RT_BIT(1));
2977	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2978	Header.flags \|= RT_H2LE_U32(RT_BIT(16) \| RT_BIT(17));
2979	Header.capacity = RT_H2LE_U64(pExtent->cSectors);
2980	Header.grainSize = RT_H2LE_U64(pExtent->cSectorsPerGrain);
2981	Header.descriptorOffset = RT_H2LE_U64(pExtent->uDescriptorSector);
2982	Header.descriptorSize = RT_H2LE_U64(pExtent->cDescriptorSectors);
2983	Header.numGTEsPerGT = RT_H2LE_U32(pExtent->cGTEntries);
2984	if (pExtent->fFooter && uOffset == 0)
2985	{
2986	if (pExtent->pRGD)
2987	{
2988	Assert(pExtent->uSectorRGD);
2989	Header.rgdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2990	Header.gdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2991	}
2992	else
2993	Header.gdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2994	}
2995	else
2996	{
2997	if (pExtent->pRGD)
2998	{
2999	Assert(pExtent->uSectorRGD);
3000	Header.rgdOffset = RT_H2LE_U64(pExtent->uSectorRGD);
3001	Header.gdOffset = RT_H2LE_U64(pExtent->uSectorGD);
3002	}
3003	else
3004	Header.gdOffset = RT_H2LE_U64(pExtent->uSectorGD);
3005	}
3006	Header.overHead = RT_H2LE_U64(pExtent->cOverheadSectors);
3007	Header.uncleanShutdown = pExtent->fUncleanShutdown;
3008	Header.singleEndLineChar = '\n';
3009	Header.nonEndLineChar = ' ';
3010	Header.doubleEndLineChar1 = '\r';
3011	Header.doubleEndLineChar2 = '\n';
3012	Header.compressAlgorithm = RT_H2LE_U16(pExtent->uCompression);
3013
3014	int rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
3015	uOffset, &Header, sizeof(Header),
3016	pIoCtx, NULL, NULL);
3017	if (RT_FAILURE(rc) && (rc != VERR_VD_ASYNC_IO_IN_PROGRESS))
3018	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error writing extent header in '%s'"), pExtent->pszFullname);
3019	return rc;
3020	}
3021
3022	/**
3023	* Internal: free the buffers used for streamOptimized images.
3024	*/
3025	static void vmdkFreeStreamBuffers(PVMDKEXTENT pExtent)
3026	{
3027	if (pExtent->pvCompGrain)
3028	{
3029	RTMemFree(pExtent->pvCompGrain);
3030	pExtent->pvCompGrain = NULL;
3031	}
3032	if (pExtent->pvGrain)
3033	{
3034	RTMemFree(pExtent->pvGrain);
3035	pExtent->pvGrain = NULL;
3036	}
3037	}
3038
3039	/**
3040	* Internal: free the memory used by the extent data structure, optionally
3041	* deleting the referenced files.
3042	*
3043	* @returns VBox status code.
3044	* @param pImage Pointer to the image instance data.
3045	* @param pExtent The extent to free.
3046	* @param fDelete Flag whether to delete the backing storage.
3047	*/
3048	static int vmdkFreeExtentData(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
3049	bool fDelete)
3050	{
3051	int rc = VINF_SUCCESS;
3052
3053	vmdkFreeGrainDirectory(pExtent);
3054	if (pExtent->pDescData)
3055	{
3056	RTMemFree(pExtent->pDescData);
3057	pExtent->pDescData = NULL;
3058	}
3059	if (pExtent->pFile != NULL)
3060	{
3061	/* Do not delete raw extents, these have full and base names equal. */
3062	rc = vmdkFileClose(pImage, &pExtent->pFile,
3063	fDelete
3064	&& pExtent->pszFullname
3065	&& pExtent->pszBasename
3066	&& strcmp(pExtent->pszFullname, pExtent->pszBasename));
3067	}
3068	if (pExtent->pszBasename)
3069	{
3070	RTMemTmpFree((void *)pExtent->pszBasename);
3071	pExtent->pszBasename = NULL;
3072	}
3073	if (pExtent->pszFullname)
3074	{
3075	RTStrFree((char )(void )pExtent->pszFullname);
3076	pExtent->pszFullname = NULL;
3077	}
3078	vmdkFreeStreamBuffers(pExtent);
3079
3080	return rc;
3081	}
3082
3083	/**
3084	* Internal: allocate grain table cache if necessary for this image.
3085	*/
3086	static int vmdkAllocateGrainTableCache(PVMDKIMAGE pImage)
3087	{
3088	PVMDKEXTENT pExtent;
3089
3090	/* Allocate grain table cache if any sparse extent is present. */
3091	for (unsigned i = 0; i < pImage->cExtents; i++)
3092	{
3093	pExtent = &pImage->pExtents[i];
3094	if (pExtent->enmType == VMDKETYPE_HOSTED_SPARSE)
3095	{
3096	/* Allocate grain table cache. */
3097	pImage->pGTCache = (PVMDKGTCACHE)RTMemAllocZ(sizeof(VMDKGTCACHE));
3098	if (!pImage->pGTCache)
3099	return VERR_NO_MEMORY;
3100	for (unsigned j = 0; j < VMDK_GT_CACHE_SIZE; j++)
3101	{
3102	PVMDKGTCACHEENTRY pGCE = &pImage->pGTCache->aGTCache[j];
3103	pGCE->uExtent = UINT32_MAX;
3104	}
3105	pImage->pGTCache->cEntries = VMDK_GT_CACHE_SIZE;
3106	break;
3107	}
3108	}
3109
3110	return VINF_SUCCESS;
3111	}
3112
3113	/**
3114	* Internal: allocate the given number of extents.
3115	*/
3116	static int vmdkCreateExtents(PVMDKIMAGE pImage, unsigned cExtents)
3117	{
3118	int rc = VINF_SUCCESS;
3119	PVMDKEXTENT pExtents = (PVMDKEXTENT)RTMemAllocZ(cExtents * sizeof(VMDKEXTENT));
3120	if (pExtents)
3121	{
3122	for (unsigned i = 0; i < cExtents; i++)
3123	{
3124	pExtents[i].pFile = NULL;
3125	pExtents[i].pszBasename = NULL;
3126	pExtents[i].pszFullname = NULL;
3127	pExtents[i].pGD = NULL;
3128	pExtents[i].pRGD = NULL;
3129	pExtents[i].pDescData = NULL;
3130	pExtents[i].uVersion = 1;
3131	pExtents[i].uCompression = VMDK_COMPRESSION_NONE;
3132	pExtents[i].uExtent = i;
3133	pExtents[i].pImage = pImage;
3134	}
3135	pImage->pExtents = pExtents;
3136	pImage->cExtents = cExtents;
3137	}
3138	else
3139	rc = VERR_NO_MEMORY;
3140
3141	return rc;
3142	}
3143
3144	/**
3145	* Internal: Create an additional file backed extent in split images.
3146	* Supports split sparse and flat images.
3147	*
3148	* @returns VBox status code.
3149	* @param pImage VMDK image instance.
3150	* @param cbSize Desiried size in bytes of new extent.
3151	*/
3152	static int vmdkAddFileBackedExtent(PVMDKIMAGE pImage, uint64_t cbSize)
3153	{
3154	int rc = VINF_SUCCESS;
3155	unsigned uImageFlags = pImage->uImageFlags;
3156
3157	/* Check for unsupported image type. */
3158	if ((uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
3159	\|\| (uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
3160	\|\| (uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK))
3161	{
3162	return VERR_NOT_SUPPORTED;
3163	}
3164
3165	/* Allocate array of extents and copy existing extents to it. */
3166	PVMDKEXTENT pNewExtents = (PVMDKEXTENT)RTMemAllocZ((pImage->cExtents + 1) * sizeof(VMDKEXTENT));
3167	if (!pNewExtents)
3168	{
3169	return VERR_NO_MEMORY;
3170	}
3171
3172	memcpy(pNewExtents, pImage->pExtents, pImage->cExtents * sizeof(VMDKEXTENT));
3173
3174	/* Locate newly created extent and populate default metadata. */
3175	PVMDKEXTENT pExtent = &pNewExtents[pImage->cExtents];
3176
3177	pExtent->pFile = NULL;
3178	pExtent->pszBasename = NULL;
3179	pExtent->pszFullname = NULL;
3180	pExtent->pGD = NULL;
3181	pExtent->pRGD = NULL;
3182	pExtent->pDescData = NULL;
3183	pExtent->uVersion = 1;
3184	pExtent->uCompression = VMDK_COMPRESSION_NONE;
3185	pExtent->uExtent = pImage->cExtents;
3186	pExtent->pImage = pImage;
3187	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
3188	pExtent->enmAccess = VMDKACCESS_READWRITE;
3189	pExtent->uSectorOffset = 0;
3190	pExtent->fMetaDirty = true;
3191
3192	/* Apply image type specific meta data. */
3193	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
3194	{
3195	pExtent->enmType = VMDKETYPE_FLAT;
3196	}
3197	else
3198	{
3199	uint64_t cSectorsPerGDE, cSectorsPerGD;
3200	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE;
3201	pExtent->cSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64(cbSize, _64K));
3202	pExtent->cSectorsPerGrain = VMDK_BYTE2SECTOR(_64K);
3203	pExtent->cGTEntries = 512;
3204	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
3205	pExtent->cSectorsPerGDE = cSectorsPerGDE;
3206	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
3207	cSectorsPerGD = (pExtent->cGDEntries + (512 / sizeof(uint32_t) - 1)) / (512 / sizeof(uint32_t));
3208	}
3209
3210	/* Allocate and set file name for extent. */
3211	char *pszBasenameSubstr = RTPathFilename(pImage->pszFilename);
3212	AssertPtr(pszBasenameSubstr);
3213
3214	char *pszBasenameSuff = RTPathSuffix(pszBasenameSubstr);
3215	char *pszBasenameBase = RTStrDup(pszBasenameSubstr);
3216	RTPathStripSuffix(pszBasenameBase);
3217	char *pszTmp;
3218	size_t cbTmp;
3219
3220	if (pImage->uImageFlags & VD_IMAGE_FLAGS_FIXED)
3221	RTStrAPrintf(&pszTmp, "%s-f%03d%s", pszBasenameBase,
3222	pExtent->uExtent + 1, pszBasenameSuff);
3223	else
3224	RTStrAPrintf(&pszTmp, "%s-s%03d%s", pszBasenameBase, pExtent->uExtent + 1,
3225	pszBasenameSuff);
3226
3227	RTStrFree(pszBasenameBase);
3228	if (!pszTmp)
3229	return VERR_NO_STR_MEMORY;
3230	cbTmp = strlen(pszTmp) + 1;
3231	char pszBasename = (char )RTMemTmpAlloc(cbTmp);
3232	if (!pszBasename)
3233	{
3234	RTStrFree(pszTmp);
3235	return VERR_NO_MEMORY;
3236	}
3237
3238	memcpy(pszBasename, pszTmp, cbTmp);
3239	RTStrFree(pszTmp);
3240
3241	pExtent->pszBasename = pszBasename;
3242
3243	char *pszBasedirectory = RTStrDup(pImage->pszFilename);
3244	if (!pszBasedirectory)
3245	return VERR_NO_STR_MEMORY;
3246	RTPathStripFilename(pszBasedirectory);
3247	char *pszFullname = RTPathJoinA(pszBasedirectory, pExtent->pszBasename);
3248	RTStrFree(pszBasedirectory);
3249	if (!pszFullname)
3250	return VERR_NO_STR_MEMORY;
3251	pExtent->pszFullname = pszFullname;
3252
3253	/* Create file for extent. */
3254	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
3255	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
3256	true /* fCreate */));
3257	if (RT_FAILURE(rc))
3258	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pExtent->pszFullname);
3259
3260	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
3261	{
3262	/* For flat images: Pre allocate file space. */
3263	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage, cbSize,
3264	0 /* fFlags */, NULL, 0, 0);
3265	if (RT_FAILURE(rc))
3266	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
3267	}
3268	else
3269	{
3270	/* For sparse images: Allocate new grain directories/tables. */
3271	/* fPreAlloc should never be false because VMware can't use such images. */
3272	rc = vmdkCreateGrainDirectory(pImage, pExtent,
3273	RT_MAX( pExtent->uDescriptorSector
3274	+ pExtent->cDescriptorSectors,
3275	1),
3276	true /* fPreAlloc */);
3277	if (RT_FAILURE(rc))
3278	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new grain directory in '%s'"), pExtent->pszFullname);
3279	}
3280
3281	/* Insert new extent into descriptor file. */
3282	rc = vmdkDescExtInsert(pImage, &pImage->Descriptor, pExtent->enmAccess,
3283	pExtent->cNominalSectors, pExtent->enmType,
3284	pExtent->pszBasename, pExtent->uSectorOffset);
3285	if (RT_FAILURE(rc))
3286	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not insert the extent list into descriptor in '%s'"), pImage->pszFilename);
3287
3288	pImage->pExtents = pNewExtents;
3289	pImage->cExtents++;
3290
3291	return rc;
3292	}
3293
3294	/**
3295	* Reads and processes the descriptor embedded in sparse images.
3296	*
3297	* @returns VBox status code.
3298	* @param pImage VMDK image instance.
3299	* @param pFile The sparse file handle.
3300	*/
3301	static int vmdkDescriptorReadSparse(PVMDKIMAGE pImage, PVMDKFILE pFile)
3302	{
3303	/* It's a hosted single-extent image. */
3304	int rc = vmdkCreateExtents(pImage, 1);
3305	if (RT_SUCCESS(rc))
3306	{
3307	/* The opened file is passed to the extent. No separate descriptor
3308	* file, so no need to keep anything open for the image. */
3309	PVMDKEXTENT pExtent = &pImage->pExtents[0];
3310	pExtent->pFile = pFile;
3311	pImage->pFile = NULL;
3312	pExtent->pszFullname = RTPathAbsDup(pImage->pszFilename);
3313	if (RT_LIKELY(pExtent->pszFullname))
3314	{
3315	/* As we're dealing with a monolithic image here, there must
3316	* be a descriptor embedded in the image file. */
3317	rc = vmdkReadBinaryMetaExtent(pImage, pExtent, true /* fMagicAlreadyRead */);
3318	if ( RT_SUCCESS(rc)
3319	&& pExtent->uDescriptorSector
3320	&& pExtent->cDescriptorSectors)
3321	{
3322	/* HACK: extend the descriptor if it is unusually small and it fits in
3323	* the unused space after the image header. Allows opening VMDK files
3324	* with extremely small descriptor in read/write mode.
3325	*
3326	* The previous version introduced a possible regression for VMDK stream
3327	* optimized images from VMware which tend to have only a single sector sized
3328	* descriptor. Increasing the descriptor size resulted in adding the various uuid
3329	* entries required to make it work with VBox but for stream optimized images
3330	* the updated binary header wasn't written to the disk creating a mismatch
3331	* between advertised and real descriptor size.
3332	*
3333	* The descriptor size will be increased even if opened readonly now if there
3334	* enough room but the new value will not be written back to the image.
3335	*/
3336	if ( pExtent->cDescriptorSectors < 3
3337	&& (int64_t)pExtent->uSectorGD - pExtent->uDescriptorSector >= 4
3338	&& (!pExtent->uSectorRGD \|\| (int64_t)pExtent->uSectorRGD - pExtent->uDescriptorSector >= 4))
3339	{
3340	uint64_t cDescriptorSectorsOld = pExtent->cDescriptorSectors;
3341
3342	pExtent->cDescriptorSectors = 4;
3343	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
3344	{
3345	/*
3346	* Update the on disk number now to make sure we don't introduce inconsistencies
3347	* in case of stream optimized images from VMware where the descriptor is just
3348	* one sector big (the binary header is not written to disk for complete
3349	* stream optimized images in vmdkFlushImage()).
3350	*/
3351	uint64_t u64DescSizeNew = RT_H2LE_U64(pExtent->cDescriptorSectors);
3352	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pFile->pStorage,
3353	RT_UOFFSETOF(SparseExtentHeader, descriptorSize),
3354	&u64DescSizeNew, sizeof(u64DescSizeNew));
3355	if (RT_FAILURE(rc))
3356	{
3357	LogFlowFunc(("Increasing the descriptor size failed with %Rrc\n", rc));
3358	/* Restore the old size and carry on. */
3359	pExtent->cDescriptorSectors = cDescriptorSectorsOld;
3360	}
3361	}
3362	}
3363	/* Read the descriptor from the extent. */
3364	pExtent->pDescData = (char *)RTMemAllocZ(VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
3365	if (RT_LIKELY(pExtent->pDescData))
3366	{
3367	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
3368	VMDK_SECTOR2BYTE(pExtent->uDescriptorSector),
3369	pExtent->pDescData,
3370	VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
3371	if (RT_SUCCESS(rc))
3372	{
3373	rc = vmdkParseDescriptor(pImage, pExtent->pDescData,
3374	VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
3375	if ( RT_SUCCESS(rc)
3376	&& ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
3377	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_ASYNC_IO)))
3378	{
3379	rc = vmdkReadMetaExtent(pImage, pExtent);
3380	if (RT_SUCCESS(rc))
3381	{
3382	/* Mark the extent as unclean if opened in read-write mode. */
3383	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3384	&& !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
3385	{
3386	pExtent->fUncleanShutdown = true;
3387	pExtent->fMetaDirty = true;
3388	}
3389	}
3390	}
3391	else if (RT_SUCCESS(rc))
3392	rc = VERR_NOT_SUPPORTED;
3393	}
3394	else
3395	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: read error for descriptor in '%s'"), pExtent->pszFullname);
3396	}
3397	else
3398	rc = VERR_NO_MEMORY;
3399	}
3400	else if (RT_SUCCESS(rc))
3401	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: monolithic image without descriptor in '%s'"), pImage->pszFilename);
3402	}
3403	else
3404	rc = VERR_NO_MEMORY;
3405	}
3406
3407	return rc;
3408	}
3409
3410	/**
3411	* Reads the descriptor from a pure text file.
3412	*
3413	* @returns VBox status code.
3414	* @param pImage VMDK image instance.
3415	* @param pFile The descriptor file handle.
3416	*/
3417	static int vmdkDescriptorReadAscii(PVMDKIMAGE pImage, PVMDKFILE pFile)
3418	{
3419	/* Allocate at least 10K, and make sure that there is 5K free space
3420	* in case new entries need to be added to the descriptor. Never
3421	* allocate more than 128K, because that's no valid descriptor file
3422	* and will result in the correct "truncated read" error handling. */
3423	uint64_t cbFileSize;
3424	int rc = vdIfIoIntFileGetSize(pImage->pIfIo, pFile->pStorage, &cbFileSize);
3425	if ( RT_SUCCESS(rc)
3426	&& cbFileSize >= 50)
3427	{
3428	uint64_t cbSize = cbFileSize;
3429	if (cbSize % VMDK_SECTOR2BYTE(10))
3430	cbSize += VMDK_SECTOR2BYTE(20) - cbSize % VMDK_SECTOR2BYTE(10);
3431	else
3432	cbSize += VMDK_SECTOR2BYTE(10);
3433	cbSize = RT_MIN(cbSize, _128K);
3434	pImage->cbDescAlloc = RT_MAX(VMDK_SECTOR2BYTE(20), cbSize);
3435	pImage->pDescData = (char *)RTMemAllocZ(pImage->cbDescAlloc);
3436	if (RT_LIKELY(pImage->pDescData))
3437	{
3438	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pFile->pStorage, 0, pImage->pDescData,
3439	RT_MIN(pImage->cbDescAlloc, cbFileSize));
3440	if (RT_SUCCESS(rc))
3441	{
3442	#if 0 /** @todo Revisit */
3443	cbRead += sizeof(u32Magic);
3444	if (cbRead == pImage->cbDescAlloc)
3445	{
3446	/* Likely the read is truncated. Better fail a bit too early
3447	* (normally the descriptor is much smaller than our buffer). */
3448	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: cannot read descriptor in '%s'"), pImage->pszFilename);
3449	goto out;
3450	}
3451	#endif
3452	rc = vmdkParseDescriptor(pImage, pImage->pDescData,
3453	pImage->cbDescAlloc);
3454	if (RT_SUCCESS(rc))
3455	{
3456	for (unsigned i = 0; i < pImage->cExtents && RT_SUCCESS(rc); i++)
3457	{
3458	PVMDKEXTENT pExtent = &pImage->pExtents[i];
3459	if (pExtent->pszBasename)
3460	{
3461	/* Hack to figure out whether the specified name in the
3462	* extent descriptor is absolute. Doesn't always work, but
3463	* should be good enough for now. */
3464	char *pszFullname;
3465	/** @todo implement proper path absolute check. */
3466	if (pExtent->pszBasename[0] == RTPATH_SLASH)
3467	{
3468	pszFullname = RTStrDup(pExtent->pszBasename);
3469	if (!pszFullname)
3470	{
3471	rc = VERR_NO_MEMORY;
3472	break;
3473	}
3474	}
3475	else
3476	{
3477	char *pszDirname = RTStrDup(pImage->pszFilename);
3478	if (!pszDirname)
3479	{
3480	rc = VERR_NO_MEMORY;
3481	break;
3482	}
3483	RTPathStripFilename(pszDirname);
3484	pszFullname = RTPathJoinA(pszDirname, pExtent->pszBasename);
3485	RTStrFree(pszDirname);
3486	if (!pszFullname)
3487	{
3488	rc = VERR_NO_STR_MEMORY;
3489	break;
3490	}
3491	}
3492	pExtent->pszFullname = pszFullname;
3493	}
3494	else
3495	pExtent->pszFullname = NULL;
3496
3497	unsigned uOpenFlags = pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0);
3498	switch (pExtent->enmType)
3499	{
3500	case VMDKETYPE_HOSTED_SPARSE:
3501	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
3502	VDOpenFlagsToFileOpenFlags(uOpenFlags, false /* fCreate */));
3503	if (RT_FAILURE(rc))
3504	{
3505	/* Do NOT signal an appropriate error here, as the VD
3506	* layer has the choice of retrying the open if it
3507	* failed. */
3508	break;
3509	}
3510	rc = vmdkReadBinaryMetaExtent(pImage, pExtent,
3511	false /* fMagicAlreadyRead */);
3512	if (RT_FAILURE(rc))
3513	break;
3514	rc = vmdkReadMetaExtent(pImage, pExtent);
3515	if (RT_FAILURE(rc))
3516	break;
3517
3518	/* Mark extent as unclean if opened in read-write mode. */
3519	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
3520	{
3521	pExtent->fUncleanShutdown = true;
3522	pExtent->fMetaDirty = true;
3523	}
3524	break;
3525	case VMDKETYPE_VMFS:
3526	case VMDKETYPE_FLAT:
3527	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
3528	VDOpenFlagsToFileOpenFlags(uOpenFlags, false /* fCreate */));
3529	if (RT_FAILURE(rc))
3530	{
3531	/* Do NOT signal an appropriate error here, as the VD
3532	* layer has the choice of retrying the open if it
3533	* failed. */
3534	break;
3535	}
3536	break;
3537	case VMDKETYPE_ZERO:
3538	/* Nothing to do. */
3539	break;
3540	default:
3541	AssertMsgFailed(("unknown vmdk extent type %d\n", pExtent->enmType));
3542	}
3543	}
3544	}
3545	}
3546	else
3547	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: read error for descriptor in '%s'"), pImage->pszFilename);
3548	}
3549	else
3550	rc = VERR_NO_MEMORY;
3551	}
3552	else if (RT_SUCCESS(rc))
3553	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor in '%s' is too short"), pImage->pszFilename);
3554
3555	return rc;
3556	}
3557
3558	/**
3559	* Read and process the descriptor based on the image type.
3560	*
3561	* @returns VBox status code.
3562	* @param pImage VMDK image instance.
3563	* @param pFile VMDK file handle.
3564	*/
3565	static int vmdkDescriptorRead(PVMDKIMAGE pImage, PVMDKFILE pFile)
3566	{
3567	uint32_t u32Magic;
3568
3569	/* Read magic (if present). */
3570	int rc = vdIfIoIntFileReadSync(pImage->pIfIo, pFile->pStorage, 0,
3571	&u32Magic, sizeof(u32Magic));
3572	if (RT_SUCCESS(rc))
3573	{
3574	/* Handle the file according to its magic number. */
3575	if (RT_LE2H_U32(u32Magic) == VMDK_SPARSE_MAGICNUMBER)
3576	rc = vmdkDescriptorReadSparse(pImage, pFile);
3577	else
3578	rc = vmdkDescriptorReadAscii(pImage, pFile);
3579	}
3580	else
3581	{
3582	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading the magic number in '%s'"), pImage->pszFilename);
3583	rc = VERR_VD_VMDK_INVALID_HEADER;
3584	}
3585
3586	return rc;
3587	}
3588
3589	/**
3590	* Internal: Open an image, constructing all necessary data structures.
3591	*/
3592	static int vmdkOpenImage(PVMDKIMAGE pImage, unsigned uOpenFlags)
3593	{
3594	pImage->uOpenFlags = uOpenFlags;
3595	pImage->pIfError = VDIfErrorGet(pImage->pVDIfsDisk);
3596	pImage->pIfIo = VDIfIoIntGet(pImage->pVDIfsImage);
3597	AssertPtrReturn(pImage->pIfIo, VERR_INVALID_PARAMETER);
3598
3599	/*
3600	* Open the image.
3601	* We don't have to check for asynchronous access because
3602	* we only support raw access and the opened file is a description
3603	* file were no data is stored.
3604	*/
3605	PVMDKFILE pFile;
3606	int rc = vmdkFileOpen(pImage, &pFile, NULL, pImage->pszFilename,
3607	VDOpenFlagsToFileOpenFlags(uOpenFlags, false /* fCreate */));
3608	if (RT_SUCCESS(rc))
3609	{
3610	pImage->pFile = pFile;
3611
3612	rc = vmdkDescriptorRead(pImage, pFile);
3613	if (RT_SUCCESS(rc))
3614	{
3615	/* Determine PCHS geometry if not set. */
3616	if (pImage->PCHSGeometry.cCylinders == 0)
3617	{
3618	uint64_t cCylinders = VMDK_BYTE2SECTOR(pImage->cbSize)
3619	/ pImage->PCHSGeometry.cHeads
3620	/ pImage->PCHSGeometry.cSectors;
3621	pImage->PCHSGeometry.cCylinders = (unsigned)RT_MIN(cCylinders, 16383);
3622	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3623	&& !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
3624	{
3625	rc = vmdkDescSetPCHSGeometry(pImage, &pImage->PCHSGeometry);
3626	AssertRC(rc);
3627	}
3628	}
3629
3630	/* Update the image metadata now in case has changed. */
3631	rc = vmdkFlushImage(pImage, NULL);
3632	if (RT_SUCCESS(rc))
3633	{
3634	/* Figure out a few per-image constants from the extents. */
3635	pImage->cbSize = 0;
3636	for (unsigned i = 0; i < pImage->cExtents; i++)
3637	{
3638	PVMDKEXTENT pExtent = &pImage->pExtents[i];
3639	if (pExtent->enmType == VMDKETYPE_HOSTED_SPARSE)
3640	{
3641	/* Here used to be a check whether the nominal size of an extent
3642	* is a multiple of the grain size. The spec says that this is
3643	* always the case, but unfortunately some files out there in the
3644	* wild violate the spec (e.g. ReactOS 0.3.1). */
3645	}
3646	else if ( pExtent->enmType == VMDKETYPE_FLAT
3647	\|\| pExtent->enmType == VMDKETYPE_ZERO)
3648	pImage->uImageFlags \|= VD_IMAGE_FLAGS_FIXED;
3649
3650	pImage->cbSize += VMDK_SECTOR2BYTE(pExtent->cNominalSectors);
3651	}
3652
3653	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
3654	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3655	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
3656	rc = vmdkAllocateGrainTableCache(pImage);
3657	}
3658	}
3659	}
3660	/* else: Do NOT signal an appropriate error here, as the VD layer has the
3661	* choice of retrying the open if it failed. */
3662
3663	if (RT_SUCCESS(rc))
3664	{
3665	PVDREGIONDESC pRegion = &pImage->RegionList.aRegions[0];
3666	pImage->RegionList.fFlags = 0;
3667	pImage->RegionList.cRegions = 1;
3668
3669	pRegion->offRegion = 0; /* Disk start. */
3670	pRegion->cbBlock = 512;
3671	pRegion->enmDataForm = VDREGIONDATAFORM_RAW;
3672	pRegion->enmMetadataForm = VDREGIONMETADATAFORM_NONE;
3673	pRegion->cbData = 512;
3674	pRegion->cbMetadata = 0;
3675	pRegion->cRegionBlocksOrBytes = pImage->cbSize;
3676	}
3677	else
3678	vmdkFreeImage(pImage, false, false /fFlush/); /* Don't try to flush anything if opening failed. */
3679	return rc;
3680	}
3681
3682	/**
3683	* Frees a raw descriptor.
3684	* @internal
3685	*/
3686	static int vmdkRawDescFree(PVDISKRAW pRawDesc)
3687	{
3688	if (!pRawDesc)
3689	return VINF_SUCCESS;
3690
3691	RTStrFree(pRawDesc->pszRawDisk);
3692	pRawDesc->pszRawDisk = NULL;
3693
3694	/* Partitions: */
3695	for (unsigned i = 0; i < pRawDesc->cPartDescs; i++)
3696	{
3697	RTStrFree(pRawDesc->pPartDescs[i].pszRawDevice);
3698	pRawDesc->pPartDescs[i].pszRawDevice = NULL;
3699
3700	RTMemFree(pRawDesc->pPartDescs[i].pvPartitionData);
3701	pRawDesc->pPartDescs[i].pvPartitionData = NULL;
3702	}
3703
3704	RTMemFree(pRawDesc->pPartDescs);
3705	pRawDesc->pPartDescs = NULL;
3706
3707	RTMemFree(pRawDesc);
3708	return VINF_SUCCESS;
3709	}
3710
3711	/**
3712	* Helper that grows the raw partition descriptor table by @a cToAdd entries,
3713	* returning the pointer to the first new entry.
3714	* @internal
3715	*/
3716	static int vmdkRawDescAppendPartDesc(PVMDKIMAGE pImage, PVDISKRAW pRawDesc, uint32_t cToAdd, PVDISKRAWPARTDESC *ppRet)
3717	{
3718	uint32_t const cOld = pRawDesc->cPartDescs;
3719	uint32_t const cNew = cOld + cToAdd;
3720	PVDISKRAWPARTDESC paNew = (PVDISKRAWPARTDESC)RTMemReallocZ(pRawDesc->pPartDescs,
3721	cOld * sizeof(pRawDesc->pPartDescs[0]),
3722	cNew * sizeof(pRawDesc->pPartDescs[0]));
3723	if (paNew)
3724	{
3725	pRawDesc->cPartDescs = cNew;
3726	pRawDesc->pPartDescs = paNew;
3727
3728	*ppRet = &paNew[cOld];
3729	return VINF_SUCCESS;
3730	}
3731	*ppRet = NULL;
3732	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS,
3733	N_("VMDK: Image path: '%s'. Out of memory growing the partition descriptors (%u -> %u)."),
3734	pImage->pszFilename, cOld, cNew);
3735	}
3736
3737	/**
3738	* @callback_method_impl{FNRTSORTCMP}
3739	*/
3740	static DECLCALLBACK(int) vmdkRawDescPartComp(void const pvElement1, void const pvElement2, void *pvUser)
3741	{
3742	RT_NOREF(pvUser);
3743	int64_t const iDelta = ((PVDISKRAWPARTDESC)pvElement1)->offStartInVDisk - ((PVDISKRAWPARTDESC)pvElement2)->offStartInVDisk;
3744	return iDelta < 0 ? -1 : iDelta > 0 ? 1 : 0;
3745	}
3746
3747	/**
3748	* Post processes the partition descriptors.
3749	*
3750	* Sorts them and check that they don't overlap.
3751	*/
3752	static int vmdkRawDescPostProcessPartitions(PVMDKIMAGE pImage, PVDISKRAW pRawDesc, uint64_t cbSize)
3753	{
3754	/*
3755	* Sort data areas in ascending order of start.
3756	*/
3757	RTSortShell(pRawDesc->pPartDescs, pRawDesc->cPartDescs, sizeof(pRawDesc->pPartDescs[0]), vmdkRawDescPartComp, NULL);
3758
3759	/*
3760	* Check that we don't have overlapping descriptors. If we do, that's an
3761	* indication that the drive is corrupt or that the RTDvm code is buggy.
3762	*/
3763	VDISKRAWPARTDESC const *paPartDescs = pRawDesc->pPartDescs;
3764	for (uint32_t i = 0; i < pRawDesc->cPartDescs; i++)
3765	{
3766	uint64_t offLast = paPartDescs[i].offStartInVDisk + paPartDescs[i].cbData;
3767	if (offLast <= paPartDescs[i].offStartInVDisk)
3768	return vdIfError(pImage->pIfError, VERR_FILESYSTEM_CORRUPT /?/, RT_SRC_POS,
3769	N_("VMDK: Image path: '%s'. Bogus partition descriptor #%u (%#RX64 LB %#RX64%s): Wrap around or zero"),
3770	pImage->pszFilename, i, paPartDescs[i].offStartInVDisk, paPartDescs[i].cbData,
3771	paPartDescs[i].pvPartitionData ? " (data)" : "");
3772	offLast -= 1;
3773
3774	if (i + 1 < pRawDesc->cPartDescs && offLast >= paPartDescs[i + 1].offStartInVDisk)
3775	return vdIfError(pImage->pIfError, VERR_FILESYSTEM_CORRUPT /?/, RT_SRC_POS,
3776	N_("VMDK: Image path: '%s'. Partition descriptor #%u (%#RX64 LB %#RX64%s) overlaps with the next (%#RX64 LB %#RX64%s)"),
3777	pImage->pszFilename, i, paPartDescs[i].offStartInVDisk, paPartDescs[i].cbData,
3778	paPartDescs[i].pvPartitionData ? " (data)" : "", paPartDescs[i + 1].offStartInVDisk,
3779	paPartDescs[i + 1].cbData, paPartDescs[i + 1].pvPartitionData ? " (data)" : "");
3780	if (offLast >= cbSize)
3781	return vdIfError(pImage->pIfError, VERR_FILESYSTEM_CORRUPT /?/, RT_SRC_POS,
3782	N_("VMDK: Image path: '%s'. Partition descriptor #%u (%#RX64 LB %#RX64%s) goes beyond the end of the drive (%#RX64)"),
3783	pImage->pszFilename, i, paPartDescs[i].offStartInVDisk, paPartDescs[i].cbData,
3784	paPartDescs[i].pvPartitionData ? " (data)" : "", cbSize);
3785	}
3786
3787	return VINF_SUCCESS;
3788	}
3789
3790
3791	#ifdef RT_OS_LINUX
3792	/**
3793	* Searches the dir specified in @a pszBlockDevDir for subdirectories with a
3794	* 'dev' file matching @a uDevToLocate.
3795	*
3796	* This is used both
3797	*
3798	* @returns IPRT status code, errors have been reported properly.
3799	* @param pImage For error reporting.
3800	* @param pszBlockDevDir Input: Path to the directory search under.
3801	* Output: Path to the directory containing information
3802	* for @a uDevToLocate.
3803	* @param cbBlockDevDir The size of the buffer @a pszBlockDevDir points to.
3804	* @param uDevToLocate The device number of the block device info dir to
3805	* locate.
3806	* @param pszDevToLocate For error reporting.
3807	*/
3808	static int vmdkFindSysBlockDevPath(PVMDKIMAGE pImage, char *pszBlockDevDir, size_t cbBlockDevDir,
3809	dev_t uDevToLocate, const char *pszDevToLocate)
3810	{
3811	size_t const cchDir = RTPathEnsureTrailingSeparator(pszBlockDevDir, cbBlockDevDir);
3812	AssertReturn(cchDir > 0, VERR_BUFFER_OVERFLOW);
3813
3814	RTDIR hDir = NIL_RTDIR;
3815	int rc = RTDirOpen(&hDir, pszBlockDevDir);
3816	if (RT_SUCCESS(rc))
3817	{
3818	for (;;)
3819	{
3820	RTDIRENTRY Entry;
3821	rc = RTDirRead(hDir, &Entry, NULL);
3822	if (RT_SUCCESS(rc))
3823	{
3824	/* We're interested in directories and symlinks. */
3825	if ( Entry.enmType == RTDIRENTRYTYPE_DIRECTORY
3826	\|\| Entry.enmType == RTDIRENTRYTYPE_SYMLINK
3827	\|\| Entry.enmType == RTDIRENTRYTYPE_UNKNOWN)
3828	{
3829	rc = RTStrCopy(&pszBlockDevDir[cchDir], cbBlockDevDir - cchDir, Entry.szName);
3830	AssertContinue(RT_SUCCESS(rc)); /* should not happen! */
3831
3832	dev_t uThisDevNo = ~uDevToLocate;
3833	rc = RTLinuxSysFsReadDevNumFile(&uThisDevNo, "%s/dev", pszBlockDevDir);
3834	if (RT_SUCCESS(rc) && uThisDevNo == uDevToLocate)
3835	break;
3836	}
3837	}
3838	else
3839	{
3840	pszBlockDevDir[cchDir] = '\0';
3841	if (rc == VERR_NO_MORE_FILES)
3842	rc = vdIfError(pImage->pIfError, VERR_NOT_FOUND, RT_SRC_POS,
3843	N_("VMDK: Image path: '%s'. Failed to locate device corresponding to '%s' under '%s'"),
3844	pImage->pszFilename, pszDevToLocate, pszBlockDevDir);
3845	else
3846	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
3847	N_("VMDK: Image path: '%s'. RTDirRead failed enumerating '%s': %Rrc"),
3848	pImage->pszFilename, pszBlockDevDir, rc);
3849	break;
3850	}
3851	}
3852	RTDirClose(hDir);
3853	}
3854	else
3855	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
3856	N_("VMDK: Image path: '%s'. Failed to open dir '%s' for listing: %Rrc"),
3857	pImage->pszFilename, pszBlockDevDir, rc);
3858	return rc;
3859	}
3860	#endif /* RT_OS_LINUX */
3861
3862	#ifdef RT_OS_FREEBSD
3863
3864
3865	/**
3866	* Reads the config data from the provider and returns offset and size
3867	*
3868	* @return IPRT status code
3869	* @param pProvider GEOM provider representing partition
3870	* @param pcbOffset Placeholder for the offset of the partition
3871	* @param pcbSize Placeholder for the size of the partition
3872	*/
3873	static int vmdkReadPartitionsParamsFromProvider(gprovider pProvider, uint64_t pcbOffset, uint64_t *pcbSize)
3874	{
3875	gconfig *pConfEntry;
3876	int rc = VERR_NOT_FOUND;
3877
3878	/*
3879	* Required parameters are located in the list containing key/value pairs.
3880	* Both key and value are in text form. Manuals tells nothing about the fact
3881	* that the both parameters should be present in the list. Thus, there are
3882	* cases when only one parameter is presented. To handle such cases we treat
3883	* absent params as zero allowing the caller decide the case is either correct
3884	* or an error.
3885	*/
3886	uint64_t cbOffset = 0;
3887	uint64_t cbSize = 0;
3888	LIST_FOREACH(pConfEntry, &pProvider->lg_config, lg_config)
3889	{
3890	if (RTStrCmp(pConfEntry->lg_name, "offset") == 0)
3891	{
3892	cbOffset = RTStrToUInt64(pConfEntry->lg_val);
3893	rc = VINF_SUCCESS;
3894	}
3895	else if (RTStrCmp(pConfEntry->lg_name, "length") == 0)
3896	{
3897	cbSize = RTStrToUInt64(pConfEntry->lg_val);
3898	rc = VINF_SUCCESS;
3899	}
3900	}
3901	if (RT_SUCCESS(rc))
3902	{
3903	*pcbOffset = cbOffset;
3904	*pcbSize = cbSize;
3905	}
3906	return rc;
3907	}
3908
3909
3910	/**
3911	* Searches the partition specified by name and calculates its size and absolute offset.
3912	*
3913	* @return IPRT status code.
3914	* @param pParentClass Class containing pParentGeom
3915	* @param pszParentGeomName Name of the parent geom where we are looking for provider
3916	* @param pszProviderName Name of the provider we are looking for
3917	* @param pcbAbsoluteOffset Placeholder for the absolute offset of the partition, i.e. offset from the beginning of the disk
3918	* @param psbSize Placeholder for the size of the partition.
3919	*/
3920	static int vmdkFindPartitionParamsByName(gclass pParentClass, const char pszParentGeomName, const char *pszProviderName,
3921	uint64_t pcbAbsoluteOffset, uint64_t pcbSize)
3922	{
3923	AssertReturn(pParentClass, VERR_INVALID_PARAMETER);
3924	AssertReturn(pszParentGeomName, VERR_INVALID_PARAMETER);
3925	AssertReturn(pszProviderName, VERR_INVALID_PARAMETER);
3926	AssertReturn(pcbAbsoluteOffset, VERR_INVALID_PARAMETER);
3927	AssertReturn(pcbSize, VERR_INVALID_PARAMETER);
3928
3929	ggeom *pParentGeom;
3930	int rc = VERR_NOT_FOUND;
3931	LIST_FOREACH(pParentGeom, &pParentClass->lg_geom, lg_geom)
3932	{
3933	if (RTStrCmp(pParentGeom->lg_name, pszParentGeomName) == 0)
3934	{
3935	rc = VINF_SUCCESS;
3936	break;
3937	}
3938	}
3939	if (RT_FAILURE(rc))
3940	return rc;
3941
3942	gprovider *pProvider;
3943	/*
3944	* First, go over providers without handling EBR or BSDLabel
3945	* partitions for case when looking provider is child
3946	* of the givng geom, to reduce searching time
3947	*/
3948	LIST_FOREACH(pProvider, &pParentGeom->lg_provider, lg_provider)
3949	{
3950	if (RTStrCmp(pProvider->lg_name, pszProviderName) == 0)
3951	return vmdkReadPartitionsParamsFromProvider(pProvider, pcbAbsoluteOffset, pcbSize);
3952	}
3953
3954	/*
3955	* No provider found. Go over the parent geom again
3956	* and make recursions if geom represents EBR or BSDLabel.
3957	* In this case given parent geom contains only EBR or BSDLabel
3958	* partition itself and their own partitions are in the separate
3959	* geoms. Also, partition offsets are relative to geom, so
3960	* we have to add offset from child provider with parent geoms
3961	* provider
3962	*/
3963
3964	LIST_FOREACH(pProvider, &pParentGeom->lg_provider, lg_provider)
3965	{
3966	uint64_t cbOffset = 0;
3967	uint64_t cbSize = 0;
3968	rc = vmdkReadPartitionsParamsFromProvider(pProvider, &cbOffset, &cbSize);
3969	if (RT_FAILURE(rc))
3970	return rc;
3971
3972	uint64_t cbProviderOffset = 0;
3973	uint64_t cbProviderSize = 0;
3974	rc = vmdkFindPartitionParamsByName(pParentClass, pProvider->lg_name, pszProviderName, &cbProviderOffset, &cbProviderSize);
3975	if (RT_SUCCESS(rc))
3976	{
3977	*pcbAbsoluteOffset = cbOffset + cbProviderOffset;
3978	*pcbSize = cbProviderSize;
3979	return rc;
3980	}
3981	}
3982
3983	return VERR_NOT_FOUND;
3984	}
3985	#endif
3986
3987
3988	/**
3989	* Attempts to verify the raw partition path.
3990	*
3991	* We don't want to trust RTDvm and the partition device node morphing blindly.
3992	*/
3993	static int vmdkRawDescVerifyPartitionPath(PVMDKIMAGE pImage, PVDISKRAWPARTDESC pPartDesc, uint32_t idxPartition,
3994	const char *pszRawDrive, RTFILE hRawDrive, uint32_t cbSector, RTDVMVOLUME hVol)
3995	{
3996	RT_NOREF(pImage, pPartDesc, idxPartition, pszRawDrive, hRawDrive, cbSector, hVol);
3997
3998	/*
3999	* Try open the raw partition device.
4000	*/
4001	RTFILE hRawPart = NIL_RTFILE;
4002	int rc = RTFileOpen(&hRawPart, pPartDesc->pszRawDevice, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE);
4003	if (RT_FAILURE(rc))
4004	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4005	N_("VMDK: Image path: '%s'. Failed to open partition #%u on '%s' via '%s' (%Rrc)"),
4006	pImage->pszFilename, idxPartition, pszRawDrive, pPartDesc->pszRawDevice, rc);
4007
4008	/*
4009	* Compare the partition UUID if we can get it.
4010	*/
4011	#ifdef RT_OS_WINDOWS
4012	DWORD cbReturned;
4013
4014	/* 1. Get the device numbers for both handles, they should have the same disk. */
4015	STORAGE_DEVICE_NUMBER DevNum1;
4016	RT_ZERO(DevNum1);
4017	if (!DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_STORAGE_GET_DEVICE_NUMBER,
4018	NULL /pvInBuffer/, 0 /cbInBuffer/, &DevNum1, sizeof(DevNum1), &cbReturned, NULL /pOverlapped/))
4019	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
4020	N_("VMDK: Image path: '%s'. IOCTL_STORAGE_GET_DEVICE_NUMBER failed on '%s': %u"),
4021	pImage->pszFilename, pszRawDrive, GetLastError());
4022
4023	STORAGE_DEVICE_NUMBER DevNum2;
4024	RT_ZERO(DevNum2);
4025	if (!DeviceIoControl((HANDLE)RTFileToNative(hRawPart), IOCTL_STORAGE_GET_DEVICE_NUMBER,
4026	NULL /pvInBuffer/, 0 /cbInBuffer/, &DevNum2, sizeof(DevNum2), &cbReturned, NULL /pOverlapped/))
4027	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
4028	N_("VMDK: Image path: '%s'. IOCTL_STORAGE_GET_DEVICE_NUMBER failed on '%s': %u"),
4029	pImage->pszFilename, pPartDesc->pszRawDevice, GetLastError());
4030	if ( RT_SUCCESS(rc)
4031	&& ( DevNum1.DeviceNumber != DevNum2.DeviceNumber
4032	\|\| DevNum1.DeviceType != DevNum2.DeviceType))
4033	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4034	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s' (%#x != %#x \|\| %#x != %#x)"),
4035	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4036	DevNum1.DeviceNumber, DevNum2.DeviceNumber, DevNum1.DeviceType, DevNum2.DeviceType);
4037	if (RT_SUCCESS(rc))
4038	{
4039	/* Get the partitions from the raw drive and match up with the volume info
4040	from RTDvm. The partition number is found in DevNum2. */
4041	DWORD cbNeeded = 0;
4042	if ( DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_DISK_GET_DRIVE_LAYOUT_EX,
4043	NULL /pvInBuffer/, 0 /cbInBuffer/, NULL, 0, &cbNeeded, NULL /pOverlapped/)
4044	\|\| cbNeeded < RT_UOFFSETOF_DYN(DRIVE_LAYOUT_INFORMATION_EX, PartitionEntry[1]))
4045	cbNeeded = RT_UOFFSETOF_DYN(DRIVE_LAYOUT_INFORMATION_EX, PartitionEntry[64]);
4046	cbNeeded += sizeof(PARTITION_INFORMATION_EX) * 2; /* just in case */
4047	DRIVE_LAYOUT_INFORMATION_EX pLayout = (DRIVE_LAYOUT_INFORMATION_EX )RTMemTmpAllocZ(cbNeeded);
4048	if (pLayout)
4049	{
4050	cbReturned = 0;
4051	if (DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_DISK_GET_DRIVE_LAYOUT_EX,
4052	NULL /pvInBuffer/, 0 /cbInBuffer/, pLayout, cbNeeded, &cbReturned, NULL /pOverlapped/))
4053	{
4054	/* Find the entry with the given partition number (it's not an index, array contains empty MBR entries ++). */
4055	unsigned iEntry = 0;
4056	while ( iEntry < pLayout->PartitionCount
4057	&& pLayout->PartitionEntry[iEntry].PartitionNumber != DevNum2.PartitionNumber)
4058	iEntry++;
4059	if (iEntry < pLayout->PartitionCount)
4060	{
4061	/* Compare the basics */
4062	PARTITION_INFORMATION_EX const * const pLayoutEntry = &pLayout->PartitionEntry[iEntry];
4063	if (pLayoutEntry->StartingOffset.QuadPart != (int64_t)pPartDesc->offStartInVDisk)
4064	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4065	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': StartingOffset %RU64, expected %RU64"),
4066	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4067	pLayoutEntry->StartingOffset.QuadPart, pPartDesc->offStartInVDisk);
4068	else if (pLayoutEntry->PartitionLength.QuadPart != (int64_t)pPartDesc->cbData)
4069	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4070	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': PartitionLength %RU64, expected %RU64"),
4071	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4072	pLayoutEntry->PartitionLength.QuadPart, pPartDesc->cbData);
4073	/** @todo We could compare the MBR type, GPT type and ID. */
4074	RT_NOREF(hVol);
4075	}
4076	else
4077	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4078	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': PartitionCount (%#x vs %#x)"),
4079	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4080	DevNum2.PartitionNumber, pLayout->PartitionCount);
4081	# ifndef LOG_ENABLED
4082	if (RT_FAILURE(rc))
4083	# endif
4084	{
4085	LogRel(("VMDK: Windows reports %u partitions for '%s':\n", pLayout->PartitionCount, pszRawDrive));
4086	PARTITION_INFORMATION_EX const *pEntry = &pLayout->PartitionEntry[0];
4087	for (DWORD i = 0; i < pLayout->PartitionCount; i++, pEntry++)
4088	{
4089	LogRel(("VMDK: #%u/%u: %016RU64 LB %016RU64 style=%d rewrite=%d",
4090	i, pEntry->PartitionNumber, pEntry->StartingOffset.QuadPart, pEntry->PartitionLength.QuadPart,
4091	pEntry->PartitionStyle, pEntry->RewritePartition));
4092	if (pEntry->PartitionStyle == PARTITION_STYLE_MBR)
4093	LogRel((" type=%#x boot=%d rec=%d hidden=%u\n", pEntry->Mbr.PartitionType, pEntry->Mbr.BootIndicator,
4094	pEntry->Mbr.RecognizedPartition, pEntry->Mbr.HiddenSectors));
4095	else if (pEntry->PartitionStyle == PARTITION_STYLE_GPT)
4096	LogRel((" type=%RTuuid id=%RTuuid aatrib=%RX64 name=%.36ls\n", &pEntry->Gpt.PartitionType,
4097	&pEntry->Gpt.PartitionId, pEntry->Gpt.Attributes, &pEntry->Gpt.Name[0]));
4098	else
4099	LogRel(("\n"));
4100	}
4101	LogRel(("VMDK: Looked for partition #%u (%u, '%s') at %RU64 LB %RU64\n", DevNum2.PartitionNumber,
4102	idxPartition, pPartDesc->pszRawDevice, pPartDesc->offStartInVDisk, pPartDesc->cbData));
4103	}
4104	}
4105	else
4106	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
4107	N_("VMDK: Image path: '%s'. IOCTL_DISK_GET_DRIVE_LAYOUT_EX failed on '%s': %u (cb %u, cbRet %u)"),
4108	pImage->pszFilename, pPartDesc->pszRawDevice, GetLastError(), cbNeeded, cbReturned);
4109	RTMemTmpFree(pLayout);
4110	}
4111	else
4112	rc = VERR_NO_TMP_MEMORY;
4113	}
4114
4115	#elif defined(RT_OS_LINUX)
4116	RT_NOREF(hVol);
4117
4118	/* Stat the two devices first to get their device numbers. (We probably
4119	could make some assumptions here about the major & minor number assignments
4120	for legacy nodes, but it doesn't hold up for nvme, so we'll skip that.) */
4121	struct stat StDrive, StPart;
4122	if (fstat((int)RTFileToNative(hRawDrive), &StDrive) != 0)
4123	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4124	N_("VMDK: Image path: '%s'. fstat failed on '%s': %d"), pImage->pszFilename, pszRawDrive, errno);
4125	else if (fstat((int)RTFileToNative(hRawPart), &StPart) != 0)
4126	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4127	N_("VMDK: Image path: '%s'. fstat failed on '%s': %d"), pImage->pszFilename, pPartDesc->pszRawDevice, errno);
4128	else
4129	{
4130	/* Scan the directories immediately under /sys/block/ for one with a
4131	'dev' file matching the drive's device number: */
4132	char szSysPath[RTPATH_MAX];
4133	rc = RTLinuxConstructPath(szSysPath, sizeof(szSysPath), "block/");
4134	AssertRCReturn(rc, rc); /* this shall not fail */
4135	if (RTDirExists(szSysPath))
4136	{
4137	rc = vmdkFindSysBlockDevPath(pImage, szSysPath, sizeof(szSysPath), StDrive.st_rdev, pszRawDrive);
4138
4139	/* Now, scan the directories under that again for a partition device
4140	matching the hRawPart device's number: */
4141	if (RT_SUCCESS(rc))
4142	rc = vmdkFindSysBlockDevPath(pImage, szSysPath, sizeof(szSysPath), StPart.st_rdev, pPartDesc->pszRawDevice);
4143
4144	/* Having found the /sys/block/device/partition/ path, we can finally
4145	read the partition attributes and compare with hVol. */
4146	if (RT_SUCCESS(rc))
4147	{
4148	/* partition number: */
4149	int64_t iLnxPartition = 0;
4150	rc = RTLinuxSysFsReadIntFile(10, &iLnxPartition, "%s/partition", szSysPath);
4151	if (RT_SUCCESS(rc) && iLnxPartition != idxPartition)
4152	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4153	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Partition number %RI64, expected %RU32"),
4154	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, iLnxPartition, idxPartition);
4155	/* else: ignore failure? */
4156
4157	/* start offset: */
4158	uint32_t const cbLnxSector = 512; /* It's hardcoded in the Linux kernel */
4159	if (RT_SUCCESS(rc))
4160	{
4161	int64_t offLnxStart = -1;
4162	rc = RTLinuxSysFsReadIntFile(10, &offLnxStart, "%s/start", szSysPath);
4163	offLnxStart *= cbLnxSector;
4164	if (RT_SUCCESS(rc) && offLnxStart != (int64_t)pPartDesc->offStartInVDisk)
4165	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4166	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RI64, expected %RU64"),
4167	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, offLnxStart, pPartDesc->offStartInVDisk);
4168	/* else: ignore failure? */
4169	}
4170
4171	/* the size: */
4172	if (RT_SUCCESS(rc))
4173	{
4174	int64_t cbLnxData = -1;
4175	rc = RTLinuxSysFsReadIntFile(10, &cbLnxData, "%s/size", szSysPath);
4176	cbLnxData *= cbLnxSector;
4177	if (RT_SUCCESS(rc) && cbLnxData != (int64_t)pPartDesc->cbData)
4178	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4179	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RI64, expected %RU64"),
4180	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbLnxData, pPartDesc->cbData);
4181	/* else: ignore failure? */
4182	}
4183	}
4184	}
4185	/* else: We've got nothing to work on, so only do content comparison. */
4186	}
4187
4188	#elif defined(RT_OS_FREEBSD)
4189	char szDriveDevName[256];
4190	char* pszDevName = fdevname_r(RTFileToNative(hRawDrive), szDriveDevName, 256);
4191	if (pszDevName == NULL)
4192	rc = vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4193	N_("VMDK: Image path: '%s'. '%s' is not a drive path"), pImage->pszFilename, pszRawDrive);
4194	char szPartDevName[256];
4195	if (RT_SUCCESS(rc))
4196	{
4197	pszDevName = fdevname_r(RTFileToNative(hRawPart), szPartDevName, 256);
4198	if (pszDevName == NULL)
4199	rc = vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4200	N_("VMDK: Image path: '%s'. '%s' is not a partition path"), pImage->pszFilename, pPartDesc->pszRawDevice);
4201	}
4202	if (RT_SUCCESS(rc))
4203	{
4204	gmesh geomMesh;
4205	int err = geom_gettree(&geomMesh);
4206	if (err == 0)
4207	{
4208	/* Find root class containg partitions info */
4209	gclass* pPartClass;
4210	LIST_FOREACH(pPartClass, &geomMesh.lg_class, lg_class)
4211	{
4212	if (RTStrCmp(pPartClass->lg_name, "PART") == 0)
4213	break;
4214	}
4215	if (pPartClass == NULL \|\| RTStrCmp(pPartClass->lg_name, "PART") != 0)
4216	rc = vdIfError(pImage->pIfError, VERR_GENERAL_FAILURE, RT_SRC_POS,
4217	N_("VMDK: Image path: '%s'. 'PART' class not found in the GEOM tree"), pImage->pszFilename);
4218
4219
4220	if (RT_SUCCESS(rc))
4221	{
4222	/* Find provider representing partition device */
4223	uint64_t cbOffset;
4224	uint64_t cbSize;
4225	rc = vmdkFindPartitionParamsByName(pPartClass, szDriveDevName, szPartDevName, &cbOffset, &cbSize);
4226	if (RT_SUCCESS(rc))
4227	{
4228	if (cbOffset != pPartDesc->offStartInVDisk)
4229	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4230	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RU64, expected %RU64"),
4231	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbOffset, pPartDesc->offStartInVDisk);
4232	if (cbSize != pPartDesc->cbData)
4233	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4234	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RU64, expected %RU64"),
4235	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbSize, pPartDesc->cbData);
4236	}
4237	else
4238	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4239	N_("VMDK: Image path: '%s'. Error getting geom provider for the partition '%s' of the drive '%s' in the GEOM tree: %Rrc"),
4240	pImage->pszFilename, pPartDesc->pszRawDevice, pszRawDrive, rc);
4241	}
4242
4243	geom_deletetree(&geomMesh);
4244	}
4245	else
4246	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(err), RT_SRC_POS,
4247	N_("VMDK: Image path: '%s'. geom_gettree failed: %d"), pImage->pszFilename, err);
4248	}
4249
4250	#elif defined(RT_OS_SOLARIS)
4251	RT_NOREF(hVol);
4252
4253	dk_cinfo dkiDriveInfo;
4254	dk_cinfo dkiPartInfo;
4255	if (ioctl(RTFileToNative(hRawDrive), DKIOCINFO, (caddr_t)&dkiDriveInfo) == -1)
4256	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4257	N_("VMDK: Image path: '%s'. DKIOCINFO failed on '%s': %d"), pImage->pszFilename, pszRawDrive, errno);
4258	else if (ioctl(RTFileToNative(hRawPart), DKIOCINFO, (caddr_t)&dkiPartInfo) == -1)
4259	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4260	N_("VMDK: Image path: '%s'. DKIOCINFO failed on '%s': %d"), pImage->pszFilename, pszRawDrive, errno);
4261	else if ( dkiDriveInfo.dki_ctype != dkiPartInfo.dki_ctype
4262	\|\| dkiDriveInfo.dki_cnum != dkiPartInfo.dki_cnum
4263	\|\| dkiDriveInfo.dki_addr != dkiPartInfo.dki_addr
4264	\|\| dkiDriveInfo.dki_unit != dkiPartInfo.dki_unit
4265	\|\| dkiDriveInfo.dki_slave != dkiPartInfo.dki_slave)
4266	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4267	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s' (%#x != %#x \|\| %#x != %#x \|\| %#x != %#x \|\| %#x != %#x \|\| %#x != %#x)"),
4268	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4269	dkiDriveInfo.dki_ctype, dkiPartInfo.dki_ctype, dkiDriveInfo.dki_cnum, dkiPartInfo.dki_cnum,
4270	dkiDriveInfo.dki_addr, dkiPartInfo.dki_addr, dkiDriveInfo.dki_unit, dkiPartInfo.dki_unit,
4271	dkiDriveInfo.dki_slave, dkiPartInfo.dki_slave);
4272	else
4273	{
4274	uint64_t cbOffset = 0;
4275	uint64_t cbSize = 0;
4276	dk_gpt *pEfi = NULL;
4277	int idxEfiPart = efi_alloc_and_read(RTFileToNative(hRawPart), &pEfi);
4278	if (idxEfiPart >= 0)
4279	{
4280	if ((uint32_t)dkiPartInfo.dki_partition + 1 == idxPartition)
4281	{
4282	cbOffset = pEfi->efi_parts[idxEfiPart].p_start * pEfi->efi_lbasize;
4283	cbSize = pEfi->efi_parts[idxEfiPart].p_size * pEfi->efi_lbasize;
4284	}
4285	else
4286	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4287	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s' (%#x != %#x)"),
4288	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4289	idxPartition, (uint32_t)dkiPartInfo.dki_partition + 1);
4290	efi_free(pEfi);
4291	}
4292	else
4293	{
4294	/*
4295	* Manual says the efi_alloc_and_read returns VT_EINVAL if no EFI partition table found.
4296	* Actually, the function returns any error, e.g. VT_ERROR. Thus, we are not sure, is it
4297	* real error or just no EFI table found. Therefore, let's try to obtain partition info
4298	* using another way. If there is an error, it returns errno which will be handled below.
4299	*/
4300
4301	uint32_t numPartition = (uint32_t)dkiPartInfo.dki_partition;
4302	if (numPartition > NDKMAP)
4303	numPartition -= NDKMAP;
4304	if (numPartition != idxPartition)
4305	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4306	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s' (%#x != %#x)"),
4307	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4308	idxPartition, numPartition);
4309	else
4310	{
4311	dk_minfo_ext mediaInfo;
4312	if (ioctl(RTFileToNative(hRawPart), DKIOCGMEDIAINFOEXT, (caddr_t)&mediaInfo) == -1)
4313	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4314	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s'. Can not obtain partition info: %d"),
4315	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4316	else
4317	{
4318	extpart_info extPartInfo;
4319	if (ioctl(RTFileToNative(hRawPart), DKIOCEXTPARTINFO, (caddr_t)&extPartInfo) != -1)
4320	{
4321	cbOffset = (uint64_t)extPartInfo.p_start * mediaInfo.dki_lbsize;
4322	cbSize = (uint64_t)extPartInfo.p_length * mediaInfo.dki_lbsize;
4323	}
4324	else
4325	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4326	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s'. Can not obtain partition info: %d"),
4327	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4328	}
4329	}
4330	}
4331	if (RT_SUCCESS(rc) && cbOffset != pPartDesc->offStartInVDisk)
4332	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4333	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RI64, expected %RU64"),
4334	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbOffset, pPartDesc->offStartInVDisk);
4335
4336	if (RT_SUCCESS(rc) && cbSize != pPartDesc->cbData)
4337	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4338	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RI64, expected %RU64"),
4339	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbSize, pPartDesc->cbData);
4340	}
4341
4342	#elif defined(RT_OS_DARWIN)
4343	/* Stat the drive get its device number. */
4344	struct stat StDrive;
4345	if (fstat((int)RTFileToNative(hRawDrive), &StDrive) != 0)
4346	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4347	N_("VMDK: Image path: '%s'. fstat failed on '%s' (errno=%d)"), pImage->pszFilename, pszRawDrive, errno);
4348	else
4349	{
4350	if (ioctl(RTFileToNative(hRawPart), DKIOCLOCKPHYSICALEXTENTS, NULL) == -1)
4351	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4352	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to lock the partition (errno=%d)"),
4353	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4354	else
4355	{
4356	uint32_t cbBlockSize = 0;
4357	uint64_t cbOffset = 0;
4358	uint64_t cbSize = 0;
4359	if (ioctl(RTFileToNative(hRawPart), DKIOCGETBLOCKSIZE, (caddr_t)&cbBlockSize) == -1)
4360	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4361	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain the sector size of the partition (errno=%d)"),
4362	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4363	else if (ioctl(RTFileToNative(hRawPart), DKIOCGETBASE, (caddr_t)&cbOffset) == -1)
4364	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4365	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain the start offset of the partition (errno=%d)"),
4366	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4367	else if (ioctl(RTFileToNative(hRawPart), DKIOCGETBLOCKCOUNT, (caddr_t)&cbSize) == -1)
4368	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4369	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain the size of the partition (errno=%d)"),
4370	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4371	else
4372	{
4373	cbSize *= (uint64_t)cbBlockSize;
4374	dk_physical_extent_t dkPartExtent = {0};
4375	dkPartExtent.offset = 0;
4376	dkPartExtent.length = cbSize;
4377	if (ioctl(RTFileToNative(hRawPart), DKIOCGETPHYSICALEXTENT, (caddr_t)&dkPartExtent) == -1)
4378	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4379	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain partition info (errno=%d)"),
4380	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4381	else
4382	{
4383	if (dkPartExtent.dev != StDrive.st_rdev)
4384	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4385	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Drive does not contain the partition"),
4386	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive);
4387	else if (cbOffset != pPartDesc->offStartInVDisk)
4388	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4389	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RU64, expected %RU64"),
4390	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbOffset, pPartDesc->offStartInVDisk);
4391	else if (cbSize != pPartDesc->cbData)
4392	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4393	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RU64, expected %RU64"),
4394	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbSize, pPartDesc->cbData);
4395	}
4396	}
4397
4398	if (ioctl(RTFileToNative(hRawPart), DKIOCUNLOCKPHYSICALEXTENTS, NULL) == -1)
4399	{
4400	int rc2 = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4401	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to unlock the partition (errno=%d)"),
4402	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4403	if (RT_SUCCESS(rc))
4404	rc = rc2;
4405	}
4406	}
4407	}
4408
4409	#else
4410	RT_NOREF(hVol); /* PORTME */
4411	rc = VERR_NOT_SUPPORTED;
4412	#endif
4413	if (RT_SUCCESS(rc))
4414	{
4415	/*
4416	* Compare the first 32 sectors of the partition.
4417	*
4418	* This might not be conclusive, but for partitions formatted with the more
4419	* common file systems it should be as they have a superblock copy at or near
4420	* the start of the partition (fat, fat32, ntfs, and ext4 does at least).
4421	*/
4422	size_t const cbToCompare = (size_t)RT_MIN(pPartDesc->cbData / cbSector, 32) * cbSector;
4423	uint8_t pbSector1 = (uint8_t )RTMemTmpAlloc(cbToCompare * 2);
4424	if (pbSector1 != NULL)
4425	{
4426	uint8_t *pbSector2 = pbSector1 + cbToCompare;
4427
4428	/* Do the comparing, we repeat if it fails and the data might be volatile. */
4429	uint64_t uPrevCrc1 = 0;
4430	uint64_t uPrevCrc2 = 0;
4431	uint32_t cStable = 0;
4432	for (unsigned iTry = 0; iTry < 256; iTry++)
4433	{
4434	rc = RTFileReadAt(hRawDrive, pPartDesc->offStartInVDisk, pbSector1, cbToCompare, NULL);
4435	if (RT_SUCCESS(rc))
4436	{
4437	rc = RTFileReadAt(hRawPart, pPartDesc->offStartInDevice, pbSector2, cbToCompare, NULL);
4438	if (RT_SUCCESS(rc))
4439	{
4440	if (memcmp(pbSector1, pbSector2, cbToCompare) != 0)
4441	{
4442	rc = VERR_MISMATCH;
4443
4444	/* Do data stability checks before repeating: */
4445	uint64_t const uCrc1 = RTCrc64(pbSector1, cbToCompare);
4446	uint64_t const uCrc2 = RTCrc64(pbSector2, cbToCompare);
4447	if ( uPrevCrc1 != uCrc1
4448	\|\| uPrevCrc2 != uCrc2)
4449	cStable = 0;
4450	else if (++cStable > 4)
4451	break;
4452	uPrevCrc1 = uCrc1;
4453	uPrevCrc2 = uCrc2;
4454	continue;
4455	}
4456	rc = VINF_SUCCESS;
4457	}
4458	else
4459	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4460	N_("VMDK: Image path: '%s'. Error reading %zu bytes from '%s' at offset %RU64 (%Rrc)"),
4461	pImage->pszFilename, cbToCompare, pPartDesc->pszRawDevice, pPartDesc->offStartInDevice, rc);
4462	}
4463	else
4464	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4465	N_("VMDK: Image path: '%s'. Error reading %zu bytes from '%s' at offset %RU64 (%Rrc)"),
4466	pImage->pszFilename, cbToCompare, pszRawDrive, pPartDesc->offStartInVDisk, rc);
4467	break;
4468	}
4469	if (rc == VERR_MISMATCH)
4470	{
4471	/* Find the first mismatching bytes: */
4472	size_t offMissmatch = 0;
4473	while (offMissmatch < cbToCompare && pbSector1[offMissmatch] == pbSector2[offMissmatch])
4474	offMissmatch++;
4475	int cbSample = (int)RT_MIN(cbToCompare - offMissmatch, 16);
4476
4477	if (cStable > 0)
4478	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4479	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s' (cStable=%d @%#zx: %.Rhxs vs %.Rhxs)"),
4480	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cStable,
4481	offMissmatch, cbSample, &pbSector1[offMissmatch], cbSample, &pbSector2[offMissmatch]);
4482	else
4483	{
4484	LogRel(("VMDK: Image path: '%s'. Partition #%u path ('%s') verification undecided on '%s' because of unstable data! (@%#zx: %.Rhxs vs %.Rhxs)\n",
4485	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4486	offMissmatch, cbSample, &pbSector1[offMissmatch], cbSample, &pbSector2[offMissmatch]));
4487	rc = -rc;
4488	}
4489	}
4490
4491	RTMemTmpFree(pbSector1);
4492	}
4493	else
4494	rc = vdIfError(pImage->pIfError, VERR_NO_TMP_MEMORY, RT_SRC_POS,
4495	N_("VMDK: Image path: '%s'. Failed to allocate %zu bytes for a temporary read buffer\n"),
4496	pImage->pszFilename, cbToCompare * 2);
4497	}
4498	RTFileClose(hRawPart);
4499	return rc;
4500	}
4501
4502	#ifdef RT_OS_WINDOWS
4503	/**
4504	* Construct the device name for the given partition number.
4505	*/
4506	static int vmdkRawDescWinMakePartitionName(PVMDKIMAGE pImage, const char *pszRawDrive, RTFILE hRawDrive, uint32_t idxPartition,
4507	char **ppszRawPartition)
4508	{
4509	int rc = VINF_SUCCESS;
4510	DWORD cbReturned = 0;
4511	STORAGE_DEVICE_NUMBER DevNum;
4512	RT_ZERO(DevNum);
4513	if (DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_STORAGE_GET_DEVICE_NUMBER,
4514	NULL /pvInBuffer/, 0 /cbInBuffer/, &DevNum, sizeof(DevNum), &cbReturned, NULL /pOverlapped/))
4515	RTStrAPrintf(ppszRawPartition, "\\\\.\\Harddisk%uPartition%u", DevNum.DeviceNumber, idxPartition);
4516	else
4517	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
4518	N_("VMDK: Image path: '%s'. IOCTL_STORAGE_GET_DEVICE_NUMBER failed on '%s': %u"),
4519	pImage->pszFilename, pszRawDrive, GetLastError());
4520	return rc;
4521	}
4522	#endif /* RT_OS_WINDOWS */
4523
4524	/**
4525	* Worker for vmdkMakeRawDescriptor that adds partition descriptors when the
4526	* 'Partitions' configuration value is present.
4527	*
4528	* @returns VBox status code, error message has been set on failure.
4529	*
4530	* @note Caller is assumed to clean up @a pRawDesc and release
4531	* @a *phVolToRelease.
4532	* @internal
4533	*/
4534	static int vmdkRawDescDoPartitions(PVMDKIMAGE pImage, RTDVM hVolMgr, PVDISKRAW pRawDesc,
4535	RTFILE hRawDrive, const char *pszRawDrive, uint32_t cbSector,
4536	uint32_t fPartitions, uint32_t fPartitionsReadOnly, bool fRelative,
4537	PRTDVMVOLUME phVolToRelease)
4538	{
4539	*phVolToRelease = NIL_RTDVMVOLUME;
4540
4541	/* Check sanity/understanding. */
4542	Assert(fPartitions);
4543	Assert((fPartitions & fPartitionsReadOnly) == fPartitionsReadOnly); /* RO should be a sub-set */
4544
4545	/*
4546	* Allocate on descriptor for each volume up front.
4547	*/
4548	uint32_t const cVolumes = RTDvmMapGetValidVolumes(hVolMgr);
4549
4550	PVDISKRAWPARTDESC paPartDescs = NULL;
4551	int rc = vmdkRawDescAppendPartDesc(pImage, pRawDesc, cVolumes, &paPartDescs);
4552	AssertRCReturn(rc, rc);
4553
4554	/*
4555	* Enumerate the partitions (volumes) on the disk and create descriptors for each of them.
4556	*/
4557	uint32_t fPartitionsLeft = fPartitions;
4558	RTDVMVOLUME hVol = NIL_RTDVMVOLUME; /* the current volume, needed for getting the next. */
4559	for (uint32_t i = 0; i < cVolumes; i++)
4560	{
4561	/*
4562	* Get the next/first volume and release the current.
4563	*/
4564	RTDVMVOLUME hVolNext = NIL_RTDVMVOLUME;
4565	if (i == 0)
4566	rc = RTDvmMapQueryFirstVolume(hVolMgr, &hVolNext);
4567	else
4568	rc = RTDvmMapQueryNextVolume(hVolMgr, hVol, &hVolNext);
4569	if (RT_FAILURE(rc))
4570	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4571	N_("VMDK: Image path: '%s'. Volume enumeration failed at volume #%u on '%s' (%Rrc)"),
4572	pImage->pszFilename, i, pszRawDrive, rc);
4573	uint32_t cRefs = RTDvmVolumeRelease(hVol);
4574	Assert(cRefs != UINT32_MAX); RT_NOREF(cRefs);
4575	*phVolToRelease = hVol = hVolNext;
4576
4577	/*
4578	* Depending on the fPartitions selector and associated read-only mask,
4579	* the guest either gets read-write or read-only access (bits set)
4580	* or no access (selector bit clear, access directed to the VMDK).
4581	*/
4582	paPartDescs[i].cbData = RTDvmVolumeGetSize(hVol);
4583
4584	uint64_t offVolumeEndIgnored = 0;
4585	rc = RTDvmVolumeQueryRange(hVol, &paPartDescs[i].offStartInVDisk, &offVolumeEndIgnored);
4586	if (RT_FAILURE(rc))
4587	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4588	N_("VMDK: Image path: '%s'. Failed to get location of volume #%u on '%s' (%Rrc)"),
4589	pImage->pszFilename, i, pszRawDrive, rc);
4590	Assert(paPartDescs[i].cbData == offVolumeEndIgnored + 1 - paPartDescs[i].offStartInVDisk);
4591
4592	/* Note! The index must match IHostDrivePartition::number. */
4593	uint32_t idxPartition = RTDvmVolumeGetIndex(hVol, RTDVMVOLIDX_HOST);
4594	if ( idxPartition < 32
4595	&& (fPartitions & RT_BIT_32(idxPartition)))
4596	{
4597	fPartitionsLeft &= ~RT_BIT_32(idxPartition);
4598	if (fPartitionsReadOnly & RT_BIT_32(idxPartition))
4599	paPartDescs[i].uFlags \|= VDISKRAW_READONLY;
4600
4601	if (!fRelative)
4602	{
4603	/*
4604	* Accessing the drive thru the main device node (pRawDesc->pszRawDisk).
4605	*/
4606	paPartDescs[i].offStartInDevice = paPartDescs[i].offStartInVDisk;
4607	paPartDescs[i].pszRawDevice = RTStrDup(pszRawDrive);
4608	AssertPtrReturn(paPartDescs[i].pszRawDevice, VERR_NO_STR_MEMORY);
4609	}
4610	else
4611	{
4612	/*
4613	* Relative means access the partition data via the device node for that
4614	* partition, allowing the sysadmin/OS to allow a user access to individual
4615	* partitions without necessarily being able to compromise the host OS.
4616	* Obviously, the creation of the VMDK requires read access to the main
4617	* device node for the drive, but that's a one-time thing and can be done
4618	* by the sysadmin. Here data starts at offset zero in the device node.
4619	*/
4620	paPartDescs[i].offStartInDevice = 0;
4621
4622	#if defined(RT_OS_DARWIN) \|\| defined(RT_OS_FREEBSD)
4623	/* /dev/rdisk1 -> /dev/rdisk1s2 (s=slice) */
4624	RTStrAPrintf(&paPartDescs[i].pszRawDevice, "%ss%u", pszRawDrive, idxPartition);
4625	#elif defined(RT_OS_LINUX)
4626	/* Two naming schemes here: /dev/nvme0n1 -> /dev/nvme0n1p1; /dev/sda -> /dev/sda1 */
4627	RTStrAPrintf(&paPartDescs[i].pszRawDevice,
4628	RT_C_IS_DIGIT(pszRawDrive[strlen(pszRawDrive) - 1]) ? "%sp%u" : "%s%u", pszRawDrive, idxPartition);
4629	#elif defined(RT_OS_WINDOWS)
4630	rc = vmdkRawDescWinMakePartitionName(pImage, pszRawDrive, hRawDrive, idxPartition, &paPartDescs[i].pszRawDevice);
4631	AssertRCReturn(rc, rc);
4632	#elif defined(RT_OS_SOLARIS)
4633	if (pRawDesc->enmPartitioningType == VDISKPARTTYPE_MBR)
4634	{
4635	/*
4636	* MBR partitions have device nodes in form /dev/(r)dsk/cXtYdZpK
4637	* where X is the controller,
4638	* Y is target (SCSI device number),
4639	* Z is disk number,
4640	* K is partition number,
4641	* where p0 is the whole disk
4642	* p1-pN are the partitions of the disk
4643	*/
4644	const char *pszRawDrivePath = pszRawDrive;
4645	char szDrivePath[RTPATH_MAX];
4646	size_t cbRawDrive = strlen(pszRawDrive);
4647	if ( cbRawDrive > 1 && strcmp(&pszRawDrive[cbRawDrive - 2], "p0") == 0)
4648	{
4649	memcpy(szDrivePath, pszRawDrive, cbRawDrive - 2);
4650	szDrivePath[cbRawDrive - 2] = '\0';
4651	pszRawDrivePath = szDrivePath;
4652	}
4653	RTStrAPrintf(&paPartDescs[i].pszRawDevice, "%sp%u", pszRawDrivePath, idxPartition);
4654	}
4655	else /* GPT */
4656	{
4657	/*
4658	* GPT partitions have device nodes in form /dev/(r)dsk/cXtYdZsK
4659	* where X is the controller,
4660	* Y is target (SCSI device number),
4661	* Z is disk number,
4662	* K is partition number, zero based. Can be only from 0 to 6.
4663	* Thus, only partitions numbered 0 through 6 have device nodes.
4664	*/
4665	if (idxPartition > 7)
4666	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4667	N_("VMDK: Image path: '%s'. the partition #%u on '%s' has no device node and can not be specified with 'Relative' property"),
4668	pImage->pszFilename, idxPartition, pszRawDrive);
4669	RTStrAPrintf(&paPartDescs[i].pszRawDevice, "%ss%u", pszRawDrive, idxPartition - 1);
4670	}
4671	#else
4672	AssertFailedReturn(VERR_INTERNAL_ERROR_4); /* The option parsing code should have prevented this - PORTME */
4673	#endif
4674	AssertPtrReturn(paPartDescs[i].pszRawDevice, VERR_NO_STR_MEMORY);
4675
4676	rc = vmdkRawDescVerifyPartitionPath(pImage, &paPartDescs[i], idxPartition, pszRawDrive, hRawDrive, cbSector, hVol);
4677	AssertRCReturn(rc, rc);
4678	}
4679	}
4680	else
4681	{
4682	/* Not accessible to the guest. */
4683	paPartDescs[i].offStartInDevice = 0;
4684	paPartDescs[i].pszRawDevice = NULL;
4685	}
4686	} /* for each volume */
4687
4688	RTDvmVolumeRelease(hVol);
4689	*phVolToRelease = NIL_RTDVMVOLUME;
4690
4691	/*
4692	* Check that we found all the partitions the user selected.
4693	*/
4694	if (fPartitionsLeft)
4695	{
4696	char szLeft[3 * sizeof(fPartitions) * 8];
4697	size_t cchLeft = 0;
4698	for (unsigned i = 0; i < sizeof(fPartitions) * 8; i++)
4699	if (fPartitionsLeft & RT_BIT_32(i))
4700	cchLeft += RTStrPrintf(&szLeft[cchLeft], sizeof(szLeft) - cchLeft, cchLeft ? "%u" : ",%u", i);
4701	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4702	N_("VMDK: Image path: '%s'. Not all the specified partitions for drive '%s' was found: %s"),
4703	pImage->pszFilename, pszRawDrive, szLeft);
4704	}
4705
4706	return VINF_SUCCESS;
4707	}
4708
4709	/**
4710	* Worker for vmdkMakeRawDescriptor that adds partition descriptors with copies
4711	* of the partition tables and associated padding areas when the 'Partitions'
4712	* configuration value is present.
4713	*
4714	* The guest is not allowed access to the partition tables, however it needs
4715	* them to be able to access the drive. So, create descriptors for each of the
4716	* tables and attach the current disk content. vmdkCreateRawImage() will later
4717	* write the content to the VMDK. Any changes the guest later makes to the
4718	* partition tables will then go to the VMDK copy, rather than the host drive.
4719	*
4720	* @returns VBox status code, error message has been set on failure.
4721	*
4722	* @note Caller is assumed to clean up @a pRawDesc
4723	* @internal
4724	*/
4725	static int vmdkRawDescDoCopyPartitionTables(PVMDKIMAGE pImage, RTDVM hVolMgr, PVDISKRAW pRawDesc,
4726	const char pszRawDrive, RTFILE hRawDrive, void pvBootSector, size_t cbBootSector)
4727	{
4728	/*
4729	* Query the locations.
4730	*/
4731	/* Determin how many locations there are: */
4732	size_t cLocations = 0;
4733	int rc = RTDvmMapQueryTableLocations(hVolMgr, RTDVMMAPQTABLOC_F_INCLUDE_LEGACY, NULL, 0, &cLocations);
4734	if (rc != VERR_BUFFER_OVERFLOW)
4735	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4736	N_("VMDK: Image path: '%s'. RTDvmMapQueryTableLocations failed on '%s' (%Rrc)"),
4737	pImage->pszFilename, pszRawDrive, rc);
4738	AssertReturn(cLocations > 0 && cLocations < _16M, VERR_INTERNAL_ERROR_5);
4739
4740	/* We can allocate the partition descriptors here to save an intentation level. */
4741	PVDISKRAWPARTDESC paPartDescs = NULL;
4742	rc = vmdkRawDescAppendPartDesc(pImage, pRawDesc, (uint32_t)cLocations, &paPartDescs);
4743	AssertRCReturn(rc, rc);
4744
4745	/* Allocate the result table and repeat the location table query: */
4746	PRTDVMTABLELOCATION paLocations = (PRTDVMTABLELOCATION)RTMemAllocZ(sizeof(paLocations[0]) * cLocations);
4747	if (!paLocations)
4748	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS, N_("VMDK: Image path: '%s'. Failed to allocate %zu bytes"),
4749	pImage->pszFilename, sizeof(paLocations[0]) * cLocations);
4750	rc = RTDvmMapQueryTableLocations(hVolMgr, RTDVMMAPQTABLOC_F_INCLUDE_LEGACY, paLocations, cLocations, NULL);
4751	if (RT_SUCCESS(rc))
4752	{
4753	/*
4754	* Translate them into descriptors.
4755	*
4756	* We restrict the amount of partition alignment padding to 4MiB as more
4757	* will just be a waste of space. The use case for including the padding
4758	* are older boot loaders and boot manager (including one by a team member)
4759	* that put data and code in the 62 sectors between the MBR and the first
4760	* partition (total of 63). Later CHS was abandond and partition started
4761	* being aligned on power of two sector boundraries (typically 64KiB or
4762	* 1MiB depending on the media size).
4763	*/
4764	for (size_t i = 0; i < cLocations && RT_SUCCESS(rc); i++)
4765	{
4766	Assert(paLocations[i].cb > 0);
4767	if (paLocations[i].cb <= _64M)
4768	{
4769	/* Create the partition descriptor entry: */
4770	//paPartDescs[i].pszRawDevice = NULL;
4771	//paPartDescs[i].offStartInDevice = 0;
4772	//paPartDescs[i].uFlags = 0;
4773	paPartDescs[i].offStartInVDisk = paLocations[i].off;
4774	paPartDescs[i].cbData = paLocations[i].cb;
4775	if (paPartDescs[i].cbData < _4M)
4776	paPartDescs[i].cbData = RT_MIN(paPartDescs[i].cbData + paLocations[i].cbPadding, _4M);
4777	paPartDescs[i].pvPartitionData = RTMemAllocZ((size_t)paPartDescs[i].cbData);
4778	if (paPartDescs[i].pvPartitionData)
4779	{
4780	/* Read the content from the drive: */
4781	rc = RTFileReadAt(hRawDrive, paPartDescs[i].offStartInVDisk, paPartDescs[i].pvPartitionData,
4782	(size_t)paPartDescs[i].cbData, NULL);
4783	if (RT_SUCCESS(rc))
4784	{
4785	/* Do we have custom boot sector code? */
4786	if (pvBootSector && cbBootSector && paPartDescs[i].offStartInVDisk == 0)
4787	{
4788	/* Note! Old code used to quietly drop the bootsector if it was considered too big.
4789	Instead we fail as we weren't able to do what the user requested us to do.
4790	Better if the user knows than starts questioning why the guest isn't
4791	booting as expected. */
4792	if (cbBootSector <= paPartDescs[i].cbData)
4793	memcpy(paPartDescs[i].pvPartitionData, pvBootSector, cbBootSector);
4794	else
4795	rc = vdIfError(pImage->pIfError, VERR_TOO_MUCH_DATA, RT_SRC_POS,
4796	N_("VMDK: Image path: '%s'. The custom boot sector is too big: %zu bytes, %RU64 bytes available"),
4797	pImage->pszFilename, cbBootSector, paPartDescs[i].cbData);
4798	}
4799	}
4800	else
4801	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4802	N_("VMDK: Image path: '%s'. Failed to read partition at off %RU64 length %zu from '%s' (%Rrc)"),
4803	pImage->pszFilename, paPartDescs[i].offStartInVDisk,
4804	(size_t)paPartDescs[i].cbData, pszRawDrive, rc);
4805	}
4806	else
4807	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4808	N_("VMDK: Image path: '%s'. Failed to allocate %zu bytes for copying the partition table at off %RU64"),
4809	pImage->pszFilename, (size_t)paPartDescs[i].cbData, paPartDescs[i].offStartInVDisk);
4810	}
4811	else
4812	rc = vdIfError(pImage->pIfError, VERR_TOO_MUCH_DATA, RT_SRC_POS,
4813	N_("VMDK: Image path: '%s'. Partition table #%u at offset %RU64 in '%s' is to big: %RU64 bytes"),
4814	pImage->pszFilename, i, paLocations[i].off, pszRawDrive, paLocations[i].cb);
4815	}
4816	}
4817	else
4818	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4819	N_("VMDK: Image path: '%s'. RTDvmMapQueryTableLocations failed on '%s' (%Rrc)"),
4820	pImage->pszFilename, pszRawDrive, rc);
4821	RTMemFree(paLocations);
4822	return rc;
4823	}
4824
4825	/**
4826	* Opens the volume manager for the raw drive when in selected-partition mode.
4827	*
4828	* @param pImage The VMDK image (for errors).
4829	* @param hRawDrive The raw drive handle.
4830	* @param pszRawDrive The raw drive device path (for errors).
4831	* @param cbSector The sector size.
4832	* @param phVolMgr Where to return the handle to the volume manager on
4833	* success.
4834	* @returns VBox status code, errors have been reported.
4835	* @internal
4836	*/
4837	static int vmdkRawDescOpenVolMgr(PVMDKIMAGE pImage, RTFILE hRawDrive, const char *pszRawDrive, uint32_t cbSector, PRTDVM phVolMgr)
4838	{
4839	*phVolMgr = NIL_RTDVM;
4840
4841	RTVFSFILE hVfsFile = NIL_RTVFSFILE;
4842	int rc = RTVfsFileFromRTFile(hRawDrive, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE, true /fLeaveOpen/, &hVfsFile);
4843	if (RT_FAILURE(rc))
4844	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4845	N_("VMDK: Image path: '%s'. RTVfsFileFromRTFile failed for '%s' handle (%Rrc)"),
4846	pImage->pszFilename, pszRawDrive, rc);
4847
4848	RTDVM hVolMgr = NIL_RTDVM;
4849	rc = RTDvmCreate(&hVolMgr, hVfsFile, cbSector, 0 /fFlags/);
4850
4851	RTVfsFileRelease(hVfsFile);
4852
4853	if (RT_FAILURE(rc))
4854	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4855	N_("VMDK: Image path: '%s'. Failed to create volume manager instance for '%s' (%Rrc)"),
4856	pImage->pszFilename, pszRawDrive, rc);
4857
4858	rc = RTDvmMapOpen(hVolMgr);
4859	if (RT_SUCCESS(rc))
4860	{
4861	*phVolMgr = hVolMgr;
4862	return VINF_SUCCESS;
4863	}
4864	RTDvmRelease(hVolMgr);
4865	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: Image path: '%s'. RTDvmMapOpen failed for '%s' (%Rrc)"),
4866	pImage->pszFilename, pszRawDrive, rc);
4867	}
4868
4869	/**
4870	* Opens the raw drive device and get the sizes for it.
4871	*
4872	* @param pImage The image (for error reporting).
4873	* @param pszRawDrive The device/whatever to open.
4874	* @param phRawDrive Where to return the file handle.
4875	* @param pcbRawDrive Where to return the size.
4876	* @param pcbSector Where to return the sector size.
4877	* @returns IPRT status code, errors have been reported.
4878	* @internal
4879	*/
4880	static int vmkdRawDescOpenDevice(PVMDKIMAGE pImage, const char *pszRawDrive,
4881	PRTFILE phRawDrive, uint64_t pcbRawDrive, uint32_t pcbSector)
4882	{
4883	/*
4884	* Open the device for the raw drive.
4885	*/
4886	RTFILE hRawDrive = NIL_RTFILE;
4887	int rc = RTFileOpen(&hRawDrive, pszRawDrive, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE);
4888	if (RT_FAILURE(rc))
4889	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4890	N_("VMDK: Image path: '%s'. Failed to open the raw dri