VMDK.cpp@ 101696

Last change on this file since 101696 was 99963, checked in by vboxsync, 19 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 drive '%s' for reading (%Rrc)"),
4891	pImage->pszFilename, pszRawDrive, rc);
4892
4893	/*
4894	* Get the sector size.
4895	*/
4896	uint32_t cbSector = 0;
4897	rc = RTFileQuerySectorSize(hRawDrive, &cbSector);
4898	if (RT_SUCCESS(rc))
4899	{
4900	/* sanity checks */
4901	if ( cbSector >= 512
4902	&& cbSector <= _64K
4903	&& RT_IS_POWER_OF_TWO(cbSector))
4904	{
4905	/*
4906	* Get the size.
4907	*/
4908	uint64_t cbRawDrive = 0;
4909	rc = RTFileQuerySize(hRawDrive, &cbRawDrive);
4910	if (RT_SUCCESS(rc))
4911	{
4912	/* Check whether cbSize is actually sensible. */
4913	if (cbRawDrive > cbSector && (cbRawDrive % cbSector) == 0)
4914	{
4915	*phRawDrive = hRawDrive;
4916	*pcbRawDrive = cbRawDrive;
4917	*pcbSector = cbSector;
4918	return VINF_SUCCESS;
4919	}
4920	rc = vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4921	N_("VMDK: Image path: '%s'. Got a bogus size for the raw drive '%s': %RU64 (sector size %u)"),
4922	pImage->pszFilename, pszRawDrive, cbRawDrive, cbSector);
4923	}
4924	else
4925	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4926	N_("VMDK: Image path: '%s'. Failed to query size of the drive '%s' (%Rrc)"),
4927	pImage->pszFilename, pszRawDrive, rc);
4928	}
4929	else
4930	rc = vdIfError(pImage->pIfError, VERR_OUT_OF_RANGE, RT_SRC_POS,
4931	N_("VMDK: Image path: '%s'. Unsupported sector size for '%s': %u (%#x)"),
4932	pImage->pszFilename, pszRawDrive, cbSector, cbSector);
4933	}
4934	else
4935	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4936	N_("VMDK: Image path: '%s'. Failed to get the sector size for '%s' (%Rrc)"),
4937	pImage->pszFilename, pszRawDrive, rc);
4938	RTFileClose(hRawDrive);
4939	return rc;
4940	}
4941
4942	/**
4943	* Reads the raw disk configuration, leaving initalization and cleanup to the
4944	* caller (regardless of return status).
4945	*
4946	* @returns VBox status code, errors properly reported.
4947	* @internal
4948	*/
4949	static int vmdkRawDescParseConfig(PVMDKIMAGE pImage, char **ppszRawDrive,
4950	uint32_t pfPartitions, uint32_t pfPartitionsReadOnly,
4951	void *ppvBootSector, size_t pcbBootSector, bool *pfRelative,
4952	char **ppszFreeMe)
4953	{
4954	PVDINTERFACECONFIG pImgCfg = VDIfConfigGet(pImage->pVDIfsImage);
4955	if (!pImgCfg)
4956	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4957	N_("VMDK: Image path: '%s'. Getting config interface failed"), pImage->pszFilename);
4958
4959	/*
4960	* RawDrive = path
4961	*/
4962	int rc = VDCFGQueryStringAlloc(pImgCfg, "RawDrive", ppszRawDrive);
4963	if (RT_FAILURE(rc))
4964	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4965	N_("VMDK: Image path: '%s'. Getting 'RawDrive' configuration failed (%Rrc)"), pImage->pszFilename, rc);
4966	AssertPtrReturn(*ppszRawDrive, VERR_INTERNAL_ERROR_3);
4967
4968	/*
4969	* Partitions=n[r][,...]
4970	*/
4971	uint32_t const cMaxPartitionBits = sizeof(pfPartitions) 8 /* ASSUMES 8 bits per char */;
4972	pfPartitions = pfPartitionsReadOnly = 0;
4973
4974	rc = VDCFGQueryStringAlloc(pImgCfg, "Partitions", ppszFreeMe);
4975	if (RT_SUCCESS(rc))
4976	{
4977	char psz = ppszFreeMe;
4978	while (*psz != '\0')
4979	{
4980	char *pszNext;
4981	uint32_t u32;
4982	rc = RTStrToUInt32Ex(psz, &pszNext, 0, &u32);
4983	if (rc == VWRN_NUMBER_TOO_BIG \|\| rc == VWRN_NEGATIVE_UNSIGNED)
4984	rc = -rc;
4985	if (RT_FAILURE(rc))
4986	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4987	N_("VMDK: Image path: '%s'. Parsing 'Partitions' config value failed. Incorrect value (%Rrc): %s"),
4988	pImage->pszFilename, rc, psz);
4989	if (u32 >= cMaxPartitionBits)
4990	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4991	N_("VMDK: Image path: '%s'. 'Partitions' config sub-value out of range: %RU32, max %RU32"),
4992	pImage->pszFilename, u32, cMaxPartitionBits);
4993	*pfPartitions \|= RT_BIT_32(u32);
4994	psz = pszNext;
4995	if (*psz == 'r')
4996	{
4997	*pfPartitionsReadOnly \|= RT_BIT_32(u32);
4998	psz++;
4999	}
5000	if (*psz == ',')
5001	psz++;
5002	else if (*psz != '\0')
5003	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5004	N_("VMDK: Image path: '%s'. Malformed 'Partitions' config value, expected separator: %s"),
5005	pImage->pszFilename, psz);
5006	}
5007
5008	RTStrFree(*ppszFreeMe);
5009	*ppszFreeMe = NULL;
5010	}
5011	else if (rc != VERR_CFGM_VALUE_NOT_FOUND)
5012	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5013	N_("VMDK: Image path: '%s'. Getting 'Partitions' configuration failed (%Rrc)"), pImage->pszFilename, rc);
5014
5015	/*
5016	* BootSector=base64
5017	*/
5018	rc = VDCFGQueryStringAlloc(pImgCfg, "BootSector", ppszFreeMe);
5019	if (RT_SUCCESS(rc))
5020	{
5021	ssize_t cbBootSector = RTBase64DecodedSize(*ppszFreeMe, NULL);
5022	if (cbBootSector < 0)
5023	return vdIfError(pImage->pIfError, VERR_INVALID_BASE64_ENCODING, RT_SRC_POS,
5024	N_("VMDK: Image path: '%s'. BASE64 decoding failed on the custom bootsector for '%s'"),
5025	pImage->pszFilename, *ppszRawDrive);
5026	if (cbBootSector == 0)
5027	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5028	N_("VMDK: Image path: '%s'. Custom bootsector for '%s' is zero bytes big"),
5029	pImage->pszFilename, *ppszRawDrive);
5030	if (cbBootSector > _4M) /* this is just a preliminary max */
5031	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5032	N_("VMDK: Image path: '%s'. Custom bootsector for '%s' is way too big: %zu bytes, max 4MB"),
5033	pImage->pszFilename, *ppszRawDrive, cbBootSector);
5034
5035	/* Refuse the boot sector if whole-drive. This used to be done quietly,
5036	however, bird disagrees and thinks the user should be told that what
5037	he/she/it tries to do isn't possible. There should be less head
5038	scratching this way when the guest doesn't do the expected thing. */
5039	if (!*pfPartitions)
5040	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5041	N_("VMDK: Image path: '%s'. Custom bootsector for '%s' is not supported for whole-drive configurations, only when selecting partitions"),
5042	pImage->pszFilename, *ppszRawDrive);
5043
5044	*pcbBootSector = (size_t)cbBootSector;
5045	*ppvBootSector = RTMemAlloc((size_t)cbBootSector);
5046	if (!*ppvBootSector)
5047	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS,
5048	N_("VMDK: Image path: '%s'. Failed to allocate %zd bytes for the custom bootsector for '%s'"),
5049	pImage->pszFilename, cbBootSector, *ppszRawDrive);
5050
5051	rc = RTBase64Decode(ppszFreeMe, ppvBootSector, cbBootSector, NULL /pcbActual/, NULL /ppszEnd/);
5052	if (RT_FAILURE(rc))
5053	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS,
5054	N_("VMDK: Image path: '%s'. Base64 decoding of the custom boot sector for '%s' failed (%Rrc)"),
5055	pImage->pszFilename, *ppszRawDrive, rc);
5056
5057	RTStrFree(*ppszFreeMe);
5058	*ppszFreeMe = NULL;
5059	}
5060	else if (rc != VERR_CFGM_VALUE_NOT_FOUND)
5061	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5062	N_("VMDK: Image path: '%s'. Getting 'BootSector' configuration failed (%Rrc)"), pImage->pszFilename, rc);
5063
5064	/*
5065	* Relative=0/1
5066	*/
5067	*pfRelative = false;
5068	rc = VDCFGQueryBool(pImgCfg, "Relative", pfRelative);
5069	if (RT_SUCCESS(rc))
5070	{
5071	if (!pfPartitions && pfRelative != false)
5072	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5073	N_("VMDK: Image path: '%s'. The 'Relative' option is not supported for whole-drive configurations, only when selecting partitions"),
5074	pImage->pszFilename);
5075	#if !defined(RT_OS_DARWIN) && !defined(RT_OS_LINUX) && !defined(RT_OS_FREEBSD) && !defined(RT_OS_WINDOWS) && !defined(RT_OS_SOLARIS) /* PORTME */
5076	if (*pfRelative == true)
5077	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5078	N_("VMDK: Image path: '%s'. The 'Relative' option is not supported on this host OS"),
5079	pImage->pszFilename);
5080	#endif
5081	}
5082	else if (rc != VERR_CFGM_VALUE_NOT_FOUND)
5083	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5084	N_("VMDK: Image path: '%s'. Getting 'Relative' configuration failed (%Rrc)"), pImage->pszFilename, rc);
5085	else
5086	#ifdef RT_OS_DARWIN /* different default on macOS, see ticketref:1461 (comment 20). */
5087	*pfRelative = true;
5088	#else
5089	*pfRelative = false;
5090	#endif
5091
5092	return VINF_SUCCESS;
5093	}
5094
5095	/**
5096	* Creates a raw drive (nee disk) descriptor.
5097	*
5098	* This was originally done in VBoxInternalManage.cpp, but was copied (not move)
5099	* here much later. That's one of the reasons why we produce a descriptor just
5100	* like it does, rather than mixing directly into the vmdkCreateRawImage code.
5101	*
5102	* @returns VBox status code.
5103	* @param pImage The image.
5104	* @param ppRaw Where to return the raw drive descriptor. Caller must
5105	* free it using vmdkRawDescFree regardless of the status
5106	* code.
5107	* @internal
5108	*/
5109	static int vmdkMakeRawDescriptor(PVMDKIMAGE pImage, PVDISKRAW *ppRaw)
5110	{
5111	/* Make sure it's NULL. */
5112	*ppRaw = NULL;
5113
5114	/*
5115	* Read the configuration.
5116	*/
5117	char *pszRawDrive = NULL;
5118	uint32_t fPartitions = 0; /* zero if whole-drive */
5119	uint32_t fPartitionsReadOnly = 0; /* (subset of fPartitions) */
5120	void *pvBootSector = NULL;
5121	size_t cbBootSector = 0;
5122	bool fRelative = false;
5123	char pszFreeMe = NULL; / lazy bird cleanup. */
5124	int rc = vmdkRawDescParseConfig(pImage, &pszRawDrive, &fPartitions, &fPartitionsReadOnly,
5125	&pvBootSector, &cbBootSector, &fRelative, &pszFreeMe);
5126	RTStrFree(pszFreeMe);
5127	if (RT_SUCCESS(rc))
5128	{
5129	/*
5130	* Open the device, getting the sector size and drive size.
5131	*/
5132	uint64_t cbSize = 0;
5133	uint32_t cbSector = 0;
5134	RTFILE hRawDrive = NIL_RTFILE;
5135	rc = vmkdRawDescOpenDevice(pImage, pszRawDrive, &hRawDrive, &cbSize, &cbSector);
5136	if (RT_SUCCESS(rc))
5137	{
5138	pImage->cbSize = cbSize;
5139	/*
5140	* Create the raw-drive descriptor
5141	*/
5142	PVDISKRAW pRawDesc = (PVDISKRAW)RTMemAllocZ(sizeof(*pRawDesc));
5143	if (pRawDesc)
5144	{
5145	pRawDesc->szSignature[0] = 'R';
5146	pRawDesc->szSignature[1] = 'A';
5147	pRawDesc->szSignature[2] = 'W';
5148	//pRawDesc->szSignature[3] = '\0';
5149	if (!fPartitions)
5150	{
5151	/*
5152	* It's simple for when doing the whole drive.
5153	*/
5154	pRawDesc->uFlags = VDISKRAW_DISK;
5155	rc = RTStrDupEx(&pRawDesc->pszRawDisk, pszRawDrive);
5156	}
5157	else
5158	{
5159	/*
5160	* In selected partitions mode we've got a lot more work ahead of us.
5161	*/
5162	pRawDesc->uFlags = VDISKRAW_NORMAL;
5163	//pRawDesc->pszRawDisk = NULL;
5164	//pRawDesc->cPartDescs = 0;
5165	//pRawDesc->pPartDescs = NULL;
5166
5167	/* We need to parse the partition map to complete the descriptor: */
5168	RTDVM hVolMgr = NIL_RTDVM;
5169	rc = vmdkRawDescOpenVolMgr(pImage, hRawDrive, pszRawDrive, cbSector, &hVolMgr);
5170	if (RT_SUCCESS(rc))
5171	{
5172	RTDVMFORMATTYPE enmFormatType = RTDvmMapGetFormatType(hVolMgr);
5173	if ( enmFormatType == RTDVMFORMATTYPE_MBR
5174	\|\| enmFormatType == RTDVMFORMATTYPE_GPT)
5175	{
5176	pRawDesc->enmPartitioningType = enmFormatType == RTDVMFORMATTYPE_MBR
5177	? VDISKPARTTYPE_MBR : VDISKPARTTYPE_GPT;
5178
5179	/* Add copies of the partition tables: */
5180	rc = vmdkRawDescDoCopyPartitionTables(pImage, hVolMgr, pRawDesc, pszRawDrive, hRawDrive,
5181	pvBootSector, cbBootSector);
5182	if (RT_SUCCESS(rc))
5183	{
5184	/* Add descriptors for the partitions/volumes, indicating which
5185	should be accessible and how to access them: */
5186	RTDVMVOLUME hVolRelease = NIL_RTDVMVOLUME;
5187	rc = vmdkRawDescDoPartitions(pImage, hVolMgr, pRawDesc, hRawDrive, pszRawDrive, cbSector,
5188	fPartitions, fPartitionsReadOnly, fRelative, &hVolRelease);
5189	RTDvmVolumeRelease(hVolRelease);
5190
5191	/* Finally, sort the partition and check consistency (overlaps, etc): */
5192	if (RT_SUCCESS(rc))
5193	rc = vmdkRawDescPostProcessPartitions(pImage, pRawDesc, cbSize);
5194	}
5195	}
5196	else
5197	rc = vdIfError(pImage->pIfError, VERR_NOT_SUPPORTED, RT_SRC_POS,
5198	N_("VMDK: Image path: '%s'. Unsupported partitioning for the disk '%s': %s"),
5199	pImage->pszFilename, pszRawDrive, RTDvmMapGetFormatType(hVolMgr));
5200	RTDvmRelease(hVolMgr);
5201	}
5202	}
5203	if (RT_SUCCESS(rc))
5204	{
5205	/*
5206	* We succeeded.
5207	*/
5208	*ppRaw = pRawDesc;
5209	Log(("vmdkMakeRawDescriptor: fFlags=%#x enmPartitioningType=%d cPartDescs=%u pszRawDisk=%s\n",
5210	pRawDesc->uFlags, pRawDesc->enmPartitioningType, pRawDesc->cPartDescs, pRawDesc->pszRawDisk));
5211	if (pRawDesc->cPartDescs)
5212	{
5213	Log(("# VMDK offset Length Device offset PartDataPtr Device\n"));
5214	for (uint32_t i = 0; i < pRawDesc->cPartDescs; i++)
5215	Log(("%2u %14RU64 %14RU64 %14RU64 %#18p %s\n", i, pRawDesc->pPartDescs[i].offStartInVDisk,
5216	pRawDesc->pPartDescs[i].cbData, pRawDesc->pPartDescs[i].offStartInDevice,
5217	pRawDesc->pPartDescs[i].pvPartitionData, pRawDesc->pPartDescs[i].pszRawDevice));
5218	}
5219	}
5220	else
5221	vmdkRawDescFree(pRawDesc);
5222	}
5223	else
5224	rc = vdIfError(pImage->pIfError, VERR_NOT_SUPPORTED, RT_SRC_POS,
5225	N_("VMDK: Image path: '%s'. Failed to allocate %u bytes for the raw drive descriptor"),
5226	pImage->pszFilename, sizeof(*pRawDesc));
5227	RTFileClose(hRawDrive);
5228	}
5229	}
5230	RTStrFree(pszRawDrive);
5231	RTMemFree(pvBootSector);
5232	return rc;
5233	}
5234
5235	/**
5236	* Internal: create VMDK images for raw disk/partition access.
5237	*/
5238	static int vmdkCreateRawImage(PVMDKIMAGE pImage, const PVDISKRAW pRaw,
5239	uint64_t cbSize)
5240	{
5241	int rc = VINF_SUCCESS;
5242	PVMDKEXTENT pExtent;
5243
5244	if (pRaw->uFlags & VDISKRAW_DISK)
5245	{
5246	/* Full raw disk access. This requires setting up a descriptor
5247	* file and open the (flat) raw disk. */
5248	rc = vmdkCreateExtents(pImage, 1);
5249	if (RT_FAILURE(rc))
5250	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
5251	pExtent = &pImage->pExtents[0];
5252	/* Create raw disk descriptor file. */
5253	rc = vmdkFileOpen(pImage, &pImage->pFile, NULL, pImage->pszFilename,
5254	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5255	true /* fCreate */));
5256	if (RT_FAILURE(rc))
5257	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pImage->pszFilename);
5258
5259	/* Set up basename for extent description. Cannot use StrDup. */
5260	size_t cbBasename = strlen(pRaw->pszRawDisk) + 1;
5261	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
5262	if (!pszBasename)
5263	return VERR_NO_MEMORY;
5264	memcpy(pszBasename, pRaw->pszRawDisk, cbBasename);
5265	pExtent->pszBasename = pszBasename;
5266	/* For raw disks the full name is identical to the base name. */
5267	pExtent->pszFullname = RTStrDup(pszBasename);
5268	if (!pExtent->pszFullname)
5269	return VERR_NO_MEMORY;
5270	pExtent->enmType = VMDKETYPE_FLAT;
5271	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
5272	pExtent->uSectorOffset = 0;
5273	pExtent->enmAccess = (pRaw->uFlags & VDISKRAW_READONLY) ? VMDKACCESS_READONLY : VMDKACCESS_READWRITE;
5274	pExtent->fMetaDirty = false;
5275
5276	/* Open flat image, the raw disk. */
5277	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5278	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
5279	false /* fCreate */));
5280	if (RT_FAILURE(rc))
5281	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not open raw disk file '%s'"), pExtent->pszFullname);
5282	}
5283	else
5284	{
5285	/* Raw partition access. This requires setting up a descriptor
5286	* file, write the partition information to a flat extent and
5287	* open all the (flat) raw disk partitions. */
5288
5289	/* First pass over the partition data areas to determine how many
5290	* extents we need. One data area can require up to 2 extents, as
5291	* it might be necessary to skip over unpartitioned space. */
5292	unsigned cExtents = 0;
5293	uint64_t uStart = 0;
5294	for (unsigned i = 0; i < pRaw->cPartDescs; i++)
5295	{
5296	PVDISKRAWPARTDESC pPart = &pRaw->pPartDescs[i];
5297	if (uStart > pPart->offStartInVDisk)
5298	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5299	N_("VMDK: incorrect partition data area ordering set up by the caller in '%s'"), pImage->pszFilename);
5300
5301	if (uStart < pPart->offStartInVDisk)
5302	cExtents++;
5303	uStart = pPart->offStartInVDisk + pPart->cbData;
5304	cExtents++;
5305	}
5306	/* Another extent for filling up the rest of the image. */
5307	if (uStart != cbSize)
5308	cExtents++;
5309
5310	rc = vmdkCreateExtents(pImage, cExtents);
5311	if (RT_FAILURE(rc))
5312	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
5313
5314	/* Create raw partition descriptor file. */
5315	rc = vmdkFileOpen(pImage, &pImage->pFile, NULL, pImage->pszFilename,
5316	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5317	true /* fCreate */));
5318	if (RT_FAILURE(rc))
5319	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pImage->pszFilename);
5320
5321	/* Create base filename for the partition table extent. */
5322	/** @todo remove fixed buffer without creating memory leaks. */
5323	char pszPartition[1024];
5324	const char *pszBase = RTPathFilename(pImage->pszFilename);
5325	const char *pszSuff = RTPathSuffix(pszBase);
5326	if (pszSuff == NULL)
5327	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: invalid filename '%s'"), pImage->pszFilename);
5328	char *pszBaseBase = RTStrDup(pszBase);
5329	if (!pszBaseBase)
5330	return VERR_NO_MEMORY;
5331	RTPathStripSuffix(pszBaseBase);
5332	RTStrPrintf(pszPartition, sizeof(pszPartition), "%s-pt%s",
5333	pszBaseBase, pszSuff);
5334	RTStrFree(pszBaseBase);
5335
5336	/* Second pass over the partitions, now define all extents. */
5337	uint64_t uPartOffset = 0;
5338	cExtents = 0;
5339	uStart = 0;
5340	for (unsigned i = 0; i < pRaw->cPartDescs; i++)
5341	{
5342	PVDISKRAWPARTDESC pPart = &pRaw->pPartDescs[i];
5343	pExtent = &pImage->pExtents[cExtents++];
5344
5345	if (uStart < pPart->offStartInVDisk)
5346	{
5347	pExtent->pszBasename = NULL;
5348	pExtent->pszFullname = NULL;
5349	pExtent->enmType = VMDKETYPE_ZERO;
5350	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->offStartInVDisk - uStart);
5351	pExtent->uSectorOffset = 0;
5352	pExtent->enmAccess = VMDKACCESS_READWRITE;
5353	pExtent->fMetaDirty = false;
5354	/* go to next extent */
5355	pExtent = &pImage->pExtents[cExtents++];
5356	}
5357	uStart = pPart->offStartInVDisk + pPart->cbData;
5358
5359	if (pPart->pvPartitionData)
5360	{
5361	/* Set up basename for extent description. Can't use StrDup. */
5362	size_t cbBasename = strlen(pszPartition) + 1;
5363	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
5364	if (!pszBasename)
5365	return VERR_NO_MEMORY;
5366	memcpy(pszBasename, pszPartition, cbBasename);
5367	pExtent->pszBasename = pszBasename;
5368
5369	/* Set up full name for partition extent. */
5370	char *pszDirname = RTStrDup(pImage->pszFilename);
5371	if (!pszDirname)
5372	return VERR_NO_STR_MEMORY;
5373	RTPathStripFilename(pszDirname);
5374	char *pszFullname = RTPathJoinA(pszDirname, pExtent->pszBasename);
5375	RTStrFree(pszDirname);
5376	if (!pszFullname)
5377	return VERR_NO_STR_MEMORY;
5378	pExtent->pszFullname = pszFullname;
5379	pExtent->enmType = VMDKETYPE_FLAT;
5380	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
5381	pExtent->uSectorOffset = uPartOffset;
5382	pExtent->enmAccess = VMDKACCESS_READWRITE;
5383	pExtent->fMetaDirty = false;
5384
5385	/* Create partition table flat image. */
5386	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5387	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
5388	true /* fCreate */));
5389	if (RT_FAILURE(rc))
5390	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new partition data file '%s'"), pExtent->pszFullname);
5391	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
5392	VMDK_SECTOR2BYTE(uPartOffset),
5393	pPart->pvPartitionData,
5394	pPart->cbData);
5395	if (RT_FAILURE(rc))
5396	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not write partition data to '%s'"), pExtent->pszFullname);
5397	uPartOffset += VMDK_BYTE2SECTOR(pPart->cbData);
5398	}
5399	else
5400	{
5401	if (pPart->pszRawDevice)
5402	{
5403	/* Set up basename for extent descr. Can't use StrDup. */
5404	size_t cbBasename = strlen(pPart->pszRawDevice) + 1;
5405	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
5406	if (!pszBasename)
5407	return VERR_NO_MEMORY;
5408	memcpy(pszBasename, pPart->pszRawDevice, cbBasename);
5409	pExtent->pszBasename = pszBasename;
5410	/* For raw disks full name is identical to base name. */
5411	pExtent->pszFullname = RTStrDup(pszBasename);
5412	if (!pExtent->pszFullname)
5413	return VERR_NO_MEMORY;
5414	pExtent->enmType = VMDKETYPE_FLAT;
5415	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
5416	pExtent->uSectorOffset = VMDK_BYTE2SECTOR(pPart->offStartInDevice);
5417	pExtent->enmAccess = (pPart->uFlags & VDISKRAW_READONLY) ? VMDKACCESS_READONLY : VMDKACCESS_READWRITE;
5418	pExtent->fMetaDirty = false;
5419
5420	/* Open flat image, the raw partition. */
5421	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5422	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
5423	false /* fCreate */));
5424	if (RT_FAILURE(rc))
5425	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not open raw partition file '%s'"), pExtent->pszFullname);
5426	}
5427	else
5428	{
5429	pExtent->pszBasename = NULL;
5430	pExtent->pszFullname = NULL;
5431	pExtent->enmType = VMDKETYPE_ZERO;
5432	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
5433	pExtent->uSectorOffset = 0;
5434	pExtent->enmAccess = VMDKACCESS_READWRITE;
5435	pExtent->fMetaDirty = false;
5436	}
5437	}
5438	}
5439	/* Another extent for filling up the rest of the image. */
5440	if (uStart != cbSize)
5441	{
5442	pExtent = &pImage->pExtents[cExtents++];
5443	pExtent->pszBasename = NULL;
5444	pExtent->pszFullname = NULL;
5445	pExtent->enmType = VMDKETYPE_ZERO;
5446	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize - uStart);
5447	pExtent->uSectorOffset = 0;
5448	pExtent->enmAccess = VMDKACCESS_READWRITE;
5449	pExtent->fMetaDirty = false;
5450	}
5451	}
5452
5453	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
5454	(pRaw->uFlags & VDISKRAW_DISK) ?
5455	"fullDevice" : "partitionedDevice");
5456	if (RT_FAILURE(rc))
5457	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
5458	return rc;
5459	}
5460
5461	/**
5462	* Internal: create a regular (i.e. file-backed) VMDK image.
5463	*/
5464	static int vmdkCreateRegularImage(PVMDKIMAGE pImage, uint64_t cbSize,
5465	unsigned uImageFlags, PVDINTERFACEPROGRESS pIfProgress,
5466	unsigned uPercentStart, unsigned uPercentSpan)
5467	{
5468	int rc = VINF_SUCCESS;
5469	unsigned cExtents = 1;
5470	uint64_t cbOffset = 0;
5471	uint64_t cbRemaining = cbSize;
5472
5473	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
5474	{
5475	cExtents = cbSize / VMDK_2G_SPLIT_SIZE;
5476	/* Do proper extent computation: need one smaller extent if the total
5477	* size isn't evenly divisible by the split size. */
5478	if (cbSize % VMDK_2G_SPLIT_SIZE)
5479	cExtents++;
5480	}
5481	rc = vmdkCreateExtents(pImage, cExtents);
5482	if (RT_FAILURE(rc))
5483	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
5484
5485	/* Basename strings needed for constructing the extent names. */
5486	char *pszBasenameSubstr = RTPathFilename(pImage->pszFilename);
5487	AssertPtr(pszBasenameSubstr);
5488	size_t cbBasenameSubstr = strlen(pszBasenameSubstr) + 1;
5489
5490	/* Create separate descriptor file if necessary. */
5491	if (cExtents != 1 \|\| (uImageFlags & VD_IMAGE_FLAGS_FIXED))
5492	{
5493	rc = vmdkFileOpen(pImage, &pImage->pFile, NULL, pImage->pszFilename,
5494	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5495	true /* fCreate */));
5496	if (RT_FAILURE(rc))
5497	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new sparse descriptor file '%s'"), pImage->pszFilename);
5498	}
5499	else
5500	pImage->pFile = NULL;
5501
5502	/* Set up all extents. */
5503	for (unsigned i = 0; i < cExtents; i++)
5504	{
5505	PVMDKEXTENT pExtent = &pImage->pExtents[i];
5506	uint64_t cbExtent = cbRemaining;
5507
5508	/* Set up fullname/basename for extent description. Cannot use StrDup
5509	* for basename, as it is not guaranteed that the memory can be freed
5510	* with RTMemTmpFree, which must be used as in other code paths
5511	* StrDup is not usable. */
5512	if (cExtents == 1 && !(uImageFlags & VD_IMAGE_FLAGS_FIXED))
5513	{
5514	char pszBasename = (char )RTMemTmpAlloc(cbBasenameSubstr);
5515	if (!pszBasename)
5516	return VERR_NO_MEMORY;
5517	memcpy(pszBasename, pszBasenameSubstr, cbBasenameSubstr);
5518	pExtent->pszBasename = pszBasename;
5519	}
5520	else
5521	{
5522	char *pszBasenameSuff = RTPathSuffix(pszBasenameSubstr);
5523	char *pszBasenameBase = RTStrDup(pszBasenameSubstr);
5524	RTPathStripSuffix(pszBasenameBase);
5525	char *pszTmp;
5526	size_t cbTmp;
5527	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
5528	{
5529	if (cExtents == 1)
5530	RTStrAPrintf(&pszTmp, "%s-flat%s", pszBasenameBase,
5531	pszBasenameSuff);
5532	else
5533	RTStrAPrintf(&pszTmp, "%s-f%03d%s", pszBasenameBase,
5534	i+1, pszBasenameSuff);
5535	}
5536	else
5537	RTStrAPrintf(&pszTmp, "%s-s%03d%s", pszBasenameBase, i+1,
5538	pszBasenameSuff);
5539	RTStrFree(pszBasenameBase);
5540	if (!pszTmp)
5541	return VERR_NO_STR_MEMORY;
5542	cbTmp = strlen(pszTmp) + 1;
5543	char pszBasename = (char )RTMemTmpAlloc(cbTmp);
5544	if (!pszBasename)
5545	{
5546	RTStrFree(pszTmp);
5547	return VERR_NO_MEMORY;
5548	}
5549	memcpy(pszBasename, pszTmp, cbTmp);
5550	RTStrFree(pszTmp);
5551	pExtent->pszBasename = pszBasename;
5552	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
5553	cbExtent = RT_MIN(cbRemaining, VMDK_2G_SPLIT_SIZE);
5554	}
5555	char *pszBasedirectory = RTStrDup(pImage->pszFilename);
5556	if (!pszBasedirectory)
5557	return VERR_NO_STR_MEMORY;
5558	RTPathStripFilename(pszBasedirectory);
5559	char *pszFullname = RTPathJoinA(pszBasedirectory, pExtent->pszBasename);
5560	RTStrFree(pszBasedirectory);
5561	if (!pszFullname)
5562	return VERR_NO_STR_MEMORY;
5563	pExtent->pszFullname = pszFullname;
5564
5565	/* Create file for extent. */
5566	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5567	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5568	true /* fCreate */));
5569	if (RT_FAILURE(rc))
5570	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pExtent->pszFullname);
5571	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
5572	{
5573	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage, cbExtent,
5574	0 /* fFlags */, pIfProgress,
5575	uPercentStart + cbOffset * uPercentSpan / cbSize,
5576	cbExtent * uPercentSpan / cbSize);
5577	if (RT_FAILURE(rc))
5578	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
5579	}
5580
5581	/* Place descriptor file information (where integrated). */
5582	if (cExtents == 1 && !(uImageFlags & VD_IMAGE_FLAGS_FIXED))
5583	{
5584	pExtent->uDescriptorSector = 1;
5585	pExtent->cDescriptorSectors = VMDK_BYTE2SECTOR(pImage->cbDescAlloc);
5586	/* The descriptor is part of the (only) extent. */
5587	pExtent->pDescData = pImage->pDescData;
5588	pImage->pDescData = NULL;
5589	}
5590
5591	if (!(uImageFlags & VD_IMAGE_FLAGS_FIXED))
5592	{
5593	uint64_t cSectorsPerGDE, cSectorsPerGD;
5594	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE;
5595	pExtent->cSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64(cbExtent, _64K));
5596	pExtent->cSectorsPerGrain = VMDK_BYTE2SECTOR(_64K);
5597	pExtent->cGTEntries = 512;
5598	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
5599	pExtent->cSectorsPerGDE = cSectorsPerGDE;
5600	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
5601	cSectorsPerGD = (pExtent->cGDEntries + (512 / sizeof(uint32_t) - 1)) / (512 / sizeof(uint32_t));
5602	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5603	{
5604	/* The spec says version is 1 for all VMDKs, but the vast
5605	* majority of streamOptimized VMDKs actually contain
5606	* version 3 - so go with the majority. Both are accepted. */
5607	pExtent->uVersion = 3;
5608	pExtent->uCompression = VMDK_COMPRESSION_DEFLATE;
5609	}
5610	}
5611	else
5612	{
5613	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
5614	pExtent->enmType = VMDKETYPE_VMFS;
5615	else
5616	pExtent->enmType = VMDKETYPE_FLAT;
5617	}
5618
5619	pExtent->enmAccess = VMDKACCESS_READWRITE;
5620	pExtent->fUncleanShutdown = true;
5621	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbExtent);
5622	pExtent->uSectorOffset = 0;
5623	pExtent->fMetaDirty = true;
5624
5625	if (!(uImageFlags & VD_IMAGE_FLAGS_FIXED))
5626	{
5627	/* fPreAlloc should never be false because VMware can't use such images. */
5628	rc = vmdkCreateGrainDirectory(pImage, pExtent,
5629	RT_MAX( pExtent->uDescriptorSector
5630	+ pExtent->cDescriptorSectors,
5631	1),
5632	true /* fPreAlloc */);
5633	if (RT_FAILURE(rc))
5634	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new grain directory in '%s'"), pExtent->pszFullname);
5635	}
5636
5637	cbOffset += cbExtent;
5638
5639	if (RT_SUCCESS(rc))
5640	vdIfProgress(pIfProgress, uPercentStart + cbOffset * uPercentSpan / cbSize);
5641
5642	cbRemaining -= cbExtent;
5643	}
5644
5645	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
5646	{
5647	/* VirtualBox doesn't care, but VMWare ESX freaks out if the wrong
5648	* controller type is set in an image. */
5649	rc = vmdkDescDDBSetStr(pImage, &pImage->Descriptor, "ddb.adapterType", "lsilogic");
5650	if (RT_FAILURE(rc))
5651	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set controller type to lsilogic in '%s'"), pImage->pszFilename);
5652	}
5653
5654	const char *pszDescType = NULL;
5655	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
5656	{
5657	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
5658	pszDescType = "vmfs";
5659	else
5660	pszDescType = (cExtents == 1)
5661	? "monolithicFlat" : "twoGbMaxExtentFlat";
5662	}
5663	else
5664	{
5665	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5666	pszDescType = "streamOptimized";
5667	else
5668	{
5669	pszDescType = (cExtents == 1)
5670	? "monolithicSparse" : "twoGbMaxExtentSparse";
5671	}
5672	}
5673	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
5674	pszDescType);
5675	if (RT_FAILURE(rc))
5676	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
5677	return rc;
5678	}
5679
5680	/**
5681	* Internal: Create a real stream optimized VMDK using only linear writes.
5682	*/
5683	static int vmdkCreateStreamImage(PVMDKIMAGE pImage, uint64_t cbSize)
5684	{
5685	int rc = vmdkCreateExtents(pImage, 1);
5686	if (RT_FAILURE(rc))
5687	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
5688
5689	/* Basename strings needed for constructing the extent names. */
5690	const char *pszBasenameSubstr = RTPathFilename(pImage->pszFilename);
5691	AssertPtr(pszBasenameSubstr);
5692	size_t cbBasenameSubstr = strlen(pszBasenameSubstr) + 1;
5693
5694	/* No separate descriptor file. */
5695	pImage->pFile = NULL;
5696
5697	/* Set up all extents. */
5698	PVMDKEXTENT pExtent = &pImage->pExtents[0];
5699
5700	/* Set up fullname/basename for extent description. Cannot use StrDup
5701	* for basename, as it is not guaranteed that the memory can be freed
5702	* with RTMemTmpFree, which must be used as in other code paths
5703	* StrDup is not usable. */
5704	char pszBasename = (char )RTMemTmpAlloc(cbBasenameSubstr);
5705	if (!pszBasename)
5706	return VERR_NO_MEMORY;
5707	memcpy(pszBasename, pszBasenameSubstr, cbBasenameSubstr);
5708	pExtent->pszBasename = pszBasename;
5709
5710	char *pszBasedirectory = RTStrDup(pImage->pszFilename);
5711	RTPathStripFilename(pszBasedirectory);
5712	char *pszFullname = RTPathJoinA(pszBasedirectory, pExtent->pszBasename);
5713	RTStrFree(pszBasedirectory);
5714	if (!pszFullname)
5715	return VERR_NO_STR_MEMORY;
5716	pExtent->pszFullname = pszFullname;
5717
5718	/* Create file for extent. Make it write only, no reading allowed. */
5719	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5720	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5721	true /* fCreate */)
5722	& ~RTFILE_O_READ);
5723	if (RT_FAILURE(rc))
5724	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pExtent->pszFullname);
5725
5726	/* Place descriptor file information. */
5727	pExtent->uDescriptorSector = 1;
5728	pExtent->cDescriptorSectors = VMDK_BYTE2SECTOR(pImage->cbDescAlloc);
5729	/* The descriptor is part of the (only) extent. */
5730	pExtent->pDescData = pImage->pDescData;
5731	pImage->pDescData = NULL;
5732
5733	uint64_t cSectorsPerGDE, cSectorsPerGD;
5734	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE;
5735	pExtent->cSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64(cbSize, _64K));
5736	pExtent->cSectorsPerGrain = VMDK_BYTE2SECTOR(_64K);
5737	pExtent->cGTEntries = 512;
5738	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
5739	pExtent->cSectorsPerGDE = cSectorsPerGDE;
5740	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
5741	cSectorsPerGD = (pExtent->cGDEntries + (512 / sizeof(uint32_t) - 1)) / (512 / sizeof(uint32_t));
5742
5743	/* The spec says version is 1 for all VMDKs, but the vast
5744	* majority of streamOptimized VMDKs actually contain
5745	* version 3 - so go with the majority. Both are accepted. */
5746	pExtent->uVersion = 3;
5747	pExtent->uCompression = VMDK_COMPRESSION_DEFLATE;
5748	pExtent->fFooter = true;
5749
5750	pExtent->enmAccess = VMDKACCESS_READONLY;
5751	pExtent->fUncleanShutdown = false;
5752	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
5753	pExtent->uSectorOffset = 0;
5754	pExtent->fMetaDirty = true;
5755
5756	/* Create grain directory, without preallocating it straight away. It will
5757	* be constructed on the fly when writing out the data and written when
5758	* closing the image. The end effect is that the full grain directory is
5759	* allocated, which is a requirement of the VMDK specs. */
5760	rc = vmdkCreateGrainDirectory(pImage, pExtent, VMDK_GD_AT_END,
5761	false /* fPreAlloc */);
5762	if (RT_FAILURE(rc))
5763	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new grain directory in '%s'"), pExtent->pszFullname);
5764
5765	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
5766	"streamOptimized");
5767	if (RT_FAILURE(rc))
5768	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
5769
5770	return rc;
5771	}
5772
5773	/**
5774	* Initializes the UUID fields in the DDB.
5775	*
5776	* @returns VBox status code.
5777	* @param pImage The VMDK image instance.
5778	*/
5779	static int vmdkCreateImageDdbUuidsInit(PVMDKIMAGE pImage)
5780	{
5781	int rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_IMAGE_UUID, &pImage->ImageUuid);
5782	if (RT_SUCCESS(rc))
5783	{
5784	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_PARENT_UUID, &pImage->ParentUuid);
5785	if (RT_SUCCESS(rc))
5786	{
5787	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_MODIFICATION_UUID,
5788	&pImage->ModificationUuid);
5789	if (RT_SUCCESS(rc))
5790	{
5791	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_PARENT_MODIFICATION_UUID,
5792	&pImage->ParentModificationUuid);
5793	if (RT_FAILURE(rc))
5794	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5795	N_("VMDK: error storing parent modification UUID in new descriptor in '%s'"), pImage->pszFilename);
5796	}
5797	else
5798	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5799	N_("VMDK: error storing modification UUID in new descriptor in '%s'"), pImage->pszFilename);
5800	}
5801	else
5802	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5803	N_("VMDK: error storing parent image UUID in new descriptor in '%s'"), pImage->pszFilename);
5804	}
5805	else
5806	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5807	N_("VMDK: error storing image UUID in new descriptor in '%s'"), pImage->pszFilename);
5808
5809	return rc;
5810	}
5811
5812	/**
5813	* Internal: The actual code for creating any VMDK variant currently in
5814	* existence on hosted environments.
5815	*/
5816	static int vmdkCreateImage(PVMDKIMAGE pImage, uint64_t cbSize,
5817	unsigned uImageFlags, const char *pszComment,
5818	PCVDGEOMETRY pPCHSGeometry,
5819	PCVDGEOMETRY pLCHSGeometry, PCRTUUID pUuid,
5820	PVDINTERFACEPROGRESS pIfProgress,
5821	unsigned uPercentStart, unsigned uPercentSpan)
5822	{
5823	pImage->uImageFlags = uImageFlags;
5824
5825	pImage->pIfError = VDIfErrorGet(pImage->pVDIfsDisk);
5826	pImage->pIfIo = VDIfIoIntGet(pImage->pVDIfsImage);
5827	AssertPtrReturn(pImage->pIfIo, VERR_INVALID_PARAMETER);
5828
5829	int rc = vmdkCreateDescriptor(pImage, pImage->pDescData, pImage->cbDescAlloc,
5830	&pImage->Descriptor);
5831	if (RT_SUCCESS(rc))
5832	{
5833	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
5834	{
5835	/* Raw disk image (includes raw partition). */
5836	PVDISKRAW pRaw = NULL;
5837	rc = vmdkMakeRawDescriptor(pImage, &pRaw);
5838	if (RT_FAILURE(rc))
5839	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create raw descriptor for '%s'"),
5840	pImage->pszFilename);
5841	if (!cbSize)
5842	cbSize = pImage->cbSize;
5843
5844	rc = vmdkCreateRawImage(pImage, pRaw, cbSize);
5845	vmdkRawDescFree(pRaw);
5846	}
5847	else if (uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5848	{
5849	/* Stream optimized sparse image (monolithic). */
5850	rc = vmdkCreateStreamImage(pImage, cbSize);
5851	}
5852	else
5853	{
5854	/* Regular fixed or sparse image (monolithic or split). */
5855	rc = vmdkCreateRegularImage(pImage, cbSize, uImageFlags,
5856	pIfProgress, uPercentStart,
5857	uPercentSpan * 95 / 100);
5858	}
5859
5860	if (RT_SUCCESS(rc))
5861	{
5862	vdIfProgress(pIfProgress, uPercentStart + uPercentSpan * 98 / 100);
5863
5864	pImage->cbSize = cbSize;
5865
5866	for (unsigned i = 0; i < pImage->cExtents; i++)
5867	{
5868	PVMDKEXTENT pExtent = &pImage->pExtents[i];
5869
5870	rc = vmdkDescExtInsert(pImage, &pImage->Descriptor, pExtent->enmAccess,
5871	pExtent->cNominalSectors, pExtent->enmType,
5872	pExtent->pszBasename, pExtent->uSectorOffset);
5873	if (RT_FAILURE(rc))
5874	{
5875	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not insert the extent list into descriptor in '%s'"), pImage->pszFilename);
5876	break;
5877	}
5878	}
5879
5880	if (RT_SUCCESS(rc))
5881	vmdkDescExtRemoveDummy(pImage, &pImage->Descriptor);
5882
5883	pImage->LCHSGeometry = *pLCHSGeometry;
5884	pImage->PCHSGeometry = *pPCHSGeometry;
5885
5886	if (RT_SUCCESS(rc))
5887	{
5888	if ( pPCHSGeometry->cCylinders != 0
5889	&& pPCHSGeometry->cHeads != 0
5890	&& pPCHSGeometry->cSectors != 0)
5891	rc = vmdkDescSetPCHSGeometry(pImage, pPCHSGeometry);
5892	else if (uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
5893	{
5894	VDGEOMETRY RawDiskPCHSGeometry;
5895	RawDiskPCHSGeometry.cCylinders = (uint32_t)RT_MIN(pImage->cbSize / 512 / 16 / 63, 16383);
5896	RawDiskPCHSGeometry.cHeads = 16;
5897	RawDiskPCHSGeometry.cSectors = 63;
5898	rc = vmdkDescSetPCHSGeometry(pImage, &RawDiskPCHSGeometry);
5899	}
5900	}
5901
5902	if ( RT_SUCCESS(rc)
5903	&& pLCHSGeometry->cCylinders != 0
5904	&& pLCHSGeometry->cHeads != 0
5905	&& pLCHSGeometry->cSectors != 0)
5906	rc = vmdkDescSetLCHSGeometry(pImage, pLCHSGeometry);
5907
5908	pImage->ImageUuid = *pUuid;
5909	RTUuidClear(&pImage->ParentUuid);
5910	RTUuidClear(&pImage->ModificationUuid);
5911	RTUuidClear(&pImage->ParentModificationUuid);
5912
5913	if (RT_SUCCESS(rc))
5914	rc = vmdkCreateImageDdbUuidsInit(pImage);
5915
5916	if (RT_SUCCESS(rc))
5917	rc = vmdkAllocateGrainTableCache(pImage);
5918
5919	if (RT_SUCCESS(rc))
5920	{
5921	rc = vmdkSetImageComment(pImage, pszComment);
5922	if (RT_FAILURE(rc))
5923	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot set image comment in '%s'"), pImage->pszFilename);
5924	}
5925
5926	if (RT_SUCCESS(rc))
5927	{
5928	vdIfProgress(pIfProgress, uPercentStart + uPercentSpan * 99 / 100);
5929
5930	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5931	{
5932	/* streamOptimized is a bit special, we cannot trigger the flush
5933	* until all data has been written. So we write the necessary
5934	* information explicitly. */
5935	pImage->pExtents[0].cDescriptorSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64( pImage->Descriptor.aLines[pImage->Descriptor.cLines]
5936	- pImage->Descriptor.aLines[0], 512));
5937	rc = vmdkWriteMetaSparseExtent(pImage, &pImage->pExtents[0], 0, NULL);
5938	if (RT_SUCCESS(rc))
5939	{
5940	rc = vmdkWriteDescriptor(pImage, NULL);
5941	if (RT_FAILURE(rc))
5942	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write VMDK descriptor in '%s'"), pImage->pszFilename);
5943	}
5944	else
5945	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write VMDK header in '%s'"), pImage->pszFilename);
5946	}
5947	else
5948	rc = vmdkFlushImage(pImage, NULL);
5949	}
5950	}
5951	}
5952	else
5953	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new descriptor in '%s'"), pImage->pszFilename);
5954
5955
5956	if (RT_SUCCESS(rc))
5957	{
5958	PVDREGIONDESC pRegion = &pImage->RegionList.aRegions[0];
5959	pImage->RegionList.fFlags = 0;
5960	pImage->RegionList.cRegions = 1;
5961
5962	pRegion->offRegion = 0; /* Disk start. */
5963	pRegion->cbBlock = 512;
5964	pRegion->enmDataForm = VDREGIONDATAFORM_RAW;
5965	pRegion->enmMetadataForm = VDREGIONMETADATAFORM_NONE;
5966	pRegion->cbData = 512;
5967	pRegion->cbMetadata = 0;
5968	pRegion->cRegionBlocksOrBytes = pImage->cbSize;
5969
5970	vdIfProgress(pIfProgress, uPercentStart + uPercentSpan);
5971	}
5972	else
5973	vmdkFreeImage(pImage, rc != VERR_ALREADY_EXISTS, false /fFlush/);
5974	return rc;
5975	}
5976
5977	/**
5978	* Internal: Update image comment.
5979	*/
5980	static int vmdkSetImageComment(PVMDKIMAGE pImage, const char *pszComment)
5981	{
5982	char *pszCommentEncoded = NULL;
5983	if (pszComment)
5984	{
5985	pszCommentEncoded = vmdkEncodeString(pszComment);
5986	if (!pszCommentEncoded)
5987	return VERR_NO_MEMORY;
5988	}
5989
5990	int rc = vmdkDescDDBSetStr(pImage, &pImage->Descriptor,
5991	"ddb.comment", pszCommentEncoded);
5992	if (pszCommentEncoded)
5993	RTStrFree(pszCommentEncoded);
5994	if (RT_FAILURE(rc))
5995	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image comment in descriptor in '%s'"), pImage->pszFilename);
5996	return VINF_SUCCESS;
5997	}
5998
5999	/**
6000	* Internal. Clear the grain table buffer for real stream optimized writing.
6001	*/
6002	static void vmdkStreamClearGT(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
6003	{
6004	uint32_t cCacheLines = RT_ALIGN(pExtent->cGTEntries, VMDK_GT_CACHELINE_SIZE) / VMDK_GT_CACHELINE_SIZE;
6005	for (uint32_t i = 0; i < cCacheLines; i++)
6006	memset(&pImage->pGTCache->aGTCache[i].aGTData[0], '\0',
6007	VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t));
6008	}
6009
6010	/**
6011	* Internal. Flush the grain table buffer for real stream optimized writing.
6012	*/
6013	static int vmdkStreamFlushGT(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
6014	uint32_t uGDEntry)
6015	{
6016	int rc = VINF_SUCCESS;
6017	uint32_t cCacheLines = RT_ALIGN(pExtent->cGTEntries, VMDK_GT_CACHELINE_SIZE) / VMDK_GT_CACHELINE_SIZE;
6018
6019	/* VMware does not write out completely empty grain tables in the case
6020	* of streamOptimized images, which according to my interpretation of
6021	* the VMDK 1.1 spec is bending the rules. Since they do it and we can
6022	* handle it without problems do it the same way and save some bytes. */
6023	bool fAllZero = true;
6024	for (uint32_t i = 0; i < cCacheLines; i++)
6025	{
6026	/* Convert the grain table to little endian in place, as it will not
6027	* be used at all after this function has been called. */
6028	uint32_t *pGTTmp = &pImage->pGTCache->aGTCache[i].aGTData[0];
6029	for (uint32_t j = 0; j < VMDK_GT_CACHELINE_SIZE; j++, pGTTmp++)
6030	if (*pGTTmp)
6031	{
6032	fAllZero = false;
6033	break;
6034	}
6035	if (!fAllZero)
6036	break;
6037	}
6038	if (fAllZero)
6039	return VINF_SUCCESS;
6040
6041	uint64_t uFileOffset = pExtent->uAppendPosition;
6042	if (!uFileOffset)
6043	return VERR_INTERNAL_ERROR;
6044	/* Align to sector, as the previous write could have been any size. */
6045	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
6046
6047	/* Grain table marker. */
6048	uint8_t aMarker[512];
6049	PVMDKMARKER pMarker = (PVMDKMARKER)&aMarker[0];
6050	memset(pMarker, '\0', sizeof(aMarker));
6051	pMarker->uSector = RT_H2LE_U64(VMDK_BYTE2SECTOR((uint64_t)pExtent->cGTEntries * sizeof(uint32_t)));
6052	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_GT);
6053	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
6054	aMarker, sizeof(aMarker));
6055	AssertRC(rc);
6056	uFileOffset += 512;
6057
6058	if (!pExtent->pGD \|\| pExtent->pGD[uGDEntry])
6059	return VERR_INTERNAL_ERROR;
6060
6061	pExtent->pGD[uGDEntry] = VMDK_BYTE2SECTOR(uFileOffset);
6062
6063	for (uint32_t i = 0; i < cCacheLines; i++)
6064	{
6065	/* Convert the grain table to little endian in place, as it will not
6066	* be used at all after this function has been called. */
6067	uint32_t *pGTTmp = &pImage->pGTCache->aGTCache[i].aGTData[0];
6068	for (uint32_t j = 0; j < VMDK_GT_CACHELINE_SIZE; j++, pGTTmp++)
6069	pGTTmp = RT_H2LE_U32(pGTTmp);
6070
6071	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
6072	&pImage->pGTCache->aGTCache[i].aGTData[0],
6073	VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t));
6074	uFileOffset += VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t);
6075	if (RT_FAILURE(rc))
6076	break;
6077	}
6078	Assert(!(uFileOffset % 512));
6079	pExtent->uAppendPosition = RT_ALIGN_64(uFileOffset, 512);
6080	return rc;
6081	}
6082
6083	/**
6084	* Internal. Free all allocated space for representing an image, and optionally
6085	* delete the image from disk.
6086	*/
6087	static int vmdkFreeImage(PVMDKIMAGE pImage, bool fDelete, bool fFlush)
6088	{
6089	int rc = VINF_SUCCESS;
6090
6091	/* Freeing a never allocated image (e.g. because the open failed) is
6092	* not signalled as an error. After all nothing bad happens. */
6093	if (pImage)
6094	{
6095	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
6096	{
6097	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6098	{
6099	/* Check if all extents are clean. */
6100	for (unsigned i = 0; i < pImage->cExtents; i++)
6101	{
6102	Assert(!pImage->pExtents[i].fUncleanShutdown);
6103	}
6104	}
6105	else
6106	{
6107	/* Mark all extents as clean. */
6108	for (unsigned i = 0; i < pImage->cExtents; i++)
6109	{
6110	if ( pImage->pExtents[i].enmType == VMDKETYPE_HOSTED_SPARSE
6111	&& pImage->pExtents[i].fUncleanShutdown)
6112	{
6113	pImage->pExtents[i].fUncleanShutdown = false;
6114	pImage->pExtents[i].fMetaDirty = true;
6115	}
6116
6117	/* From now on it's not safe to append any more data. */
6118	pImage->pExtents[i].uAppendPosition = 0;
6119	}
6120	}
6121	}
6122
6123	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6124	{
6125	/* No need to write any pending data if the file will be deleted
6126	* or if the new file wasn't successfully created. */
6127	if ( !fDelete && pImage->pExtents
6128	&& pImage->pExtents[0].cGTEntries
6129	&& pImage->pExtents[0].uAppendPosition)
6130	{
6131	PVMDKEXTENT pExtent = &pImage->pExtents[0];
6132	uint32_t uLastGDEntry = pExtent->uLastGrainAccess / pExtent->cGTEntries;
6133	rc = vmdkStreamFlushGT(pImage, pExtent, uLastGDEntry);
6134	AssertRC(rc);
6135	vmdkStreamClearGT(pImage, pExtent);
6136	for (uint32_t i = uLastGDEntry + 1; i < pExtent->cGDEntries; i++)
6137	{
6138	rc = vmdkStreamFlushGT(pImage, pExtent, i);
6139	AssertRC(rc);
6140	}
6141
6142	uint64_t uFileOffset = pExtent->uAppendPosition;
6143	if (!uFileOffset)
6144	return VERR_INTERNAL_ERROR;
6145	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
6146
6147	/* From now on it's not safe to append any more data. */
6148	pExtent->uAppendPosition = 0;
6149
6150	/* Grain directory marker. */
6151	uint8_t aMarker[512];
6152	PVMDKMARKER pMarker = (PVMDKMARKER)&aMarker[0];
6153	memset(pMarker, '\0', sizeof(aMarker));
6154	pMarker->uSector = VMDK_BYTE2SECTOR(RT_ALIGN_64(RT_H2LE_U64((uint64_t)pExtent->cGDEntries * sizeof(uint32_t)), 512));
6155	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_GD);
6156	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
6157	aMarker, sizeof(aMarker));
6158	AssertRC(rc);
6159	uFileOffset += 512;
6160
6161	/* Write grain directory in little endian style. The array will
6162	* not be used after this, so convert in place. */
6163	uint32_t *pGDTmp = pExtent->pGD;
6164	for (uint32_t i = 0; i < pExtent->cGDEntries; i++, pGDTmp++)
6165	pGDTmp = RT_H2LE_U32(pGDTmp);
6166	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
6167	uFileOffset, pExtent->pGD,
6168	pExtent->cGDEntries * sizeof(uint32_t));
6169	AssertRC(rc);
6170
6171	pExtent->uSectorGD = VMDK_BYTE2SECTOR(uFileOffset);
6172	pExtent->uSectorRGD = VMDK_BYTE2SECTOR(uFileOffset);
6173	uFileOffset = RT_ALIGN_64( uFileOffset
6174	+ pExtent->cGDEntries * sizeof(uint32_t),
6175	512);
6176
6177	/* Footer marker. */
6178	memset(pMarker, '\0', sizeof(aMarker));
6179	pMarker->uSector = VMDK_BYTE2SECTOR(512);
6180	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_FOOTER);
6181	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
6182	uFileOffset, aMarker, sizeof(aMarker));
6183	AssertRC(rc);
6184
6185	uFileOffset += 512;
6186	rc = vmdkWriteMetaSparseExtent(pImage, pExtent, uFileOffset, NULL);
6187	AssertRC(rc);
6188
6189	uFileOffset += 512;
6190	/* End-of-stream marker. */
6191	memset(pMarker, '\0', sizeof(aMarker));
6192	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
6193	uFileOffset, aMarker, sizeof(aMarker));
6194	AssertRC(rc);
6195	}
6196	}
6197	else if (!fDelete && fFlush)
6198	vmdkFlushImage(pImage, NULL);
6199
6200	if (pImage->pExtents != NULL)
6201	{
6202	for (unsigned i = 0 ; i < pImage->cExtents; i++)
6203	{
6204	int rc2 = vmdkFreeExtentData(pImage, &pImage->pExtents[i], fDelete);
6205	if (RT_SUCCESS(rc))
6206	rc = rc2; /* Propogate any error when closing the file. */
6207	}
6208	RTMemFree(pImage->pExtents);
6209	pImage->pExtents = NULL;
6210	}
6211	pImage->cExtents = 0;
6212	if (pImage->pFile != NULL)
6213	{
6214	int rc2 = vmdkFileClose(pImage, &pImage->pFile, fDelete);
6215	if (RT_SUCCESS(rc))
6216	rc = rc2; /* Propogate any error when closing the file. */
6217	}
6218	int rc2 = vmdkFileCheckAllClose(pImage);
6219	if (RT_SUCCESS(rc))
6220	rc = rc2; /* Propogate any error when closing the file. */
6221
6222	if (pImage->pGTCache)
6223	{
6224	RTMemFree(pImage->pGTCache);
6225	pImage->pGTCache = NULL;
6226	}
6227	if (pImage->pDescData)
6228	{
6229	RTMemFree(pImage->pDescData);
6230	pImage->pDescData = NULL;
6231	}
6232	}
6233
6234	LogFlowFunc(("returns %Rrc\n", rc));
6235	return rc;
6236	}
6237
6238	/**
6239	* Internal. Flush image data (and metadata) to disk.
6240	*/
6241	static int vmdkFlushImage(PVMDKIMAGE pImage, PVDIOCTX pIoCtx)
6242	{
6243	PVMDKEXTENT pExtent;
6244	int rc = VINF_SUCCESS;
6245
6246	/* Update descriptor if changed. */
6247	if (pImage->Descriptor.fDirty)
6248	rc = vmdkWriteDescriptor(pImage, pIoCtx);
6249
6250	if (RT_SUCCESS(rc))
6251	{
6252	for (unsigned i = 0; i < pImage->cExtents; i++)
6253	{
6254	pExtent = &pImage->pExtents[i];
6255	if (pExtent->pFile != NULL && pExtent->fMetaDirty)
6256	{
6257	switch (pExtent->enmType)
6258	{
6259	case VMDKETYPE_HOSTED_SPARSE:
6260	if (!pExtent->fFooter)
6261	rc = vmdkWriteMetaSparseExtent(pImage, pExtent, 0, pIoCtx);
6262	else
6263	{
6264	uint64_t uFileOffset = pExtent->uAppendPosition;
6265	/* Simply skip writing anything if the streamOptimized
6266	* image hasn't been just created. */
6267	if (!uFileOffset)
6268	break;
6269	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
6270	rc = vmdkWriteMetaSparseExtent(pImage, pExtent,
6271	uFileOffset, pIoCtx);
6272	}
6273	break;
6274	case VMDKETYPE_VMFS:
6275	case VMDKETYPE_FLAT:
6276	/* Nothing to do. */
6277	break;
6278	case VMDKETYPE_ZERO:
6279	default:
6280	AssertMsgFailed(("extent with type %d marked as dirty\n",
6281	pExtent->enmType));
6282	break;
6283	}
6284	}
6285
6286	if (RT_FAILURE(rc))
6287	break;
6288
6289	switch (pExtent->enmType)
6290	{
6291	case VMDKETYPE_HOSTED_SPARSE:
6292	case VMDKETYPE_VMFS:
6293	case VMDKETYPE_FLAT:
6294	/** @todo implement proper path absolute check. */
6295	if ( pExtent->pFile != NULL
6296	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
6297	&& !(pExtent->pszBasename[0] == RTPATH_SLASH))
6298	rc = vdIfIoIntFileFlush(pImage->pIfIo, pExtent->pFile->pStorage, pIoCtx,
6299	NULL, NULL);
6300	break;
6301	case VMDKETYPE_ZERO:
6302	/* No need to do anything for this extent. */
6303	break;
6304	default:
6305	AssertMsgFailed(("unknown extent type %d\n", pExtent->enmType));
6306	break;
6307	}
6308	}
6309	}
6310
6311	return rc;
6312	}
6313
6314	/**
6315	* Internal. Find extent corresponding to the sector number in the disk.
6316	*/
6317	static int vmdkFindExtent(PVMDKIMAGE pImage, uint64_t offSector,
6318	PVMDKEXTENT ppExtent, uint64_t puSectorInExtent)
6319	{
6320	PVMDKEXTENT pExtent = NULL;
6321	int rc = VINF_SUCCESS;
6322
6323	for (unsigned i = 0; i < pImage->cExtents; i++)
6324	{
6325	if (offSector < pImage->pExtents[i].cNominalSectors)
6326	{
6327	pExtent = &pImage->pExtents[i];
6328	*puSectorInExtent = offSector + pImage->pExtents[i].uSectorOffset;
6329	break;
6330	}
6331	offSector -= pImage->pExtents[i].cNominalSectors;
6332	}
6333
6334	if (pExtent)
6335	*ppExtent = pExtent;
6336	else
6337	rc = VERR_IO_SECTOR_NOT_FOUND;
6338
6339	return rc;
6340	}
6341
6342	/**
6343	* Internal. Hash function for placing the grain table hash entries.
6344	*/
6345	static uint32_t vmdkGTCacheHash(PVMDKGTCACHE pCache, uint64_t uSector,
6346	unsigned uExtent)
6347	{
6348	/** @todo this hash function is quite simple, maybe use a better one which
6349	* scrambles the bits better. */
6350	return (uSector + uExtent) % pCache->cEntries;
6351	}
6352
6353	/**
6354	* Internal. Get sector number in the extent file from the relative sector
6355	* number in the extent.
6356	*/
6357	static int vmdkGetSector(PVMDKIMAGE pImage, PVDIOCTX pIoCtx,
6358	PVMDKEXTENT pExtent, uint64_t uSector,
6359	uint64_t *puExtentSector)
6360	{
6361	PVMDKGTCACHE pCache = pImage->pGTCache;
6362	uint64_t uGDIndex, uGTSector, uGTBlock;
6363	uint32_t uGTHash, uGTBlockIndex;
6364	PVMDKGTCACHEENTRY pGTCacheEntry;
6365	uint32_t aGTDataTmp[VMDK_GT_CACHELINE_SIZE];
6366	int rc;
6367
6368	/* For newly created and readonly/sequentially opened streamOptimized
6369	* images this must be a no-op, as the grain directory is not there. */
6370	if ( ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
6371	&& pExtent->uAppendPosition)
6372	\|\| ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
6373	&& pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY
6374	&& pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
6375	{
6376	*puExtentSector = 0;
6377	return VINF_SUCCESS;
6378	}
6379
6380	uGDIndex = uSector / pExtent->cSectorsPerGDE;
6381	if (uGDIndex >= pExtent->cGDEntries)
6382	return VERR_OUT_OF_RANGE;
6383	uGTSector = pExtent->pGD[uGDIndex];
6384	if (!uGTSector)
6385	{
6386	/* There is no grain table referenced by this grain directory
6387	* entry. So there is absolutely no data in this area. */
6388	*puExtentSector = 0;
6389	return VINF_SUCCESS;
6390	}
6391
6392	uGTBlock = uSector / (pExtent->cSectorsPerGrain * VMDK_GT_CACHELINE_SIZE);
6393	uGTHash = vmdkGTCacheHash(pCache, uGTBlock, pExtent->uExtent);
6394	pGTCacheEntry = &pCache->aGTCache[uGTHash];
6395	if ( pGTCacheEntry->uExtent != pExtent->uExtent
6396	\|\| pGTCacheEntry->uGTBlock != uGTBlock)
6397	{
6398	/* Cache miss, fetch data from disk. */
6399	PVDMETAXFER pMetaXfer;
6400	rc = vdIfIoIntFileReadMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6401	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
6402	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx, &pMetaXfer, NULL, NULL);
6403	if (RT_FAILURE(rc))
6404	return rc;
6405	/* We can release the metadata transfer immediately. */
6406	vdIfIoIntMetaXferRelease(pImage->pIfIo, pMetaXfer);
6407	pGTCacheEntry->uExtent = pExtent->uExtent;
6408	pGTCacheEntry->uGTBlock = uGTBlock;
6409	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
6410	pGTCacheEntry->aGTData[i] = RT_LE2H_U32(aGTDataTmp[i]);
6411	}
6412	uGTBlockIndex = (uSector / pExtent->cSectorsPerGrain) % VMDK_GT_CACHELINE_SIZE;
6413	uint32_t uGrainSector = pGTCacheEntry->aGTData[uGTBlockIndex];
6414	if (uGrainSector)
6415	*puExtentSector = uGrainSector + uSector % pExtent->cSectorsPerGrain;
6416	else
6417	*puExtentSector = 0;
6418	return VINF_SUCCESS;
6419	}
6420
6421	/**
6422	* Internal. Writes the grain and also if necessary the grain tables.
6423	* Uses the grain table cache as a true grain table.
6424	*/
6425	static int vmdkStreamAllocGrain(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
6426	uint64_t uSector, PVDIOCTX pIoCtx,
6427	uint64_t cbWrite)
6428	{
6429	uint32_t uGrain;
6430	uint32_t uGDEntry, uLastGDEntry;
6431	uint32_t cbGrain = 0;
6432	uint32_t uCacheLine, uCacheEntry;
6433	const void *pData;
6434	int rc;
6435
6436	/* Very strict requirements: always write at least one full grain, with
6437	* proper alignment. Everything else would require reading of already
6438	* written data, which we don't support for obvious reasons. The only
6439	* exception is the last grain, and only if the image size specifies
6440	* that only some portion holds data. In any case the write must be
6441	* within the image limits, no "overshoot" allowed. */
6442	if ( cbWrite == 0
6443	\|\| ( cbWrite < VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain)
6444	&& pExtent->cNominalSectors - uSector >= pExtent->cSectorsPerGrain)
6445	\|\| uSector % pExtent->cSectorsPerGrain
6446	\|\| uSector + VMDK_BYTE2SECTOR(cbWrite) > pExtent->cNominalSectors)
6447	return VERR_INVALID_PARAMETER;
6448
6449	/* Clip write range to at most the rest of the grain. */
6450	cbWrite = RT_MIN(cbWrite, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain - uSector % pExtent->cSectorsPerGrain));
6451
6452	/* Do not allow to go back. */
6453	uGrain = uSector / pExtent->cSectorsPerGrain;
6454	uCacheLine = uGrain % pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE;
6455	uCacheEntry = uGrain % VMDK_GT_CACHELINE_SIZE;
6456	uGDEntry = uGrain / pExtent->cGTEntries;
6457	uLastGDEntry = pExtent->uLastGrainAccess / pExtent->cGTEntries;
6458	if (uGrain < pExtent->uLastGrainAccess)
6459	return VERR_VD_VMDK_INVALID_WRITE;
6460
6461	/* Zero byte write optimization. Since we don't tell VBoxHDD that we need
6462	* to allocate something, we also need to detect the situation ourself. */
6463	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_HONOR_ZEROES)
6464	&& vdIfIoIntIoCtxIsZero(pImage->pIfIo, pIoCtx, cbWrite, true /* fAdvance */))
6465	return VINF_SUCCESS;
6466
6467	if (uGDEntry != uLastGDEntry)
6468	{
6469	rc = vmdkStreamFlushGT(pImage, pExtent, uLastGDEntry);
6470	if (RT_FAILURE(rc))
6471	return rc;
6472	vmdkStreamClearGT(pImage, pExtent);
6473	for (uint32_t i = uLastGDEntry + 1; i < uGDEntry; i++)
6474	{
6475	rc = vmdkStreamFlushGT(pImage, pExtent, i);
6476	if (RT_FAILURE(rc))
6477	return rc;
6478	}
6479	}
6480
6481	uint64_t uFileOffset;
6482	uFileOffset = pExtent->uAppendPosition;
6483	if (!uFileOffset)
6484	return VERR_INTERNAL_ERROR;
6485	/* Align to sector, as the previous write could have been any size. */
6486	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
6487
6488	/* Paranoia check: extent type, grain table buffer presence and
6489	* grain table buffer space. Also grain table entry must be clear. */
6490	if ( pExtent->enmType != VMDKETYPE_HOSTED_SPARSE
6491	\|\| !pImage->pGTCache
6492	\|\| pExtent->cGTEntries > VMDK_GT_CACHE_SIZE * VMDK_GT_CACHELINE_SIZE
6493	\|\| pImage->pGTCache->aGTCache[uCacheLine].aGTData[uCacheEntry])
6494	return VERR_INTERNAL_ERROR;
6495
6496	/* Update grain table entry. */
6497	pImage->pGTCache->aGTCache[uCacheLine].aGTData[uCacheEntry] = VMDK_BYTE2SECTOR(uFileOffset);
6498
6499	if (cbWrite != VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
6500	{
6501	vdIfIoIntIoCtxCopyFrom(pImage->pIfIo, pIoCtx, pExtent->pvGrain, cbWrite);
6502	memset((char *)pExtent->pvGrain + cbWrite, '\0',
6503	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain) - cbWrite);
6504	pData = pExtent->pvGrain;
6505	}
6506	else
6507	{
6508	RTSGSEG Segment;
6509	unsigned cSegments = 1;
6510	size_t cbSeg = 0;
6511
6512	cbSeg = vdIfIoIntIoCtxSegArrayCreate(pImage->pIfIo, pIoCtx, &Segment,
6513	&cSegments, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
6514	Assert(cbSeg == VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
6515	pData = Segment.pvSeg;
6516	}
6517	rc = vmdkFileDeflateSync(pImage, pExtent, uFileOffset, pData,
6518	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
6519	uSector, &cbGrain);
6520	if (RT_FAILURE(rc))
6521	{
6522	pExtent->uGrainSectorAbs = 0;
6523	AssertRC(rc);
6524	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write compressed data block in '%s'"), pExtent->pszFullname);
6525	}
6526	pExtent->uLastGrainAccess = uGrain;
6527	pExtent->uAppendPosition += cbGrain;
6528
6529	return rc;
6530	}
6531
6532	/**
6533	* Internal: Updates the grain table during grain allocation.
6534	*/
6535	static int vmdkAllocGrainGTUpdate(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, PVDIOCTX pIoCtx,
6536	PVMDKGRAINALLOCASYNC pGrainAlloc)
6537	{
6538	int rc = VINF_SUCCESS;
6539	PVMDKGTCACHE pCache = pImage->pGTCache;
6540	uint32_t aGTDataTmp[VMDK_GT_CACHELINE_SIZE];
6541	uint32_t uGTHash, uGTBlockIndex;
6542	uint64_t uGTSector, uRGTSector, uGTBlock;
6543	uint64_t uSector = pGrainAlloc->uSector;
6544	PVMDKGTCACHEENTRY pGTCacheEntry;
6545
6546	LogFlowFunc(("pImage=%#p pExtent=%#p pCache=%#p pIoCtx=%#p pGrainAlloc=%#p\n",
6547	pImage, pExtent, pCache, pIoCtx, pGrainAlloc));
6548
6549	uGTSector = pGrainAlloc->uGTSector;
6550	uRGTSector = pGrainAlloc->uRGTSector;
6551	LogFlow(("uGTSector=%llu uRGTSector=%llu\n", uGTSector, uRGTSector));
6552
6553	/* Update the grain table (and the cache). */
6554	uGTBlock = uSector / (pExtent->cSectorsPerGrain * VMDK_GT_CACHELINE_SIZE);
6555	uGTHash = vmdkGTCacheHash(pCache, uGTBlock, pExtent->uExtent);
6556	pGTCacheEntry = &pCache->aGTCache[uGTHash];
6557	if ( pGTCacheEntry->uExtent != pExtent->uExtent
6558	\|\| pGTCacheEntry->uGTBlock != uGTBlock)
6559	{
6560	/* Cache miss, fetch data from disk. */
6561	LogFlow(("Cache miss, fetch data from disk\n"));
6562	PVDMETAXFER pMetaXfer = NULL;
6563	rc = vdIfIoIntFileReadMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6564	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
6565	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
6566	&pMetaXfer, vmdkAllocGrainComplete, pGrainAlloc);
6567	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6568	{
6569	pGrainAlloc->cIoXfersPending++;
6570	pGrainAlloc->fGTUpdateNeeded = true;
6571	/* Leave early, we will be called again after the read completed. */
6572	LogFlowFunc(("Metadata read in progress, leaving\n"));
6573	return rc;
6574	}
6575	else if (RT_FAILURE(rc))
6576	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot read allocated grain table entry in '%s'"), pExtent->pszFullname);
6577	vdIfIoIntMetaXferRelease(pImage->pIfIo, pMetaXfer);
6578	pGTCacheEntry->uExtent = pExtent->uExtent;
6579	pGTCacheEntry->uGTBlock = uGTBlock;
6580	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
6581	pGTCacheEntry->aGTData[i] = RT_LE2H_U32(aGTDataTmp[i]);
6582	}
6583	else
6584	{
6585	/* Cache hit. Convert grain table block back to disk format, otherwise
6586	* the code below will write garbage for all but the updated entry. */
6587	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
6588	aGTDataTmp[i] = RT_H2LE_U32(pGTCacheEntry->aGTData[i]);
6589	}
6590	pGrainAlloc->fGTUpdateNeeded = false;
6591	uGTBlockIndex = (uSector / pExtent->cSectorsPerGrain) % VMDK_GT_CACHELINE_SIZE;
6592	aGTDataTmp[uGTBlockIndex] = RT_H2LE_U32(VMDK_BYTE2SECTOR(pGrainAlloc->uGrainOffset));
6593	pGTCacheEntry->aGTData[uGTBlockIndex] = VMDK_BYTE2SECTOR(pGrainAlloc->uGrainOffset);
6594	/* Update grain table on disk. */
6595	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6596	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
6597	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
6598	vmdkAllocGrainComplete, pGrainAlloc);
6599	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6600	pGrainAlloc->cIoXfersPending++;
6601	else if (RT_FAILURE(rc))
6602	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write updated grain table in '%s'"), pExtent->pszFullname);
6603	if (pExtent->pRGD)
6604	{
6605	/* Update backup grain table on disk. */
6606	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6607	VMDK_SECTOR2BYTE(uRGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
6608	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
6609	vmdkAllocGrainComplete, pGrainAlloc);
6610	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6611	pGrainAlloc->cIoXfersPending++;
6612	else if (RT_FAILURE(rc))
6613	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write updated backup grain table in '%s'"), pExtent->pszFullname);
6614	}
6615
6616	LogFlowFunc(("leaving rc=%Rrc\n", rc));
6617	return rc;
6618	}
6619
6620	/**
6621	* Internal - complete the grain allocation by updating disk grain table if required.
6622	*/
6623	static DECLCALLBACK(int) vmdkAllocGrainComplete(void pBackendData, PVDIOCTX pIoCtx, void pvUser, int rcReq)
6624	{
6625	RT_NOREF1(rcReq);
6626	int rc = VINF_SUCCESS;
6627	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6628	PVMDKGRAINALLOCASYNC pGrainAlloc = (PVMDKGRAINALLOCASYNC)pvUser;
6629
6630	LogFlowFunc(("pBackendData=%#p pIoCtx=%#p pvUser=%#p rcReq=%Rrc\n",
6631	pBackendData, pIoCtx, pvUser, rcReq));
6632
6633	pGrainAlloc->cIoXfersPending--;
6634	if (!pGrainAlloc->cIoXfersPending && pGrainAlloc->fGTUpdateNeeded)
6635	rc = vmdkAllocGrainGTUpdate(pImage, pGrainAlloc->pExtent, pIoCtx, pGrainAlloc);
6636
6637	if (!pGrainAlloc->cIoXfersPending)
6638	{
6639	/* Grain allocation completed. */
6640	RTMemFree(pGrainAlloc);
6641	}
6642
6643	LogFlowFunc(("Leaving rc=%Rrc\n", rc));
6644	return rc;
6645	}
6646
6647	/**
6648	* Internal. Allocates a new grain table (if necessary).
6649	*/
6650	static int vmdkAllocGrain(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, PVDIOCTX pIoCtx,
6651	uint64_t uSector, uint64_t cbWrite)
6652	{
6653	PVMDKGTCACHE pCache = pImage->pGTCache; NOREF(pCache);
6654	uint64_t uGDIndex, uGTSector, uRGTSector;
6655	uint64_t uFileOffset;
6656	PVMDKGRAINALLOCASYNC pGrainAlloc = NULL;
6657	int rc;
6658
6659	LogFlowFunc(("pCache=%#p pExtent=%#p pIoCtx=%#p uSector=%llu cbWrite=%llu\n",
6660	pCache, pExtent, pIoCtx, uSector, cbWrite));
6661
6662	pGrainAlloc = (PVMDKGRAINALLOCASYNC)RTMemAllocZ(sizeof(VMDKGRAINALLOCASYNC));
6663	if (!pGrainAlloc)
6664	return VERR_NO_MEMORY;
6665
6666	pGrainAlloc->pExtent = pExtent;
6667	pGrainAlloc->uSector = uSector;
6668
6669	uGDIndex = uSector / pExtent->cSectorsPerGDE;
6670	if (uGDIndex >= pExtent->cGDEntries)
6671	{
6672	RTMemFree(pGrainAlloc);
6673	return VERR_OUT_OF_RANGE;
6674	}
6675	uGTSector = pExtent->pGD[uGDIndex];
6676	if (pExtent->pRGD)
6677	uRGTSector = pExtent->pRGD[uGDIndex];
6678	else
6679	uRGTSector = 0; /*< avoid compiler warning /
6680	if (!uGTSector)
6681	{
6682	LogFlow(("Allocating new grain table\n"));
6683
6684	/* There is no grain table referenced by this grain directory
6685	* entry. So there is absolutely no data in this area. Allocate
6686	* a new grain table and put the reference to it in the GDs. */
6687	uFileOffset = pExtent->uAppendPosition;
6688	if (!uFileOffset)
6689	{
6690	RTMemFree(pGrainAlloc);
6691	return VERR_INTERNAL_ERROR;
6692	}
6693	Assert(!(uFileOffset % 512));
6694
6695	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
6696	uGTSector = VMDK_BYTE2SECTOR(uFileOffset);
6697
6698	/* Normally the grain table is preallocated for hosted sparse extents
6699	* that support more than 32 bit sector numbers. So this shouldn't
6700	* ever happen on a valid extent. */
6701	if (uGTSector > UINT32_MAX)
6702	{
6703	RTMemFree(pGrainAlloc);
6704	return VERR_VD_VMDK_INVALID_HEADER;
6705	}
6706
6707	/* Write grain table by writing the required number of grain table
6708	* cache chunks. Allocate memory dynamically here or we flood the
6709	* metadata cache with very small entries. */
6710	size_t cbGTDataTmp = pExtent->cGTEntries * sizeof(uint32_t);
6711	uint32_t paGTDataTmp = (uint32_t )RTMemTmpAllocZ(cbGTDataTmp);
6712
6713	if (!paGTDataTmp)
6714	{
6715	RTMemFree(pGrainAlloc);
6716	return VERR_NO_MEMORY;
6717	}
6718
6719	memset(paGTDataTmp, '\0', cbGTDataTmp);
6720	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6721	VMDK_SECTOR2BYTE(uGTSector),
6722	paGTDataTmp, cbGTDataTmp, pIoCtx,
6723	vmdkAllocGrainComplete, pGrainAlloc);
6724	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6725	pGrainAlloc->cIoXfersPending++;
6726	else if (RT_FAILURE(rc))
6727	{
6728	RTMemTmpFree(paGTDataTmp);
6729	RTMemFree(pGrainAlloc);
6730	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write grain table allocation in '%s'"), pExtent->pszFullname);
6731	}
6732	pExtent->uAppendPosition = RT_ALIGN_64( pExtent->uAppendPosition
6733	+ cbGTDataTmp, 512);
6734
6735	if (pExtent->pRGD)
6736	{
6737	AssertReturn(!uRGTSector, VERR_VD_VMDK_INVALID_HEADER);
6738	uFileOffset = pExtent->uAppendPosition;
6739	if (!uFileOffset)
6740	return VERR_INTERNAL_ERROR;
6741	Assert(!(uFileOffset % 512));
6742	uRGTSector = VMDK_BYTE2SECTOR(uFileOffset);
6743
6744	/* Normally the redundant grain table is preallocated for hosted
6745	* sparse extents that support more than 32 bit sector numbers. So
6746	* this shouldn't ever happen on a valid extent. */
6747	if (uRGTSector > UINT32_MAX)
6748	{
6749	RTMemTmpFree(paGTDataTmp);
6750	return VERR_VD_VMDK_INVALID_HEADER;
6751	}
6752
6753	/* Write grain table by writing the required number of grain table
6754	* cache chunks. Allocate memory dynamically here or we flood the
6755	* metadata cache with very small entries. */
6756	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6757	VMDK_SECTOR2BYTE(uRGTSector),
6758	paGTDataTmp, cbGTDataTmp, pIoCtx,
6759	vmdkAllocGrainComplete, pGrainAlloc);
6760	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6761	pGrainAlloc->cIoXfersPending++;
6762	else if (RT_FAILURE(rc))
6763	{
6764	RTMemTmpFree(paGTDataTmp);
6765	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write backup grain table allocation in '%s'"), pExtent->pszFullname);
6766	}
6767
6768	pExtent->uAppendPosition = pExtent->uAppendPosition + cbGTDataTmp;
6769	}
6770
6771	RTMemTmpFree(paGTDataTmp);
6772
6773	/* Update the grain directory on disk (doing it before writing the
6774	* grain table will result in a garbled extent if the operation is
6775	* aborted for some reason. Otherwise the worst that can happen is
6776	* some unused sectors in the extent. */
6777	uint32_t uGTSectorLE = RT_H2LE_U64(uGTSector);
6778	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6779	VMDK_SECTOR2BYTE(pExtent->uSectorGD) + uGDIndex * sizeof(uGTSectorLE),
6780	&uGTSectorLE, sizeof(uGTSectorLE), pIoCtx,
6781	vmdkAllocGrainComplete, pGrainAlloc);
6782	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6783	pGrainAlloc->cIoXfersPending++;
6784	else if (RT_FAILURE(rc))
6785	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write grain directory entry in '%s'"), pExtent->pszFullname);
6786	if (pExtent->pRGD)
6787	{
6788	uint32_t uRGTSectorLE = RT_H2LE_U64(uRGTSector);
6789	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6790	VMDK_SECTOR2BYTE(pExtent->uSectorRGD) + uGDIndex * sizeof(uGTSectorLE),
6791	&uRGTSectorLE, sizeof(uRGTSectorLE), pIoCtx,
6792	vmdkAllocGrainComplete, pGrainAlloc);
6793	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6794	pGrainAlloc->cIoXfersPending++;
6795	else if (RT_FAILURE(rc))
6796	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write backup grain directory entry in '%s'"), pExtent->pszFullname);
6797	}
6798
6799	/* As the final step update the in-memory copy of the GDs. */
6800	pExtent->pGD[uGDIndex] = uGTSector;
6801	if (pExtent->pRGD)
6802	pExtent->pRGD[uGDIndex] = uRGTSector;
6803	}
6804
6805	LogFlow(("uGTSector=%llu uRGTSector=%llu\n", uGTSector, uRGTSector));
6806	pGrainAlloc->uGTSector = uGTSector;
6807	pGrainAlloc->uRGTSector = uRGTSector;
6808
6809	uFileOffset = pExtent->uAppendPosition;
6810	if (!uFileOffset)
6811	return VERR_INTERNAL_ERROR;
6812	Assert(!(uFileOffset % 512));
6813
6814	pGrainAlloc->uGrainOffset = uFileOffset;
6815
6816	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6817	{
6818	AssertMsgReturn(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
6819	("Accesses to stream optimized images must be synchronous\n"),
6820	VERR_INVALID_STATE);
6821
6822	if (cbWrite != VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
6823	return vdIfError(pImage->pIfError, VERR_INTERNAL_ERROR, RT_SRC_POS, N_("VMDK: not enough data for a compressed data block in '%s'"), pExtent->pszFullname);
6824
6825	/* Invalidate cache, just in case some code incorrectly allows mixing
6826	* of reads and writes. Normally shouldn't be needed. */
6827	pExtent->uGrainSectorAbs = 0;
6828
6829	/* Write compressed data block and the markers. */
6830	uint32_t cbGrain = 0;
6831	size_t cbSeg = 0;
6832	RTSGSEG Segment;
6833	unsigned cSegments = 1;
6834
6835	cbSeg = vdIfIoIntIoCtxSegArrayCreate(pImage->pIfIo, pIoCtx, &Segment,
6836	&cSegments, cbWrite);
6837	Assert(cbSeg == cbWrite);
6838
6839	rc = vmdkFileDeflateSync(pImage, pExtent, uFileOffset,
6840	Segment.pvSeg, cbWrite, uSector, &cbGrain);
6841	if (RT_FAILURE(rc))
6842	{
6843	AssertRC(rc);
6844	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write allocated compressed data block in '%s'"), pExtent->pszFullname);
6845	}
6846	pExtent->uLastGrainAccess = uSector / pExtent->cSectorsPerGrain;
6847	pExtent->uAppendPosition += cbGrain;
6848	}
6849	else
6850	{
6851	/* Write the data. Always a full grain, or we're in big trouble. */
6852	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
6853	uFileOffset, pIoCtx, cbWrite,
6854	vmdkAllocGrainComplete, pGrainAlloc);
6855	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6856	pGrainAlloc->cIoXfersPending++;
6857	else if (RT_FAILURE(rc))
6858	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write allocated data block in '%s'"), pExtent->pszFullname);
6859
6860	pExtent->uAppendPosition += cbWrite;
6861	}
6862
6863	rc = vmdkAllocGrainGTUpdate(pImage, pExtent, pIoCtx, pGrainAlloc);
6864
6865	if (!pGrainAlloc->cIoXfersPending)
6866	{
6867	/* Grain allocation completed. */
6868	RTMemFree(pGrainAlloc);
6869	}
6870
6871	LogFlowFunc(("leaving rc=%Rrc\n", rc));
6872
6873	return rc;
6874	}
6875
6876	/**
6877	* Internal. Reads the contents by sequentially going over the compressed
6878	* grains (hoping that they are in sequence).
6879	*/
6880	static int vmdkStreamReadSequential(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
6881	uint64_t uSector, PVDIOCTX pIoCtx,
6882	uint64_t cbRead)
6883	{
6884	int rc;
6885
6886	LogFlowFunc(("pImage=%#p pExtent=%#p uSector=%llu pIoCtx=%#p cbRead=%llu\n",
6887	pImage, pExtent, uSector, pIoCtx, cbRead));
6888
6889	AssertMsgReturn(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
6890	("Async I/O not supported for sequential stream optimized images\n"),
6891	VERR_INVALID_STATE);
6892
6893	/* Do not allow to go back. */
6894	uint32_t uGrain = uSector / pExtent->cSectorsPerGrain;
6895	if (uGrain < pExtent->uLastGrainAccess)
6896	return VERR_VD_VMDK_INVALID_STATE;
6897	pExtent->uLastGrainAccess = uGrain;
6898
6899	/* After a previous error do not attempt to recover, as it would need
6900	* seeking (in the general case backwards which is forbidden). */
6901	if (!pExtent->uGrainSectorAbs)
6902	return VERR_VD_VMDK_INVALID_STATE;
6903
6904	/* Check if we need to read something from the image or if what we have
6905	* in the buffer is good to fulfill the request. */
6906	if (!pExtent->cbGrainStreamRead \|\| uGrain > pExtent->uGrain)
6907	{
6908	uint32_t uGrainSectorAbs = pExtent->uGrainSectorAbs
6909	+ VMDK_BYTE2SECTOR(pExtent->cbGrainStreamRead);
6910
6911	/* Get the marker from the next data block - and skip everything which
6912	* is not a compressed grain. If it's a compressed grain which is for
6913	* the requested sector (or after), read it. */
6914	VMDKMARKER Marker;
6915	do
6916	{
6917	RT_ZERO(Marker);
6918	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
6919	VMDK_SECTOR2BYTE(uGrainSectorAbs),
6920	&Marker, RT_UOFFSETOF(VMDKMARKER, uType));
6921	if (RT_FAILURE(rc))
6922	return rc;
6923	Marker.uSector = RT_LE2H_U64(Marker.uSector);
6924	Marker.cbSize = RT_LE2H_U32(Marker.cbSize);
6925
6926	if (Marker.cbSize == 0)
6927	{
6928	/* A marker for something else than a compressed grain. */
6929	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
6930	VMDK_SECTOR2BYTE(uGrainSectorAbs)
6931	+ RT_UOFFSETOF(VMDKMARKER, uType),
6932	&Marker.uType, sizeof(Marker.uType));
6933	if (RT_FAILURE(rc))
6934	return rc;
6935	Marker.uType = RT_LE2H_U32(Marker.uType);
6936	switch (Marker.uType)
6937	{
6938	case VMDK_MARKER_EOS:
6939	uGrainSectorAbs++;
6940	/* Read (or mostly skip) to the end of file. Uses the
6941	* Marker (LBA sector) as it is unused anyway. This
6942	* makes sure that really everything is read in the
6943	* success case. If this read fails it means the image
6944	* is truncated, but this is harmless so ignore. */
6945	vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
6946	VMDK_SECTOR2BYTE(uGrainSectorAbs)
6947	+ 511,
6948	&Marker.uSector, 1);
6949	break;
6950	case VMDK_MARKER_GT:
6951	uGrainSectorAbs += 1 + VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
6952	break;
6953	case VMDK_MARKER_GD:
6954	uGrainSectorAbs += 1 + VMDK_BYTE2SECTOR(RT_ALIGN(pExtent->cGDEntries * sizeof(uint32_t), 512));
6955	break;
6956	case VMDK_MARKER_FOOTER:
6957	uGrainSectorAbs += 2;
6958	break;
6959	case VMDK_MARKER_UNSPECIFIED:
6960	/* Skip over the contents of the unspecified marker
6961	* type 4 which exists in some vSphere created files. */
6962	/** @todo figure out what the payload means. */
6963	uGrainSectorAbs += 1;
6964	break;
6965	default:
6966	AssertMsgFailed(("VMDK: corrupted marker, type=%#x\n", Marker.uType));
6967	pExtent->uGrainSectorAbs = 0;
6968	return VERR_VD_VMDK_INVALID_STATE;
6969	}
6970	pExtent->cbGrainStreamRead = 0;
6971	}
6972	else
6973	{
6974	/* A compressed grain marker. If it is at/after what we're
6975	* interested in read and decompress data. */
6976	if (uSector > Marker.uSector + pExtent->cSectorsPerGrain)
6977	{
6978	uGrainSectorAbs += VMDK_BYTE2SECTOR(RT_ALIGN(Marker.cbSize + RT_UOFFSETOF(VMDKMARKER, uType), 512));
6979	continue;
6980	}
6981	uint64_t uLBA = 0;
6982	uint32_t cbGrainStreamRead = 0;
6983	rc = vmdkFileInflateSync(pImage, pExtent,
6984	VMDK_SECTOR2BYTE(uGrainSectorAbs),
6985	pExtent->pvGrain,
6986	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
6987	&Marker, &uLBA, &cbGrainStreamRead);
6988	if (RT_FAILURE(rc))
6989	{
6990	pExtent->uGrainSectorAbs = 0;
6991	return rc;
6992	}
6993	if ( pExtent->uGrain
6994	&& uLBA / pExtent->cSectorsPerGrain <= pExtent->uGrain)
6995	{
6996	pExtent->uGrainSectorAbs = 0;
6997	return VERR_VD_VMDK_INVALID_STATE;
6998	}
6999	pExtent->uGrain = uLBA / pExtent->cSectorsPerGrain;
7000	pExtent->cbGrainStreamRead = cbGrainStreamRead;
7001	break;
7002	}
7003	} while (Marker.uType != VMDK_MARKER_EOS);
7004
7005	pExtent->uGrainSectorAbs = uGrainSectorAbs;
7006
7007	if (!pExtent->cbGrainStreamRead && Marker.uType == VMDK_MARKER_EOS)
7008	{
7009	pExtent->uGrain = UINT32_MAX;
7010	/* Must set a non-zero value for pExtent->cbGrainStreamRead or
7011	* the next read would try to get more data, and we're at EOF. */
7012	pExtent->cbGrainStreamRead = 1;
7013	}
7014	}
7015
7016	if (pExtent->uGrain > uSector / pExtent->cSectorsPerGrain)
7017	{
7018	/* The next data block we have is not for this area, so just return
7019	* that there is no data. */
7020	LogFlowFunc(("returns VERR_VD_BLOCK_FREE\n"));
7021	return VERR_VD_BLOCK_FREE;
7022	}
7023
7024	uint32_t uSectorInGrain = uSector % pExtent->cSectorsPerGrain;
7025	vdIfIoIntIoCtxCopyTo(pImage->pIfIo, pIoCtx,
7026	(uint8_t *)pExtent->pvGrain + VMDK_SECTOR2BYTE(uSectorInGrain),
7027	cbRead);
7028	LogFlowFunc(("returns VINF_SUCCESS\n"));
7029	return VINF_SUCCESS;
7030	}
7031
7032	/**
7033	* Replaces a fragment of a string with the specified string.
7034	*
7035	* @returns Pointer to the allocated UTF-8 string.
7036	* @param pszWhere UTF-8 string to search in.
7037	* @param pszWhat UTF-8 string to search for.
7038	* @param pszByWhat UTF-8 string to replace the found string with.
7039	*
7040	* @note r=bird: This is only used by vmdkRenameWorker(). The first use is
7041	* for updating the base name in the descriptor, the second is for
7042	* generating new filenames for extents. This code borked when
7043	* RTPathAbs started correcting the driver letter case on windows,
7044	* when strstr failed because the pExtent->pszFullname was not
7045	* subjected to RTPathAbs but while pExtent->pszFullname was. I fixed
7046	* this by apply RTPathAbs to the places it wasn't applied.
7047	*
7048	* However, this highlights some undocumented ASSUMPTIONS as well as
7049	* terrible short commings of the approach.
7050	*
7051	* Given the right filename, it may also screw up the descriptor. Take
7052	* the descriptor text 'RW 2048 SPARSE "Test0.vmdk"' for instance,
7053	* we'll be asked to replace "Test0" with something, no problem. No,
7054	* imagine 'RW 2048 SPARSE "SPARSE.vmdk"', 'RW 2048 SPARSE "RW.vmdk"'
7055	* or 'RW 2048 SPARSE "2048.vmdk"', and the strstr approach falls on
7056	* its bum. The descriptor string must be parsed and reconstructed,
7057	* the lazy strstr approach doesn't cut it.
7058	*
7059	* I'm also curious as to what would be the correct escaping of '"' in
7060	* the file name and how that is supposed to be handled, because it
7061	* needs to be or such names must be rejected in several places (maybe
7062	* they are, I didn't check).
7063	*
7064	* When this function is used to replace the start of a path, I think
7065	* the assumption from the prep/setup code is that we kind of knows
7066	* what we're working on (I could be wrong). However, using strstr
7067	* instead of strncmp/RTStrNICmp makes no sense and isn't future proof.
7068	* Especially on unix systems, weird stuff could happen if someone
7069	* unwittingly tinkers with the prep/setup code. What should really be
7070	* done here is using a new RTPathStartEx function that (via flags)
7071	* allows matching partial final component and returns the length of
7072	* what it matched up (in case it skipped slashes and '.' components).
7073	*
7074	*/
7075	static char vmdkStrReplace(const char pszWhere, const char *pszWhat,
7076	const char *pszByWhat)
7077	{
7078	AssertPtr(pszWhere);
7079	AssertPtr(pszWhat);
7080	AssertPtr(pszByWhat);
7081	const char *pszFoundStr = strstr(pszWhere, pszWhat);
7082	if (!pszFoundStr)
7083	{
7084	LogFlowFunc(("Failed to find '%s' in '%s'!\n", pszWhat, pszWhere));
7085	return NULL;
7086	}
7087	size_t cbFinal = strlen(pszWhere) + 1 + strlen(pszByWhat) - strlen(pszWhat);
7088	char *pszNewStr = RTStrAlloc(cbFinal);
7089	if (pszNewStr)
7090	{
7091	char *pszTmp = pszNewStr;
7092	memcpy(pszTmp, pszWhere, pszFoundStr - pszWhere);
7093	pszTmp += pszFoundStr - pszWhere;
7094	memcpy(pszTmp, pszByWhat, strlen(pszByWhat));
7095	pszTmp += strlen(pszByWhat);
7096	strcpy(pszTmp, pszFoundStr + strlen(pszWhat));
7097	}
7098	return pszNewStr;
7099	}
7100
7101
7102	/** @copydoc VDIMAGEBACKEND::pfnProbe */
7103	static DECLCALLBACK(int) vmdkProbe(const char *pszFilename, PVDINTERFACE pVDIfsDisk,
7104	PVDINTERFACE pVDIfsImage, VDTYPE enmDesiredType, VDTYPE *penmType)
7105	{
7106	RT_NOREF(enmDesiredType);
7107	LogFlowFunc(("pszFilename=\"%s\" pVDIfsDisk=%#p pVDIfsImage=%#p penmType=%#p\n",
7108	pszFilename, pVDIfsDisk, pVDIfsImage, penmType));
7109	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
7110	AssertReturn(*pszFilename != '\0', VERR_INVALID_PARAMETER);
7111
7112	int rc = VINF_SUCCESS;
7113	PVMDKIMAGE pImage = (PVMDKIMAGE)RTMemAllocZ(RT_UOFFSETOF(VMDKIMAGE, RegionList.aRegions[1]));
7114	if (RT_LIKELY(pImage))
7115	{
7116	pImage->pszFilename = pszFilename;
7117	pImage->pFile = NULL;
7118	pImage->pExtents = NULL;
7119	pImage->pFiles = NULL;
7120	pImage->pGTCache = NULL;
7121	pImage->pDescData = NULL;
7122	pImage->pVDIfsDisk = pVDIfsDisk;
7123	pImage->pVDIfsImage = pVDIfsImage;
7124	/** @todo speed up this test open (VD_OPEN_FLAGS_INFO) by skipping as
7125	* much as possible in vmdkOpenImage. */
7126	rc = vmdkOpenImage(pImage, VD_OPEN_FLAGS_INFO \| VD_OPEN_FLAGS_READONLY);
7127	vmdkFreeImage(pImage, false, false /fFlush/);
7128	RTMemFree(pImage);
7129
7130	if (RT_SUCCESS(rc))
7131	*penmType = VDTYPE_HDD;
7132	}
7133	else
7134	rc = VERR_NO_MEMORY;
7135
7136	LogFlowFunc(("returns %Rrc\n", rc));
7137	return rc;
7138	}
7139
7140	/** @copydoc VDIMAGEBACKEND::pfnOpen */
7141	static DECLCALLBACK(int) vmdkOpen(const char *pszFilename, unsigned uOpenFlags,
7142	PVDINTERFACE pVDIfsDisk, PVDINTERFACE pVDIfsImage,
7143	VDTYPE enmType, void **ppBackendData)
7144	{
7145	RT_NOREF1(enmType); /*< @todo r=klaus make use of the type info. /
7146
7147	LogFlowFunc(("pszFilename=\"%s\" uOpenFlags=%#x pVDIfsDisk=%#p pVDIfsImage=%#p enmType=%u ppBackendData=%#p\n",
7148	pszFilename, uOpenFlags, pVDIfsDisk, pVDIfsImage, enmType, ppBackendData));
7149	int rc;
7150
7151	/* Check open flags. All valid flags are supported. */
7152	AssertReturn(!(uOpenFlags & ~VD_OPEN_FLAGS_MASK), VERR_INVALID_PARAMETER);
7153	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
7154	AssertReturn(*pszFilename != '\0', VERR_INVALID_PARAMETER);
7155
7156
7157	PVMDKIMAGE pImage = (PVMDKIMAGE)RTMemAllocZ(RT_UOFFSETOF(VMDKIMAGE, RegionList.aRegions[1]));
7158	if (RT_LIKELY(pImage))
7159	{
7160	pImage->pszFilename = pszFilename;
7161	pImage->pFile = NULL;
7162	pImage->pExtents = NULL;
7163	pImage->pFiles = NULL;
7164	pImage->pGTCache = NULL;
7165	pImage->pDescData = NULL;
7166	pImage->pVDIfsDisk = pVDIfsDisk;
7167	pImage->pVDIfsImage = pVDIfsImage;
7168
7169	rc = vmdkOpenImage(pImage, uOpenFlags);
7170	if (RT_SUCCESS(rc))
7171	*ppBackendData = pImage;
7172	else
7173	RTMemFree(pImage);
7174	}
7175	else
7176	rc = VERR_NO_MEMORY;
7177
7178	LogFlowFunc(("returns %Rrc (pBackendData=%#p)\n", rc, *ppBackendData));
7179	return rc;
7180	}
7181
7182	/** @copydoc VDIMAGEBACKEND::pfnCreate */
7183	static DECLCALLBACK(int) vmdkCreate(const char *pszFilename, uint64_t cbSize,
7184	unsigned uImageFlags, const char *pszComment,
7185	PCVDGEOMETRY pPCHSGeometry, PCVDGEOMETRY pLCHSGeometry,
7186	PCRTUUID pUuid, unsigned uOpenFlags,
7187	unsigned uPercentStart, unsigned uPercentSpan,
7188	PVDINTERFACE pVDIfsDisk, PVDINTERFACE pVDIfsImage,
7189	PVDINTERFACE pVDIfsOperation, VDTYPE enmType,
7190	void **ppBackendData)
7191	{
7192	LogFlowFunc(("pszFilename=\"%s\" cbSize=%llu uImageFlags=%#x pszComment=\"%s\" pPCHSGeometry=%#p pLCHSGeometry=%#p Uuid=%RTuuid uOpenFlags=%#x uPercentStart=%u uPercentSpan=%u pVDIfsDisk=%#p pVDIfsImage=%#p pVDIfsOperation=%#p enmType=%u ppBackendData=%#p\n",
7193	pszFilename, cbSize, uImageFlags, pszComment, pPCHSGeometry, pLCHSGeometry, pUuid, uOpenFlags, uPercentStart, uPercentSpan, pVDIfsDisk, pVDIfsImage, pVDIfsOperation, enmType, ppBackendData));
7194	int rc;
7195
7196	/* Check the VD container type and image flags. */
7197	if ( enmType != VDTYPE_HDD
7198	\|\| (uImageFlags & ~VD_VMDK_IMAGE_FLAGS_MASK) != 0)
7199	return VERR_VD_INVALID_TYPE;
7200
7201	/* Check size. Maximum 256TB-64K for sparse images, otherwise unlimited. */
7202	if ( !(uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
7203	&& ( !cbSize
7204	\|\| (!(uImageFlags & VD_IMAGE_FLAGS_FIXED) && cbSize >= _1T * 256 - _64K)))
7205	return VERR_VD_INVALID_SIZE;
7206
7207	/* Check image flags for invalid combinations. */
7208	if ( (uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7209	&& (uImageFlags & ~(VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED \| VD_IMAGE_FLAGS_DIFF)))
7210	return VERR_INVALID_PARAMETER;
7211
7212	/* Check open flags. All valid flags are supported. */
7213	AssertReturn(!(uOpenFlags & ~VD_OPEN_FLAGS_MASK), VERR_INVALID_PARAMETER);
7214	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
7215	AssertReturn(*pszFilename != '\0', VERR_INVALID_PARAMETER);
7216	AssertPtrReturn(pPCHSGeometry, VERR_INVALID_POINTER);
7217	AssertPtrReturn(pLCHSGeometry, VERR_INVALID_POINTER);
7218	AssertReturn(!( uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX
7219	&& !(uImageFlags & VD_IMAGE_FLAGS_FIXED)),
7220	VERR_INVALID_PARAMETER);
7221
7222	PVMDKIMAGE pImage = (PVMDKIMAGE)RTMemAllocZ(RT_UOFFSETOF(VMDKIMAGE, RegionList.aRegions[1]));
7223	if (RT_LIKELY(pImage))
7224	{
7225	PVDINTERFACEPROGRESS pIfProgress = VDIfProgressGet(pVDIfsOperation);
7226
7227	pImage->pszFilename = pszFilename;
7228	pImage->pFile = NULL;
7229	pImage->pExtents = NULL;
7230	pImage->pFiles = NULL;
7231	pImage->pGTCache = NULL;
7232	pImage->pDescData = NULL;
7233	pImage->pVDIfsDisk = pVDIfsDisk;
7234	pImage->pVDIfsImage = pVDIfsImage;
7235	/* Descriptors for split images can be pretty large, especially if the
7236	* filename is long. So prepare for the worst, and allocate quite some
7237	* memory for the descriptor in this case. */
7238	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
7239	pImage->cbDescAlloc = VMDK_SECTOR2BYTE(200);
7240	else
7241	pImage->cbDescAlloc = VMDK_SECTOR2BYTE(20);
7242	pImage->pDescData = (char *)RTMemAllocZ(pImage->cbDescAlloc);
7243	if (RT_LIKELY(pImage->pDescData))
7244	{
7245	rc = vmdkCreateImage(pImage, cbSize, uImageFlags, pszComment,
7246	pPCHSGeometry, pLCHSGeometry, pUuid,
7247	pIfProgress, uPercentStart, uPercentSpan);
7248	if (RT_SUCCESS(rc))
7249	{
7250	/* So far the image is opened in read/write mode. Make sure the
7251	* image is opened in read-only mode if the caller requested that. */
7252	if (uOpenFlags & VD_OPEN_FLAGS_READONLY)
7253	{
7254	vmdkFreeImage(pImage, false, true /fFlush/);
7255	rc = vmdkOpenImage(pImage, uOpenFlags);
7256	}
7257
7258	if (RT_SUCCESS(rc))
7259	*ppBackendData = pImage;
7260	}
7261
7262	if (RT_FAILURE(rc))
7263	RTMemFree(pImage->pDescData);
7264	}
7265	else
7266	rc = VERR_NO_MEMORY;
7267
7268	if (RT_FAILURE(rc))
7269	RTMemFree(pImage);
7270	}
7271	else
7272	rc = VERR_NO_MEMORY;
7273
7274	LogFlowFunc(("returns %Rrc (pBackendData=%#p)\n", rc, *ppBackendData));
7275	return rc;
7276	}
7277
7278	/**
7279	* Prepares the state for renaming a VMDK image, setting up the state and allocating
7280	* memory.
7281	*
7282	* @returns VBox status code.
7283	* @param pImage VMDK image instance.
7284	* @param pRenameState The state to initialize.
7285	* @param pszFilename The new filename.
7286	*/
7287	static int vmdkRenameStatePrepare(PVMDKIMAGE pImage, PVMDKRENAMESTATE pRenameState, const char *pszFilename)
7288	{
7289	AssertReturn(RTPathFilename(pszFilename) != NULL, VERR_INVALID_PARAMETER);
7290
7291	int rc = VINF_SUCCESS;
7292
7293	memset(&pRenameState->DescriptorCopy, 0, sizeof(pRenameState->DescriptorCopy));
7294
7295	/*
7296	* Allocate an array to store both old and new names of renamed files
7297	* in case we have to roll back the changes. Arrays are initialized
7298	* with zeros. We actually save stuff when and if we change it.
7299	*/
7300	pRenameState->cExtents = pImage->cExtents;
7301	pRenameState->apszOldName = (char *)RTMemTmpAllocZ((pRenameState->cExtents + 1) sizeof(char *));
7302	pRenameState->apszNewName = (char *)RTMemTmpAllocZ((pRenameState->cExtents + 1) sizeof(char *));
7303	pRenameState->apszNewLines = (char *)RTMemTmpAllocZ(pRenameState->cExtents sizeof(char *));
7304	if ( pRenameState->apszOldName
7305	&& pRenameState->apszNewName
7306	&& pRenameState->apszNewLines)
7307	{
7308	/* Save the descriptor size and position. */
7309	if (pImage->pDescData)
7310	{
7311	/* Separate descriptor file. */
7312	pRenameState->fEmbeddedDesc = false;
7313	}
7314	else
7315	{
7316	/* Embedded descriptor file. */
7317	pRenameState->ExtentCopy = pImage->pExtents[0];
7318	pRenameState->fEmbeddedDesc = true;
7319	}
7320
7321	/* Save the descriptor content. */
7322	pRenameState->DescriptorCopy.cLines = pImage->Descriptor.cLines;
7323	for (unsigned i = 0; i < pRenameState->DescriptorCopy.cLines; i++)
7324	{
7325	pRenameState->DescriptorCopy.aLines[i] = RTStrDup(pImage->Descriptor.aLines[i]);
7326	if (!pRenameState->DescriptorCopy.aLines[i])
7327	{
7328	rc = VERR_NO_MEMORY;
7329	break;
7330	}
7331	}
7332
7333	if (RT_SUCCESS(rc))
7334	{
7335	/* Prepare both old and new base names used for string replacement. */
7336	pRenameState->pszNewBaseName = RTStrDup(RTPathFilename(pszFilename));
7337	AssertReturn(pRenameState->pszNewBaseName, VERR_NO_STR_MEMORY);
7338	RTPathStripSuffix(pRenameState->pszNewBaseName);
7339
7340	pRenameState->pszOldBaseName = RTStrDup(RTPathFilename(pImage->pszFilename));
7341	AssertReturn(pRenameState->pszOldBaseName, VERR_NO_STR_MEMORY);
7342	RTPathStripSuffix(pRenameState->pszOldBaseName);
7343
7344	/* Prepare both old and new full names used for string replacement.
7345	Note! Must abspath the stuff here, so the strstr weirdness later in
7346	the renaming process get a match against abspath'ed extent paths.
7347	See RTPathAbsDup call in vmdkDescriptorReadSparse(). */
7348	pRenameState->pszNewFullName = RTPathAbsDup(pszFilename);
7349	AssertReturn(pRenameState->pszNewFullName, VERR_NO_STR_MEMORY);
7350	RTPathStripSuffix(pRenameState->pszNewFullName);
7351
7352	pRenameState->pszOldFullName = RTPathAbsDup(pImage->pszFilename);
7353	AssertReturn(pRenameState->pszOldFullName, VERR_NO_STR_MEMORY);
7354	RTPathStripSuffix(pRenameState->pszOldFullName);
7355
7356	/* Save the old name for easy access to the old descriptor file. */
7357	pRenameState->pszOldDescName = RTStrDup(pImage->pszFilename);
7358	AssertReturn(pRenameState->pszOldDescName, VERR_NO_STR_MEMORY);
7359
7360	/* Save old image name. */
7361	pRenameState->pszOldImageName = pImage->pszFilename;
7362	}
7363	}
7364	else
7365	rc = VERR_NO_TMP_MEMORY;
7366
7367	return rc;
7368	}
7369
7370	/**
7371	* Destroys the given rename state, freeing all allocated memory.
7372	*
7373	* @param pRenameState The rename state to destroy.
7374	*/
7375	static void vmdkRenameStateDestroy(PVMDKRENAMESTATE pRenameState)
7376	{
7377	for (unsigned i = 0; i < pRenameState->DescriptorCopy.cLines; i++)
7378	if (pRenameState->DescriptorCopy.aLines[i])
7379	RTStrFree(pRenameState->DescriptorCopy.aLines[i]);
7380	if (pRenameState->apszOldName)
7381	{
7382	for (unsigned i = 0; i <= pRenameState->cExtents; i++)
7383	if (pRenameState->apszOldName[i])
7384	RTStrFree(pRenameState->apszOldName[i]);
7385	RTMemTmpFree(pRenameState->apszOldName);
7386	}
7387	if (pRenameState->apszNewName)
7388	{
7389	for (unsigned i = 0; i <= pRenameState->cExtents; i++)
7390	if (pRenameState->apszNewName[i])
7391	RTStrFree(pRenameState->apszNewName[i]);
7392	RTMemTmpFree(pRenameState->apszNewName);
7393	}
7394	if (pRenameState->apszNewLines)
7395	{
7396	for (unsigned i = 0; i < pRenameState->cExtents; i++)
7397	if (pRenameState->apszNewLines[i])
7398	RTStrFree(pRenameState->apszNewLines[i]);
7399	RTMemTmpFree(pRenameState->apszNewLines);
7400	}
7401	if (pRenameState->pszOldDescName)
7402	RTStrFree(pRenameState->pszOldDescName);
7403	if (pRenameState->pszOldBaseName)
7404	RTStrFree(pRenameState->pszOldBaseName);
7405	if (pRenameState->pszNewBaseName)
7406	RTStrFree(pRenameState->pszNewBaseName);
7407	if (pRenameState->pszOldFullName)
7408	RTStrFree(pRenameState->pszOldFullName);
7409	if (pRenameState->pszNewFullName)
7410	RTStrFree(pRenameState->pszNewFullName);
7411	}
7412
7413	/**
7414	* Rolls back the rename operation to the original state.
7415	*
7416	* @returns VBox status code.
7417	* @param pImage VMDK image instance.
7418	* @param pRenameState The rename state.
7419	*/
7420	static int vmdkRenameRollback(PVMDKIMAGE pImage, PVMDKRENAMESTATE pRenameState)
7421	{
7422	int rc = VINF_SUCCESS;
7423
7424	if (!pRenameState->fImageFreed)
7425	{
7426	/*
7427	* Some extents may have been closed, close the rest. We will
7428	* re-open the whole thing later.
7429	*/
7430	vmdkFreeImage(pImage, false, true /fFlush/);
7431	}
7432
7433	/* Rename files back. */
7434	for (unsigned i = 0; i <= pRenameState->cExtents; i++)
7435	{
7436	if (pRenameState->apszOldName[i])
7437	{
7438	rc = vdIfIoIntFileMove(pImage->pIfIo, pRenameState->apszNewName[i], pRenameState->apszOldName[i], 0);
7439	AssertRC(rc);
7440	}
7441	}
7442	/* Restore the old descriptor. */
7443	PVMDKFILE pFile;
7444	rc = vmdkFileOpen(pImage, &pFile, NULL, pRenameState->pszOldDescName,
7445	VDOpenFlagsToFileOpenFlags(VD_OPEN_FLAGS_NORMAL,
7446	false /* fCreate */));
7447	AssertRC(rc);
7448	if (pRenameState->fEmbeddedDesc)
7449	{
7450	pRenameState->ExtentCopy.pFile = pFile;
7451	pImage->pExtents = &pRenameState->ExtentCopy;
7452	}
7453	else
7454	{
7455	/* Shouldn't be null for separate descriptor.
7456	* There will be no access to the actual content.
7457	*/
7458	pImage->pDescData = pRenameState->pszOldDescName;
7459	pImage->pFile = pFile;
7460	}
7461	pImage->Descriptor = pRenameState->DescriptorCopy;
7462	vmdkWriteDescriptor(pImage, NULL);
7463	vmdkFileClose(pImage, &pFile, false);
7464	/* Get rid of the stuff we implanted. */
7465	pImage->pExtents = NULL;
7466	pImage->pFile = NULL;
7467	pImage->pDescData = NULL;
7468	/* Re-open the image back. */
7469	pImage->pszFilename = pRenameState->pszOldImageName;
7470	rc = vmdkOpenImage(pImage, pImage->uOpenFlags);
7471
7472	return rc;
7473	}
7474
7475	/**
7476	* Rename worker doing the real work.
7477	*
7478	* @returns VBox status code.
7479	* @param pImage VMDK image instance.
7480	* @param pRenameState The rename state.
7481	* @param pszFilename The new filename.
7482	*/
7483	static int vmdkRenameWorker(PVMDKIMAGE pImage, PVMDKRENAMESTATE pRenameState, const char *pszFilename)
7484	{
7485	int rc = VINF_SUCCESS;
7486	unsigned i, line;
7487
7488	/* Update the descriptor with modified extent names. */
7489	for (i = 0, line = pImage->Descriptor.uFirstExtent;
7490	i < pRenameState->cExtents;
7491	i++, line = pImage->Descriptor.aNextLines[line])
7492	{
7493	/* Update the descriptor. */
7494	pRenameState->apszNewLines[i] = vmdkStrReplace(pImage->Descriptor.aLines[line],
7495	pRenameState->pszOldBaseName,
7496	pRenameState->pszNewBaseName);
7497	if (!pRenameState->apszNewLines[i])
7498	{
7499	rc = VERR_NO_MEMORY;
7500	break;
7501	}
7502	pImage->Descriptor.aLines[line] = pRenameState->apszNewLines[i];
7503	}
7504
7505	if (RT_SUCCESS(rc))
7506	{
7507	/* Make sure the descriptor gets written back. */
7508	pImage->Descriptor.fDirty = true;
7509	/* Flush the descriptor now, in case it is embedded. */
7510	vmdkFlushImage(pImage, NULL);
7511
7512	/* Close and rename/move extents. */
7513	for (i = 0; i < pRenameState->cExtents; i++)
7514	{
7515	PVMDKEXTENT pExtent = &pImage->pExtents[i];
7516	/* Compose new name for the extent. */
7517	pRenameState->apszNewName[i] = vmdkStrReplace(pExtent->pszFullname,
7518	pRenameState->pszOldFullName,
7519	pRenameState->pszNewFullName);
7520	if (!pRenameState->apszNewName[i])
7521	{
7522	rc = VERR_NO_MEMORY;
7523	break;
7524	}
7525	/* Close the extent file. */
7526	rc = vmdkFileClose(pImage, &pExtent->pFile, false);
7527	if (RT_FAILURE(rc))
7528	break;;
7529
7530	/* Rename the extent file. */
7531	rc = vdIfIoIntFileMove(pImage->pIfIo, pExtent->pszFullname, pRenameState->apszNewName[i], 0);
7532	if (RT_FAILURE(rc))
7533	break;
7534	/* Remember the old name. */
7535	pRenameState->apszOldName[i] = RTStrDup(pExtent->pszFullname);
7536	}
7537
7538	if (RT_SUCCESS(rc))
7539	{
7540	/* Release all old stuff. */
7541	rc = vmdkFreeImage(pImage, false, true /fFlush/);
7542	if (RT_SUCCESS(rc))
7543	{
7544	pRenameState->fImageFreed = true;
7545
7546	/* Last elements of new/old name arrays are intended for
7547	* storing descriptor's names.
7548	*/
7549	pRenameState->apszNewName[pRenameState->cExtents] = RTStrDup(pszFilename);
7550	/* Rename the descriptor file if it's separate. */
7551	if (!pRenameState->fEmbeddedDesc)
7552	{
7553	rc = vdIfIoIntFileMove(pImage->pIfIo, pImage->pszFilename, pRenameState->apszNewName[pRenameState->cExtents], 0);
7554	if (RT_SUCCESS(rc))
7555	{
7556	/* Save old name only if we may need to change it back. */
7557	pRenameState->apszOldName[pRenameState->cExtents] = RTStrDup(pszFilename);
7558	}
7559	}
7560
7561	/* Update pImage with the new information. */
7562	pImage->pszFilename = pszFilename;
7563
7564	/* Open the new image. */
7565	rc = vmdkOpenImage(pImage, pImage->uOpenFlags);
7566	}
7567	}
7568	}
7569
7570	return rc;
7571	}
7572
7573	/** @copydoc VDIMAGEBACKEND::pfnRename */
7574	static DECLCALLBACK(int) vmdkRename(void pBackendData, const char pszFilename)
7575	{
7576	LogFlowFunc(("pBackendData=%#p pszFilename=%#p\n", pBackendData, pszFilename));
7577
7578	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7579	VMDKRENAMESTATE RenameState;
7580
7581	memset(&RenameState, 0, sizeof(RenameState));
7582
7583	/* Check arguments. */
7584	AssertPtrReturn(pImage, VERR_INVALID_POINTER);
7585	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
7586	AssertReturn(*pszFilename != '\0', VERR_INVALID_PARAMETER);
7587	AssertReturn(!(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK), VERR_INVALID_PARAMETER);
7588
7589	int rc = vmdkRenameStatePrepare(pImage, &RenameState, pszFilename);
7590	if (RT_SUCCESS(rc))
7591	{
7592	/* --- Up to this point we have not done any damage yet. --- */
7593
7594	rc = vmdkRenameWorker(pImage, &RenameState, pszFilename);
7595	/* Roll back all changes in case of failure. */
7596	if (RT_FAILURE(rc))
7597	{
7598	int rrc = vmdkRenameRollback(pImage, &RenameState);
7599	AssertRC(rrc);
7600	}
7601	}
7602
7603	vmdkRenameStateDestroy(&RenameState);
7604	LogFlowFunc(("returns %Rrc\n", rc));
7605	return rc;
7606	}
7607
7608	/** @copydoc VDIMAGEBACKEND::pfnClose */
7609	static DECLCALLBACK(int) vmdkClose(void *pBackendData, bool fDelete)
7610	{
7611	LogFlowFunc(("pBackendData=%#p fDelete=%d\n", pBackendData, fDelete));
7612	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7613
7614	int rc = vmdkFreeImage(pImage, fDelete, true /fFlush/);
7615	RTMemFree(pImage);
7616
7617	LogFlowFunc(("returns %Rrc\n", rc));
7618	return rc;
7619	}
7620
7621	/** @copydoc VDIMAGEBACKEND::pfnRead */
7622	static DECLCALLBACK(int) vmdkRead(void *pBackendData, uint64_t uOffset, size_t cbToRead,
7623	PVDIOCTX pIoCtx, size_t *pcbActuallyRead)
7624	{
7625	LogFlowFunc(("pBackendData=%#p uOffset=%llu pIoCtx=%#p cbToRead=%zu pcbActuallyRead=%#p\n",
7626	pBackendData, uOffset, pIoCtx, cbToRead, pcbActuallyRead));
7627	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7628
7629	AssertPtr(pImage);
7630	Assert(uOffset % 512 == 0);
7631	Assert(cbToRead % 512 == 0);
7632	AssertPtrReturn(pIoCtx, VERR_INVALID_POINTER);
7633	AssertReturn(cbToRead, VERR_INVALID_PARAMETER);
7634	AssertReturn(uOffset + cbToRead <= pImage->cbSize, VERR_INVALID_PARAMETER);
7635
7636	/* Find the extent and check access permissions as defined in the extent descriptor. */
7637	PVMDKEXTENT pExtent;
7638	uint64_t uSectorExtentRel;
7639	int rc = vmdkFindExtent(pImage, VMDK_BYTE2SECTOR(uOffset),
7640	&pExtent, &uSectorExtentRel);
7641	if ( RT_SUCCESS(rc)
7642	&& pExtent->enmAccess != VMDKACCESS_NOACCESS)
7643	{
7644	/* Clip read range to remain in this extent. */
7645	cbToRead = RT_MIN(cbToRead, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
7646
7647	/* Handle the read according to the current extent type. */
7648	switch (pExtent->enmType)
7649	{
7650	case VMDKETYPE_HOSTED_SPARSE:
7651	{
7652	uint64_t uSectorExtentAbs;
7653
7654	rc = vmdkGetSector(pImage, pIoCtx, pExtent, uSectorExtentRel, &uSectorExtentAbs);
7655	if (RT_FAILURE(rc))
7656	break;
7657	/* Clip read range to at most the rest of the grain. */
7658	cbToRead = RT_MIN(cbToRead, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain - uSectorExtentRel % pExtent->cSectorsPerGrain));
7659	Assert(!(cbToRead % 512));
7660	if (uSectorExtentAbs == 0)
7661	{
7662	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7663	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
7664	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
7665	rc = VERR_VD_BLOCK_FREE;
7666	else
7667	rc = vmdkStreamReadSequential(pImage, pExtent,
7668	uSectorExtentRel,
7669	pIoCtx, cbToRead);
7670	}
7671	else
7672	{
7673	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7674	{
7675	AssertMsg(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
7676	("Async I/O is not supported for stream optimized VMDK's\n"));
7677
7678	uint32_t uSectorInGrain = uSectorExtentRel % pExtent->cSectorsPerGrain;
7679	uSectorExtentAbs -= uSectorInGrain;
7680	if (pExtent->uGrainSectorAbs != uSectorExtentAbs)
7681	{
7682	uint64_t uLBA = 0; /* gcc maybe uninitialized */
7683	rc = vmdkFileInflateSync(pImage, pExtent,
7684	VMDK_SECTOR2BYTE(uSectorExtentAbs),
7685	pExtent->pvGrain,
7686	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
7687	NULL, &uLBA, NULL);
7688	if (RT_FAILURE(rc))
7689	{
7690	pExtent->uGrainSectorAbs = 0;
7691	break;
7692	}
7693	pExtent->uGrainSectorAbs = uSectorExtentAbs;
7694	pExtent->uGrain = uSectorExtentRel / pExtent->cSectorsPerGrain;
7695	Assert(uLBA == uSectorExtentRel);
7696	}
7697	vdIfIoIntIoCtxCopyTo(pImage->pIfIo, pIoCtx,
7698	(uint8_t *)pExtent->pvGrain
7699	+ VMDK_SECTOR2BYTE(uSectorInGrain),
7700	cbToRead);
7701	}
7702	else
7703	rc = vdIfIoIntFileReadUser(pImage->pIfIo, pExtent->pFile->pStorage,
7704	VMDK_SECTOR2BYTE(uSectorExtentAbs),
7705	pIoCtx, cbToRead);
7706	}
7707	break;
7708	}
7709	case VMDKETYPE_VMFS:
7710	case VMDKETYPE_FLAT:
7711	rc = vdIfIoIntFileReadUser(pImage->pIfIo, pExtent->pFile->pStorage,
7712	VMDK_SECTOR2BYTE(uSectorExtentRel),
7713	pIoCtx, cbToRead);
7714	break;
7715	case VMDKETYPE_ZERO:
7716	{
7717	size_t cbSet;
7718
7719	cbSet = vdIfIoIntIoCtxSet(pImage->pIfIo, pIoCtx, 0, cbToRead);
7720	Assert(cbSet == cbToRead);
7721	break;
7722	}
7723	}
7724	if (pcbActuallyRead)
7725	*pcbActuallyRead = cbToRead;
7726	}
7727	else if (RT_SUCCESS(rc))
7728	rc = VERR_VD_VMDK_INVALID_STATE;
7729
7730	LogFlowFunc(("returns %Rrc\n", rc));
7731	return rc;
7732	}
7733
7734	/** @copydoc VDIMAGEBACKEND::pfnWrite */
7735	static DECLCALLBACK(int) vmdkWrite(void *pBackendData, uint64_t uOffset, size_t cbToWrite,
7736	PVDIOCTX pIoCtx, size_t pcbWriteProcess, size_t pcbPreRead,
7737	size_t *pcbPostRead, unsigned fWrite)
7738	{
7739	LogFlowFunc(("pBackendData=%#p uOffset=%llu pIoCtx=%#p cbToWrite=%zu pcbWriteProcess=%#p pcbPreRead=%#p pcbPostRead=%#p\n",
7740	pBackendData, uOffset, pIoCtx, cbToWrite, pcbWriteProcess, pcbPreRead, pcbPostRead));
7741	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7742	int rc;
7743
7744	AssertPtr(pImage);
7745	Assert(uOffset % 512 == 0);
7746	Assert(cbToWrite % 512 == 0);
7747	AssertPtrReturn(pIoCtx, VERR_INVALID_POINTER);
7748	AssertReturn(cbToWrite, VERR_INVALID_PARAMETER);
7749
7750	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7751	{
7752	PVMDKEXTENT pExtent;
7753	uint64_t uSectorExtentRel;
7754	uint64_t uSectorExtentAbs;
7755
7756	/* No size check here, will do that later when the extent is located.
7757	* There are sparse images out there which according to the spec are
7758	* invalid, because the total size is not a multiple of the grain size.
7759	* Also for sparse images which are stitched together in odd ways (not at
7760	* grain boundaries, and with the nominal size not being a multiple of the
7761	* grain size), this would prevent writing to the last grain. */
7762
7763	rc = vmdkFindExtent(pImage, VMDK_BYTE2SECTOR(uOffset),
7764	&pExtent, &uSectorExtentRel);
7765	if (RT_SUCCESS(rc))
7766	{
7767	if ( pExtent->enmAccess != VMDKACCESS_READWRITE
7768	&& ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7769	&& !pImage->pExtents[0].uAppendPosition
7770	&& pExtent->enmAccess != VMDKACCESS_READONLY))
7771	rc = VERR_VD_VMDK_INVALID_STATE;
7772	else
7773	{
7774	/* Handle the write according to the current extent type. */
7775	switch (pExtent->enmType)
7776	{
7777	case VMDKETYPE_HOSTED_SPARSE:
7778	rc = vmdkGetSector(pImage, pIoCtx, pExtent, uSectorExtentRel, &uSectorExtentAbs);
7779	if (RT_SUCCESS(rc))
7780	{
7781	if ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
7782	&& uSectorExtentRel < (uint64_t)pExtent->uLastGrainAccess * pExtent->cSectorsPerGrain)
7783	rc = VERR_VD_VMDK_INVALID_WRITE;
7784	else
7785	{
7786	/* Clip write range to at most the rest of the grain. */
7787	cbToWrite = RT_MIN(cbToWrite,
7788	VMDK_SECTOR2BYTE( pExtent->cSectorsPerGrain
7789	- uSectorExtentRel % pExtent->cSectorsPerGrain));
7790	if (uSectorExtentAbs == 0)
7791	{
7792	if (!(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7793	{
7794	if (cbToWrite == VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
7795	{
7796	/* Full block write to a previously unallocated block.
7797	* Check if the caller wants to avoid the automatic alloc. */
7798	if (!(fWrite & VD_WRITE_NO_ALLOC))
7799	{
7800	/* Allocate GT and find out where to store the grain. */
7801	rc = vmdkAllocGrain(pImage, pExtent, pIoCtx,
7802	uSectorExtentRel, cbToWrite);
7803	}
7804	else
7805	rc = VERR_VD_BLOCK_FREE;
7806	*pcbPreRead = 0;
7807	*pcbPostRead = 0;
7808	}
7809	else
7810	{
7811	/* Clip write range to remain in this extent. */
7812	cbToWrite = RT_MIN(cbToWrite,
7813	VMDK_SECTOR2BYTE( pExtent->uSectorOffset
7814	+ pExtent->cNominalSectors - uSectorExtentRel));
7815	*pcbPreRead = VMDK_SECTOR2BYTE(uSectorExtentRel % pExtent->cSectorsPerGrain);
7816	pcbPostRead = VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain) - cbToWrite - pcbPreRead;
7817	rc = VERR_VD_BLOCK_FREE;
7818	}
7819	}
7820	else
7821	rc = vmdkStreamAllocGrain(pImage, pExtent, uSectorExtentRel,
7822	pIoCtx, cbToWrite);
7823	}
7824	else
7825	{
7826	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7827	{
7828	/* A partial write to a streamOptimized image is simply
7829	* invalid. It requires rewriting already compressed data
7830	* which is somewhere between expensive and impossible. */
7831	rc = VERR_VD_VMDK_INVALID_STATE;
7832	pExtent->uGrainSectorAbs = 0;
7833	AssertRC(rc);
7834	}
7835	else
7836	{
7837	Assert(!(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED));
7838	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
7839	VMDK_SECTOR2BYTE(uSectorExtentAbs),
7840	pIoCtx, cbToWrite, NULL, NULL);
7841	}
7842	}
7843	}
7844	}
7845	break;
7846	case VMDKETYPE_VMFS:
7847	case VMDKETYPE_FLAT:
7848	/* Clip write range to remain in this extent. */
7849	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
7850	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
7851	VMDK_SECTOR2BYTE(uSectorExtentRel),
7852	pIoCtx, cbToWrite, NULL, NULL);
7853	break;
7854	case VMDKETYPE_ZERO:
7855	/* Clip write range to remain in this extent. */
7856	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
7857	break;
7858	}
7859	}
7860
7861	if (pcbWriteProcess)
7862	*pcbWriteProcess = cbToWrite;
7863	}
7864	}
7865	else
7866	rc = VERR_VD_IMAGE_READ_ONLY;
7867
7868	LogFlowFunc(("returns %Rrc\n", rc));
7869	return rc;
7870	}
7871
7872	/** @copydoc VDIMAGEBACKEND::pfnFlush */
7873	static DECLCALLBACK(int) vmdkFlush(void *pBackendData, PVDIOCTX pIoCtx)
7874	{
7875	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7876
7877	return vmdkFlushImage(pImage, pIoCtx);
7878	}
7879
7880	/** @copydoc VDIMAGEBACKEND::pfnGetVersion */
7881	static DECLCALLBACK(unsigned) vmdkGetVersion(void *pBackendData)
7882	{
7883	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
7884	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7885
7886	AssertPtrReturn(pImage, 0);
7887
7888	return VMDK_IMAGE_VERSION;
7889	}
7890
7891	/** @copydoc VDIMAGEBACKEND::pfnGetFileSize */
7892	static DECLCALLBACK(uint64_t) vmdkGetFileSize(void *pBackendData)
7893	{
7894	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
7895	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7896	uint64_t cb = 0;
7897
7898	AssertPtrReturn(pImage, 0);
7899
7900	if (pImage->pFile != NULL)
7901	{
7902	uint64_t cbFile;
7903	int rc = vdIfIoIntFileGetSize(pImage->pIfIo, pImage->pFile->pStorage, &cbFile);
7904	if (RT_SUCCESS(rc))
7905	cb += cbFile;
7906	}
7907	for (unsigned i = 0; i < pImage->cExtents; i++)
7908	{
7909	if (pImage->pExtents[i].pFile != NULL)
7910	{
7911	uint64_t cbFile;
7912	int rc = vdIfIoIntFileGetSize(pImage->pIfIo, pImage->pExtents[i].pFile->pStorage, &cbFile);
7913	if (RT_SUCCESS(rc))
7914	cb += cbFile;
7915	}
7916	}
7917
7918	LogFlowFunc(("returns %lld\n", cb));
7919	return cb;
7920	}
7921
7922	/** @copydoc VDIMAGEBACKEND::pfnGetPCHSGeometry */
7923	static DECLCALLBACK(int) vmdkGetPCHSGeometry(void *pBackendData, PVDGEOMETRY pPCHSGeometry)
7924	{
7925	LogFlowFunc(("pBackendData=%#p pPCHSGeometry=%#p\n", pBackendData, pPCHSGeometry));
7926	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7927	int rc = VINF_SUCCESS;
7928
7929	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7930
7931	if (pImage->PCHSGeometry.cCylinders)
7932	*pPCHSGeometry = pImage->PCHSGeometry;
7933	else
7934	rc = VERR_VD_GEOMETRY_NOT_SET;
7935
7936	LogFlowFunc(("returns %Rrc (PCHS=%u/%u/%u)\n", rc, pPCHSGeometry->cCylinders, pPCHSGeometry->cHeads, pPCHSGeometry->cSectors));
7937	return rc;
7938	}
7939
7940	/** @copydoc VDIMAGEBACKEND::pfnSetPCHSGeometry */
7941	static DECLCALLBACK(int) vmdkSetPCHSGeometry(void *pBackendData, PCVDGEOMETRY pPCHSGeometry)
7942	{
7943	LogFlowFunc(("pBackendData=%#p pPCHSGeometry=%#p PCHS=%u/%u/%u\n",
7944	pBackendData, pPCHSGeometry, pPCHSGeometry->cCylinders, pPCHSGeometry->cHeads, pPCHSGeometry->cSectors));
7945	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7946	int rc = VINF_SUCCESS;
7947
7948	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7949
7950	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7951	{
7952	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7953	{
7954	rc = vmdkDescSetPCHSGeometry(pImage, pPCHSGeometry);
7955	if (RT_SUCCESS(rc))
7956	pImage->PCHSGeometry = *pPCHSGeometry;
7957	}
7958	else
7959	rc = VERR_NOT_SUPPORTED;
7960	}
7961	else
7962	rc = VERR_VD_IMAGE_READ_ONLY;
7963
7964	LogFlowFunc(("returns %Rrc\n", rc));
7965	return rc;
7966	}
7967
7968	/** @copydoc VDIMAGEBACKEND::pfnGetLCHSGeometry */
7969	static DECLCALLBACK(int) vmdkGetLCHSGeometry(void *pBackendData, PVDGEOMETRY pLCHSGeometry)
7970	{
7971	LogFlowFunc(("pBackendData=%#p pLCHSGeometry=%#p\n", pBackendData, pLCHSGeometry));
7972	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7973	int rc = VINF_SUCCESS;
7974
7975	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7976
7977	if (pImage->LCHSGeometry.cCylinders)
7978	*pLCHSGeometry = pImage->LCHSGeometry;
7979	else
7980	rc = VERR_VD_GEOMETRY_NOT_SET;
7981
7982	LogFlowFunc(("returns %Rrc (LCHS=%u/%u/%u)\n", rc, pLCHSGeometry->cCylinders, pLCHSGeometry->cHeads, pLCHSGeometry->cSectors));
7983	return rc;
7984	}
7985
7986	/** @copydoc VDIMAGEBACKEND::pfnSetLCHSGeometry */
7987	static DECLCALLBACK(int) vmdkSetLCHSGeometry(void *pBackendData, PCVDGEOMETRY pLCHSGeometry)
7988	{
7989	LogFlowFunc(("pBackendData=%#p pLCHSGeometry=%#p LCHS=%u/%u/%u\n",
7990	pBackendData, pLCHSGeometry, pLCHSGeometry->cCylinders, pLCHSGeometry->cHeads, pLCHSGeometry->cSectors));
7991	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7992	int rc = VINF_SUCCESS;
7993
7994	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7995
7996	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7997	{
7998	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7999	{
8000	rc = vmdkDescSetLCHSGeometry(pImage, pLCHSGeometry);
8001	if (RT_SUCCESS(rc))
8002	pImage->LCHSGeometry = *pLCHSGeometry;
8003	}
8004	else
8005	rc = VERR_NOT_SUPPORTED;
8006	}
8007	else
8008	rc = VERR_VD_IMAGE_READ_ONLY;
8009
8010	LogFlowFunc(("returns %Rrc\n", rc));
8011	return rc;
8012	}
8013
8014	/** @copydoc VDIMAGEBACKEND::pfnQueryRegions */
8015	static DECLCALLBACK(int) vmdkQueryRegions(void pBackendData, PCVDREGIONLIST ppRegionList)
8016	{
8017	LogFlowFunc(("pBackendData=%#p ppRegionList=%#p\n", pBackendData, ppRegionList));
8018	PVMDKIMAGE pThis = (PVMDKIMAGE)pBackendData;
8019
8020	AssertPtrReturn(pThis, VERR_VD_NOT_OPENED);
8021
8022	*ppRegionList = &pThis->RegionList;
8023	LogFlowFunc(("returns %Rrc\n", VINF_SUCCESS));
8024	return VINF_SUCCESS;
8025	}
8026
8027	/** @copydoc VDIMAGEBACKEND::pfnRegionListRelease */
8028	static DECLCALLBACK(void) vmdkRegionListRelease(void *pBackendData, PCVDREGIONLIST pRegionList)
8029	{
8030	RT_NOREF1(pRegionList);
8031	LogFlowFunc(("pBackendData=%#p pRegionList=%#p\n", pBackendData, pRegionList));
8032	PVMDKIMAGE pThis = (PVMDKIMAGE)pBackendData;
8033	AssertPtr(pThis); RT_NOREF(pThis);
8034
8035	/* Nothing to do here. */
8036	}
8037
8038	/** @copydoc VDIMAGEBACKEND::pfnGetImageFlags */
8039	static DECLCALLBACK(unsigned) vmdkGetImageFlags(void *pBackendData)
8040	{
8041	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
8042	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8043
8044	AssertPtrReturn(pImage, 0);
8045
8046	LogFlowFunc(("returns %#x\n", pImage->uImageFlags));
8047	return pImage->uImageFlags;
8048	}
8049
8050	/** @copydoc VDIMAGEBACKEND::pfnGetOpenFlags */
8051	static DECLCALLBACK(unsigned) vmdkGetOpenFlags(void *pBackendData)
8052	{
8053	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
8054	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8055
8056	AssertPtrReturn(pImage, 0);
8057
8058	LogFlowFunc(("returns %#x\n", pImage->uOpenFlags));
8059	return pImage->uOpenFlags;
8060	}
8061
8062	/** @copydoc VDIMAGEBACKEND::pfnSetOpenFlags */
8063	static DECLCALLBACK(int) vmdkSetOpenFlags(void *pBackendData, unsigned uOpenFlags)
8064	{
8065	LogFlowFunc(("pBackendData=%#p uOpenFlags=%#x\n", pBackendData, uOpenFlags));
8066	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8067	int rc;
8068
8069	/* Image must be opened and the new flags must be valid. */
8070	if (!pImage \|\| (uOpenFlags & ~( VD_OPEN_FLAGS_READONLY \| VD_OPEN_FLAGS_INFO
8071	\| VD_OPEN_FLAGS_ASYNC_IO \| VD_OPEN_FLAGS_SHAREABLE
8072	\| VD_OPEN_FLAGS_SEQUENTIAL \| VD_OPEN_FLAGS_SKIP_CONSISTENCY_CHECKS)))
8073	rc = VERR_INVALID_PARAMETER;
8074	else
8075	{
8076	/* StreamOptimized images need special treatment: reopen is prohibited. */
8077	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
8078	{
8079	if (pImage->uOpenFlags == uOpenFlags)
8080	rc = VINF_SUCCESS;
8081	else
8082	rc = VERR_INVALID_PARAMETER;
8083	}
8084	else
8085	{
8086	/* Implement this operation via reopening the image. */
8087	vmdkFreeImage(pImage, false, true /fFlush/);
8088	rc = vmdkOpenImage(pImage, uOpenFlags);
8089	}
8090	}
8091
8092	LogFlowFunc(("returns %Rrc\n", rc));
8093	return rc;
8094	}
8095
8096	/** @copydoc VDIMAGEBACKEND::pfnGetComment */
8097	static DECLCALLBACK(int) vmdkGetComment(void pBackendData, char pszComment, size_t cbComment)
8098	{
8099	LogFlowFunc(("pBackendData=%#p pszComment=%#p cbComment=%zu\n", pBackendData, pszComment, cbComment));
8100	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8101
8102	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8103
8104	char *pszCommentEncoded = NULL;
8105	int rc = vmdkDescDDBGetStr(pImage, &pImage->Descriptor,
8106	"ddb.comment", &pszCommentEncoded);
8107	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
8108	{
8109	pszCommentEncoded = NULL;
8110	rc = VINF_SUCCESS;
8111	}
8112
8113	if (RT_SUCCESS(rc))
8114	{
8115	if (pszComment && pszCommentEncoded)
8116	rc = vmdkDecodeString(pszCommentEncoded, pszComment, cbComment);
8117	else if (pszComment)
8118	*pszComment = '\0';
8119
8120	if (pszCommentEncoded)
8121	RTMemTmpFree(pszCommentEncoded);
8122	}
8123
8124	LogFlowFunc(("returns %Rrc comment='%s'\n", rc, pszComment));
8125	return rc;
8126	}
8127
8128	/** @copydoc VDIMAGEBACKEND::pfnSetComment */
8129	static DECLCALLBACK(int) vmdkSetComment(void pBackendData, const char pszComment)
8130	{
8131	LogFlowFunc(("pBackendData=%#p pszComment=\"%s\"\n", pBackendData, pszComment));
8132	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8133	int rc;
8134
8135	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8136
8137	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8138	{
8139	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8140	rc = vmdkSetImageComment(pImage, pszComment);
8141	else
8142	rc = VERR_NOT_SUPPORTED;
8143	}
8144	else
8145	rc = VERR_VD_IMAGE_READ_ONLY;
8146
8147	LogFlowFunc(("returns %Rrc\n", rc));
8148	return rc;
8149	}
8150
8151	/** @copydoc VDIMAGEBACKEND::pfnGetUuid */
8152	static DECLCALLBACK(int) vmdkGetUuid(void *pBackendData, PRTUUID pUuid)
8153	{
8154	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
8155	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8156
8157	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8158
8159	*pUuid = pImage->ImageUuid;
8160
8161	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
8162	return VINF_SUCCESS;
8163	}
8164
8165	/** @copydoc VDIMAGEBACKEND::pfnSetUuid */
8166	static DECLCALLBACK(int) vmdkSetUuid(void *pBackendData, PCRTUUID pUuid)
8167	{
8168	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
8169	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8170	int rc = VINF_SUCCESS;
8171
8172	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8173
8174	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8175	{
8176	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8177	{
8178	pImage->ImageUuid = *pUuid;
8179	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
8180	VMDK_DDB_IMAGE_UUID, pUuid);
8181	if (RT_FAILURE(rc))
8182	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
8183	N_("VMDK: error storing image UUID in descriptor in '%s'"), pImage->pszFilename);
8184	}
8185	else
8186	rc = VERR_NOT_SUPPORTED;
8187	}
8188	else
8189	rc = VERR_VD_IMAGE_READ_ONLY;
8190
8191	LogFlowFunc(("returns %Rrc\n", rc));
8192	return rc;
8193	}
8194
8195	/** @copydoc VDIMAGEBACKEND::pfnGetModificationUuid */
8196	static DECLCALLBACK(int) vmdkGetModificationUuid(void *pBackendData, PRTUUID pUuid)
8197	{
8198	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
8199	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8200
8201	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8202
8203	*pUuid = pImage->ModificationUuid;
8204
8205	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
8206	return VINF_SUCCESS;
8207	}
8208
8209	/** @copydoc VDIMAGEBACKEND::pfnSetModificationUuid */
8210	static DECLCALLBACK(int) vmdkSetModificationUuid(void *pBackendData, PCRTUUID pUuid)
8211	{
8212	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
8213	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8214	int rc = VINF_SUCCESS;
8215
8216	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8217
8218	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8219	{
8220	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8221	{
8222	/* Only touch the modification uuid if it changed. */
8223	if (RTUuidCompare(&pImage->ModificationUuid, pUuid))
8224	{
8225	pImage->ModificationUuid = *pUuid;
8226	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
8227	VMDK_DDB_MODIFICATION_UUID, pUuid);
8228	if (RT_FAILURE(rc))
8229	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing modification UUID in descriptor in '%s'"), pImage->pszFilename);
8230	}
8231	}
8232	else
8233	rc = VERR_NOT_SUPPORTED;
8234	}
8235	else
8236	rc = VERR_VD_IMAGE_READ_ONLY;
8237
8238	LogFlowFunc(("returns %Rrc\n", rc));
8239	return rc;
8240	}
8241
8242	/** @copydoc VDIMAGEBACKEND::pfnGetParentUuid */
8243	static DECLCALLBACK(int) vmdkGetParentUuid(void *pBackendData, PRTUUID pUuid)
8244	{
8245	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
8246	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8247
8248	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8249
8250	*pUuid = pImage->ParentUuid;
8251
8252	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
8253	return VINF_SUCCESS;
8254	}
8255
8256	/** @copydoc VDIMAGEBACKEND::pfnSetParentUuid */
8257	static DECLCALLBACK(int) vmdkSetParentUuid(void *pBackendData, PCRTUUID pUuid)
8258	{
8259	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
8260	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8261	int rc = VINF_SUCCESS;
8262
8263	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8264
8265	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8266	{
8267	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8268	{
8269	pImage->ParentUuid = *pUuid;
8270	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
8271	VMDK_DDB_PARENT_UUID, pUuid);
8272	if (RT_FAILURE(rc))
8273	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
8274	N_("VMDK: error storing parent image UUID in descriptor in '%s'"), pImage->pszFilename);
8275	}
8276	else
8277	rc = VERR_NOT_SUPPORTED;
8278	}
8279	else
8280	rc = VERR_VD_IMAGE_READ_ONLY;
8281
8282	LogFlowFunc(("returns %Rrc\n", rc));
8283	return rc;
8284	}
8285
8286	/** @copydoc VDIMAGEBACKEND::pfnGetParentModificationUuid */
8287	static DECLCALLBACK(int) vmdkGetParentModificationUuid(void *pBackendData, PRTUUID pUuid)
8288	{
8289	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
8290	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8291
8292	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8293
8294	*pUuid = pImage->ParentModificationUuid;
8295
8296	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
8297	return VINF_SUCCESS;
8298	}
8299
8300	/** @copydoc VDIMAGEBACKEND::pfnSetParentModificationUuid */
8301	static DECLCALLBACK(int) vmdkSetParentModificationUuid(void *pBackendData, PCRTUUID pUuid)
8302	{
8303	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
8304	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8305	int rc = VINF_SUCCESS;
8306
8307	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8308
8309	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8310	{
8311	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8312	{
8313	pImage->ParentModificationUuid = *pUuid;
8314	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
8315	VMDK_DDB_PARENT_MODIFICATION_UUID, pUuid);
8316	if (RT_FAILURE(rc))
8317	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent image UUID in descriptor in '%s'"), pImage->pszFilename);
8318	}
8319	else
8320	rc = VERR_NOT_SUPPORTED;
8321	}
8322	else
8323	rc = VERR_VD_IMAGE_READ_ONLY;
8324
8325	LogFlowFunc(("returns %Rrc\n", rc));
8326	return rc;
8327	}
8328
8329	/** @copydoc VDIMAGEBACKEND::pfnDump */
8330	static DECLCALLBACK(void) vmdkDump(void *pBackendData)
8331	{
8332	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8333
8334	AssertPtrReturnVoid(pImage);
8335	vdIfErrorMessage(pImage->pIfError, "Header: Geometry PCHS=%u/%u/%u LCHS=%u/%u/%u cbSector=%llu\n",
8336	pImage->PCHSGeometry.cCylinders, pImage->PCHSGeometry.cHeads, pImage->PCHSGeometry.cSectors,
8337	pImage->LCHSGeometry.cCylinders, pImage->LCHSGeometry.cHeads, pImage->LCHSGeometry.cSectors,
8338	VMDK_BYTE2SECTOR(pImage->cbSize));
8339	vdIfErrorMessage(pImage->pIfError, "Header: uuidCreation={%RTuuid}\n", &pImage->ImageUuid);
8340	vdIfErrorMessage(pImage->pIfError, "Header: uuidModification={%RTuuid}\n", &pImage->ModificationUuid);
8341	vdIfErrorMessage(pImage->pIfError, "Header: uuidParent={%RTuuid}\n", &pImage->ParentUuid);
8342	vdIfErrorMessage(pImage->pIfError, "Header: uuidParentModification={%RTuuid}\n", &pImage->ParentModificationUuid);
8343	}
8344
8345
8346	/**
8347	* Returns the size, in bytes, of the sparse extent overhead for
8348	* the number of desired total sectors and based on the current
8349	* sectors of the extent.
8350	*
8351	* @returns uint64_t size of new overhead in bytes.
8352	* @param pExtent VMDK extent instance.
8353	* @param cSectorsNew Number of desired total sectors.
8354	*/
8355	static uint64_t vmdkGetNewOverhead(PVMDKEXTENT pExtent, uint64_t cSectorsNew)
8356	{
8357	uint64_t cNewDirEntries = cSectorsNew / pExtent->cSectorsPerGDE;
8358	if (cSectorsNew % pExtent->cSectorsPerGDE)
8359	cNewDirEntries++;
8360
8361	size_t cbNewGD = cNewDirEntries * sizeof(uint32_t);
8362	uint64_t cbNewDirSize = RT_ALIGN_64(cbNewGD, 512);
8363	uint64_t cbNewAllTablesSize = RT_ALIGN_64(cNewDirEntries * pExtent->cGTEntries * sizeof(uint32_t), 512);
8364	uint64_t cbNewOverhead = RT_ALIGN_Z(RT_MAX(pExtent->uDescriptorSector
8365	+ pExtent->cDescriptorSectors, 1)
8366	+ cbNewDirSize + cbNewAllTablesSize, 512);
8367	cbNewOverhead += cbNewDirSize + cbNewAllTablesSize;
8368	cbNewOverhead = RT_ALIGN_64(cbNewOverhead,
8369	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
8370
8371	return cbNewOverhead;
8372	}
8373
8374	/**
8375	* Internal: Replaces the size (in sectors) of an extent in the descriptor file.
8376	*
8377	* @returns VBox status code.
8378	* @param pImage VMDK image instance.
8379	* @param pExtent VMDK extent instance.
8380	* @param uLine Line number of descriptor to change.
8381	* @param cSectorsOld Existing number of sectors.
8382	* @param cSectorsNew New number of sectors.
8383	*/
8384	static int vmdkReplaceExtentSize(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, unsigned uLine, uint64_t cSectorsOld,
8385	uint64_t cSectorsNew)
8386	{
8387	char szOldExtentSectors[UINT64_MAX_BUFF_SIZE];
8388	char szNewExtentSectors[UINT64_MAX_BUFF_SIZE];
8389
8390	ssize_t cbWritten = RTStrPrintf2(szOldExtentSectors, sizeof(szOldExtentSectors), "%llu", cSectorsOld);
8391	if (cbWritten <= 0 \|\| cbWritten > (ssize_t)sizeof(szOldExtentSectors))
8392	return VERR_BUFFER_OVERFLOW;
8393
8394	cbWritten = RTStrPrintf2(szNewExtentSectors, sizeof(szNewExtentSectors), "%llu", cSectorsNew);
8395	if (cbWritten <= 0 \|\| cbWritten > (ssize_t)sizeof(szNewExtentSectors))
8396	return VERR_BUFFER_OVERFLOW;
8397
8398	char *pszNewExtentLine = vmdkStrReplace(pImage->Descriptor.aLines[uLine],
8399	szOldExtentSectors,
8400	szNewExtentSectors);
8401
8402	if (RT_UNLIKELY(!pszNewExtentLine))
8403	return VERR_INVALID_PARAMETER;
8404
8405	vmdkDescExtRemoveByLine(pImage, &pImage->Descriptor, uLine);
8406	vmdkDescExtInsert(pImage, &pImage->Descriptor,
8407	pExtent->enmAccess, cSectorsNew,
8408	pExtent->enmType, pExtent->pszBasename, pExtent->uSectorOffset);
8409
8410	RTStrFree(pszNewExtentLine);
8411	pszNewExtentLine = NULL;
8412
8413	pImage->Descriptor.fDirty = true;
8414
8415	return VINF_SUCCESS;
8416	}
8417
8418	/**
8419	* Moves sectors down to make room for new overhead.
8420	* Used for sparse extent resize.
8421	*
8422	* @returns VBox status code.
8423	* @param pImage VMDK image instance.
8424	* @param pExtent VMDK extent instance.
8425	* @param cSectorsNew Number of sectors after resize.
8426	*/
8427	static int vmdkRelocateSectorsForSparseResize(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
8428	uint64_t cSectorsNew)
8429	{
8430	int rc = VINF_SUCCESS;
8431
8432	uint64_t cbNewOverhead = vmdkGetNewOverhead(pExtent, cSectorsNew);
8433
8434	uint64_t cNewOverheadSectors = VMDK_BYTE2SECTOR(cbNewOverhead);
8435	uint64_t cOverheadSectorDiff = cNewOverheadSectors - pExtent->cOverheadSectors;
8436
8437	uint64_t cbFile = 0;
8438	rc = vdIfIoIntFileGetSize(pImage->pIfIo, pExtent->pFile->pStorage, &cbFile);
8439
8440	uint64_t uNewAppendPosition;
8441
8442	/* Calculate how many sectors need to be relocated. */
8443	unsigned cSectorsReloc = cOverheadSectorDiff;
8444	if (cbNewOverhead % VMDK_SECTOR_SIZE)
8445	cSectorsReloc++;
8446
8447	if (cSectorsReloc < pExtent->cSectors)
8448	uNewAppendPosition = RT_ALIGN_Z(cbFile + VMDK_SECTOR2BYTE(cOverheadSectorDiff), 512);
8449	else
8450	uNewAppendPosition = cbFile;
8451
8452	/*
8453	* Get the blocks we need to relocate first, they are appended to the end
8454	* of the image.
8455	*/
8456	void pvBuf = NULL, pvZero = NULL;
8457	do
8458	{
8459	/* Allocate data buffer. */
8460	pvBuf = RTMemAllocZ(VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
8461	if (!pvBuf)
8462	{
8463	rc = VERR_NO_MEMORY;
8464	break;
8465	}
8466
8467	/* Allocate buffer for overwriting with zeroes. */
8468	pvZero = RTMemAllocZ(VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
8469	if (!pvZero)
8470	{
8471	RTMemFree(pvBuf);
8472	pvBuf = NULL;
8473
8474	rc = VERR_NO_MEMORY;
8475	break;
8476	}
8477
8478	uint32_t aGTDataTmp = (uint32_t )RTMemAllocZ(sizeof(uint32_t) * pExtent->cGTEntries);
8479	if(!aGTDataTmp)
8480	{
8481	RTMemFree(pvBuf);
8482	pvBuf = NULL;
8483
8484	RTMemFree(pvZero);
8485	pvZero = NULL;
8486
8487	rc = VERR_NO_MEMORY;
8488	break;
8489	}
8490
8491	uint32_t aRGTDataTmp = (uint32_t )RTMemAllocZ(sizeof(uint32_t) * pExtent->cGTEntries);
8492	if(!aRGTDataTmp)
8493	{
8494	RTMemFree(pvBuf);
8495	pvBuf = NULL;
8496
8497	RTMemFree(pvZero);
8498	pvZero = NULL;
8499
8500	RTMemFree(aGTDataTmp);
8501	aGTDataTmp = NULL;
8502
8503	rc = VERR_NO_MEMORY;
8504	break;
8505	}
8506
8507	/* Search for overlap sector in the grain table. */
8508	for (uint32_t idxGD = 0; idxGD < pExtent->cGDEntries; idxGD++)
8509	{
8510	uint64_t uGTSector = pExtent->pGD[idxGD];
8511	uint64_t uRGTSector = pExtent->pRGD[idxGD];
8512
8513	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
8514	VMDK_SECTOR2BYTE(uGTSector),
8515	aGTDataTmp, sizeof(uint32_t) * pExtent->cGTEntries);
8516
8517	if (RT_FAILURE(rc))
8518	break;
8519
8520	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
8521	VMDK_SECTOR2BYTE(uRGTSector),
8522	aRGTDataTmp, sizeof(uint32_t) * pExtent->cGTEntries);
8523
8524	if (RT_FAILURE(rc))
8525	break;
8526
8527	for (uint32_t idxGT = 0; idxGT < pExtent->cGTEntries; idxGT++)
8528	{
8529	uint64_t aGTEntryLE = RT_LE2H_U64(aGTDataTmp[idxGT]);
8530	uint64_t aRGTEntryLE = RT_LE2H_U64(aRGTDataTmp[idxGT]);
8531
8532	/**
8533	* Check if grain table is valid. If not dump out with an error.
8534	* Shoudln't ever get here (given other checks) but good sanity check.
8535	*/
8536	if (aGTEntryLE != aRGTEntryLE)
8537	{
8538	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
8539	N_("VMDK: inconsistent references within grain table in '%s'"), pExtent->pszFullname);
8540	break;
8541	}
8542
8543	if (aGTEntryLE < cNewOverheadSectors
8544	&& aGTEntryLE != 0)
8545	{
8546	/* Read data and append grain to the end of the image. */
8547	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
8548	VMDK_SECTOR2BYTE(aGTEntryLE), pvBuf,
8549	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
8550	if (RT_FAILURE(rc))
8551	break;
8552
8553	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8554	uNewAppendPosition, pvBuf,
8555	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
8556	if (RT_FAILURE(rc))
8557	break;
8558
8559	/* Zero out the old block area. */
8560	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8561	VMDK_SECTOR2BYTE(aGTEntryLE), pvZero,
8562	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
8563	if (RT_FAILURE(rc))
8564	break;
8565
8566	/* Write updated grain tables to file */
8567	aGTDataTmp[idxGT] = VMDK_BYTE2SECTOR(uNewAppendPosition);
8568	aRGTDataTmp[idxGT] = VMDK_BYTE2SECTOR(uNewAppendPosition);
8569
8570	if (memcmp(aGTDataTmp, aRGTDataTmp, sizeof(uint32_t) * pExtent->cGTEntries))
8571	{
8572	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
8573	N_("VMDK: inconsistency between grain table and backup grain table in '%s'"), pExtent->pszFullname);
8574	break;
8575	}
8576
8577	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8578	VMDK_SECTOR2BYTE(uGTSector),
8579	aGTDataTmp, sizeof(uint32_t) * pExtent->cGTEntries);
8580
8581	if (RT_FAILURE(rc))
8582	break;
8583
8584	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8585	VMDK_SECTOR2BYTE(uRGTSector),
8586	aRGTDataTmp, sizeof(uint32_t) * pExtent->cGTEntries);
8587
8588	break;
8589	}
8590	}
8591	}
8592
8593	RTMemFree(aGTDataTmp);
8594	aGTDataTmp = NULL;
8595
8596	RTMemFree(aRGTDataTmp);
8597	aRGTDataTmp = NULL;
8598
8599	if (RT_FAILURE(rc))
8600	break;
8601
8602	uNewAppendPosition += VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain);
8603	} while (0);
8604
8605	if (pvBuf)
8606	{
8607	RTMemFree(pvBuf);
8608	pvBuf = NULL;
8609	}
8610
8611	if (pvZero)
8612	{
8613	RTMemFree(pvZero);
8614	pvZero = NULL;
8615	}
8616
8617	// Update append position for extent
8618	pExtent->uAppendPosition = uNewAppendPosition;
8619
8620	return rc;
8621	}
8622
8623	/**
8624	* Resizes meta/overhead for sparse extent resize.
8625	*
8626	* @returns VBox status code.
8627	* @param pImage VMDK image instance.
8628	* @param pExtent VMDK extent instance.
8629	* @param cSectorsNew Number of sectors after resize.
8630	*/
8631	static int vmdkResizeSparseMeta(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
8632	uint64_t cSectorsNew)
8633	{
8634	int rc = VINF_SUCCESS;
8635	uint32_t cOldGDEntries = pExtent->cGDEntries;
8636
8637	uint64_t cNewDirEntries = cSectorsNew / pExtent->cSectorsPerGDE;
8638	if (cSectorsNew % pExtent->cSectorsPerGDE)
8639	cNewDirEntries++;
8640
8641	size_t cbNewGD = cNewDirEntries * sizeof(uint32_t);
8642
8643	uint64_t cbNewDirSize = RT_ALIGN_64(cbNewGD, 512);
8644	uint64_t cbCurrDirSize = RT_ALIGN_64(pExtent->cGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE, 512);
8645	uint64_t cDirSectorDiff = VMDK_BYTE2SECTOR(cbNewDirSize - cbCurrDirSize);
8646
8647	uint64_t cbNewAllTablesSize = RT_ALIGN_64(cNewDirEntries * pExtent->cGTEntries * sizeof(uint32_t), 512);
8648	uint64_t cbCurrAllTablesSize = RT_ALIGN_64(pExtent->cGDEntries * VMDK_GRAIN_TABLE_SIZE, 512);
8649	uint64_t cTableSectorDiff = VMDK_BYTE2SECTOR(cbNewAllTablesSize - cbCurrAllTablesSize);
8650
8651	uint64_t cbNewOverhead = vmdkGetNewOverhead(pExtent, cSectorsNew);
8652	uint64_t cNewOverheadSectors = VMDK_BYTE2SECTOR(cbNewOverhead);
8653	uint64_t cOverheadSectorDiff = cNewOverheadSectors - pExtent->cOverheadSectors;
8654
8655	/*
8656	* Get the blocks we need to relocate first, they are appended to the end
8657	* of the image.
8658	*/
8659	void pvBuf = NULL, pvZero = NULL;
8660
8661	do
8662	{
8663	/* Allocate data buffer. */
8664	pvBuf = RTMemAllocZ(VMDK_GRAIN_TABLE_SIZE);
8665	if (!pvBuf)
8666	{
8667	rc = VERR_NO_MEMORY;
8668	break;
8669	}
8670
8671	/* Allocate buffer for overwriting with zeroes. */
8672	pvZero = RTMemAllocZ(VMDK_GRAIN_TABLE_SIZE);
8673	if (!pvZero)
8674	{
8675	RTMemFree(pvBuf);
8676	pvBuf = NULL;
8677
8678	rc = VERR_NO_MEMORY;
8679	break;
8680	}
8681
8682	uint32_t uGTStart = VMDK_SECTOR2BYTE(pExtent->uSectorGD) + (cOldGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8683
8684	// points to last element in the grain table
8685	uint32_t uGTTail = uGTStart + (pExtent->cGDEntries * VMDK_GRAIN_TABLE_SIZE) - VMDK_GRAIN_TABLE_SIZE;
8686	uint32_t cbGTOff = RT_ALIGN_Z(VMDK_SECTOR2BYTE(cDirSectorDiff + cTableSectorDiff + cDirSectorDiff), 512);
8687
8688	for (int i = pExtent->cGDEntries - 1; i >= 0; i--)
8689	{
8690	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
8691	uGTTail, pvBuf,
8692	VMDK_GRAIN_TABLE_SIZE);
8693	if (RT_FAILURE(rc))
8694	break;
8695
8696	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8697	RT_ALIGN_Z(uGTTail + cbGTOff, 512), pvBuf,
8698	VMDK_GRAIN_TABLE_SIZE);
8699	if (RT_FAILURE(rc))
8700	break;
8701
8702	// This overshoots when i == 0, but we don't need it anymore.
8703	uGTTail -= VMDK_GRAIN_TABLE_SIZE;
8704	}
8705
8706
8707	/* Find the end of the grain directory and start bumping everything down. Update locations of GT entries. */
8708	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
8709	VMDK_SECTOR2BYTE(pExtent->uSectorGD), pvBuf,
8710	pExtent->cGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8711	if (RT_FAILURE(rc))
8712	break;
8713
8714	int * tmpBuf = (int *)pvBuf;
8715
8716	for (uint32_t i = 0; i < pExtent->cGDEntries; i++)
8717	{
8718	tmpBuf[i] = tmpBuf[i] + VMDK_BYTE2SECTOR(cbGTOff);
8719	pExtent->pGD[i] = pExtent->pGD[i] + VMDK_BYTE2SECTOR(cbGTOff);
8720	}
8721
8722	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8723	RT_ALIGN_Z(VMDK_SECTOR2BYTE(pExtent->uSectorGD + cTableSectorDiff + cDirSectorDiff), 512), pvBuf,
8724	pExtent->cGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8725	if (RT_FAILURE(rc))
8726	break;
8727
8728	pExtent->uSectorGD = pExtent->uSectorGD + cDirSectorDiff + cTableSectorDiff;
8729
8730	/* Repeat both steps with the redundant grain table/directory. */
8731
8732	uint32_t uRGTStart = VMDK_SECTOR2BYTE(pExtent->uSectorRGD) + (cOldGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8733
8734	// points to last element in the grain table
8735	uint32_t uRGTTail = uRGTStart + (pExtent->cGDEntries * VMDK_GRAIN_TABLE_SIZE) - VMDK_GRAIN_TABLE_SIZE;
8736	uint32_t cbRGTOff = RT_ALIGN_Z(VMDK_SECTOR2BYTE(cDirSectorDiff), 512);
8737
8738	for (int i = pExtent->cGDEntries - 1; i >= 0; i--)
8739	{
8740	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
8741	uRGTTail, pvBuf,
8742	VMDK_GRAIN_TABLE_SIZE);
8743	if (RT_FAILURE(rc))
8744	break;
8745
8746	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8747	RT_ALIGN_Z(uRGTTail + cbRGTOff, 512), pvBuf,
8748	VMDK_GRAIN_TABLE_SIZE);
8749	if (RT_FAILURE(rc))
8750	break;
8751
8752	// This overshoots when i == 0, but we don't need it anymore.
8753	uRGTTail -= VMDK_GRAIN_TABLE_SIZE;
8754	}
8755
8756	/* Update locations of GT entries. */
8757	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
8758	VMDK_SECTOR2BYTE(pExtent->uSectorRGD), pvBuf,
8759	pExtent->cGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8760	if (RT_FAILURE(rc))
8761	break;
8762
8763	tmpBuf = (int *)pvBuf;
8764
8765	for (uint32_t i = 0; i < pExtent->cGDEntries; i++)
8766	{
8767	tmpBuf[i] = tmpBuf[i] + cDirSectorDiff;
8768	pExtent->pRGD[i] = pExtent->pRGD[i] + cDirSectorDiff;
8769	}
8770
8771	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8772	VMDK_SECTOR2BYTE(pExtent->uSectorRGD), pvBuf,
8773	pExtent->cGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8774	if (RT_FAILURE(rc))
8775	break;
8776
8777	pExtent->uSectorRGD = pExtent->uSectorRGD;
8778	pExtent->cOverheadSectors += cOverheadSectorDiff;
8779
8780	} while (0);
8781
8782	if (pvBuf)
8783	{
8784	RTMemFree(pvBuf);
8785	pvBuf = NULL;
8786	}
8787
8788	if (pvZero)
8789	{
8790	RTMemFree(pvZero);
8791	pvZero = NULL;
8792	}
8793
8794	pExtent->cGDEntries = cNewDirEntries;
8795
8796	/* Allocate buffer for overwriting with zeroes. */
8797	pvZero = RTMemAllocZ(VMDK_GRAIN_TABLE_SIZE);
8798	if (!pvZero)
8799	return VERR_NO_MEMORY;
8800
8801	// Allocate additional grain dir
8802	pExtent->pGD = (uint32_t ) RTMemReallocZ(pExtent->pGD, pExtent->cGDEntries sizeof(uint32_t), cbNewGD);
8803	if (RT_LIKELY(pExtent->pGD))
8804	{
8805	if (pExtent->uSectorRGD)
8806	{
8807	pExtent->pRGD = (uint32_t )RTMemReallocZ(pExtent->pRGD, pExtent->cGDEntries sizeof(uint32_t), cbNewGD);
8808	if (RT_UNLIKELY(!pExtent->pRGD))
8809	rc = VERR_NO_MEMORY;
8810	}
8811	}
8812	else
8813	return VERR_NO_MEMORY;
8814
8815
8816	uint32_t uTmpDirVal = pExtent->pGD[cOldGDEntries - 1] + VMDK_GRAIN_DIR_ENTRY_SIZE;
8817	for (uint32_t i = cOldGDEntries; i < pExtent->cGDEntries; i++)
8818	{
8819	pExtent->pGD[i] = uTmpDirVal;
8820
8821	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8822	VMDK_SECTOR2BYTE(uTmpDirVal), pvZero,
8823	VMDK_GRAIN_TABLE_SIZE);
8824
8825	if (RT_FAILURE(rc))
8826	return rc;
8827
8828	uTmpDirVal += VMDK_GRAIN_DIR_ENTRY_SIZE;
8829	}
8830
8831	uint32_t uRTmpDirVal = pExtent->pRGD[cOldGDEntries - 1] + VMDK_GRAIN_DIR_ENTRY_SIZE;
8832	for (uint32_t i = cOldGDEntries; i < pExtent->cGDEntries; i++)
8833	{
8834	pExtent->pRGD[i] = uRTmpDirVal;
8835
8836	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8837	VMDK_SECTOR2BYTE(uRTmpDirVal), pvZero,
8838	VMDK_GRAIN_TABLE_SIZE);
8839
8840	if (RT_FAILURE(rc))
8841	return rc;
8842
8843	uRTmpDirVal += VMDK_GRAIN_DIR_ENTRY_SIZE;
8844	}
8845
8846	RTMemFree(pvZero);
8847	pvZero = NULL;
8848
8849	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8850	VMDK_SECTOR2BYTE(pExtent->uSectorGD), pExtent->pGD,
8851	pExtent->cGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8852	if (RT_FAILURE(rc))
8853	return rc;
8854
8855	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8856	VMDK_SECTOR2BYTE(pExtent->uSectorRGD), pExtent->pRGD,
8857	pExtent->cGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8858	if (RT_FAILURE(rc))
8859	return rc;
8860
8861	rc = vmdkReplaceExtentSize(pImage, pExtent, pImage->Descriptor.uFirstExtent + pExtent->uExtent,
8862	pExtent->cNominalSectors, cSectorsNew);
8863	if (RT_FAILURE(rc))
8864	return rc;
8865
8866	return rc;
8867	}
8868
8869	/** @copydoc VDIMAGEBACKEND::pfnResize */
8870	static DECLCALLBACK(int) vmdkResize(void *pBackendData, uint64_t cbSize,
8871	PCVDGEOMETRY pPCHSGeometry, PCVDGEOMETRY pLCHSGeometry,
8872	unsigned uPercentStart, unsigned uPercentSpan,
8873	PVDINTERFACE pVDIfsDisk, PVDINTERFACE pVDIfsImage,
8874	PVDINTERFACE pVDIfsOperation)
8875	{
8876	RT_NOREF5(uPercentStart, uPercentSpan, pVDIfsDisk, pVDIfsImage, pVDIfsOperation);
8877
8878	// Establish variables and objects needed
8879	int rc = VINF_SUCCESS;
8880	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8881	unsigned uImageFlags = pImage->uImageFlags;
8882	PVMDKEXTENT pExtent = &pImage->pExtents[0];
8883	pExtent->fMetaDirty = true;
8884
8885	uint64_t cSectorsNew = cbSize / VMDK_SECTOR_SIZE; /** < New number of sectors in the image after the resize */
8886	if (cbSize % VMDK_SECTOR_SIZE)
8887	cSectorsNew++;
8888
8889	uint64_t cSectorsOld = pImage->cbSize / VMDK_SECTOR_SIZE; /** < Number of sectors before the resize. Only for FLAT images. */
8890	if (pImage->cbSize % VMDK_SECTOR_SIZE)
8891	cSectorsOld++;
8892	unsigned cExtents = pImage->cExtents;
8893
8894	/* Check size is within min/max bounds. */
8895	if ( !(uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
8896	&& ( !cbSize
8897	\|\| (!(uImageFlags & VD_IMAGE_FLAGS_FIXED) && cbSize >= _1T * 256 - _64K)) )
8898	return VERR_VD_INVALID_SIZE;
8899
8900	/*
8901	* Making the image smaller is not supported at the moment.
8902	*/
8903	/** @todo implement making the image smaller, it is the responsibility of
8904	* the user to know what they're doing. */
8905	if (cbSize < pImage->cbSize)
8906	rc = VERR_VD_SHRINK_NOT_SUPPORTED;
8907	else if (cbSize > pImage->cbSize)
8908	{
8909	/**
8910	* monolithicFlat. FIXED flag and not split up into 2 GB parts.
8911	*/
8912	if ((uImageFlags & VD_IMAGE_FLAGS_FIXED) && !(uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G))
8913	{
8914	/** Required space in bytes for the extent after the resize. */
8915	uint64_t cbSectorSpaceNew = cSectorsNew * VMDK_SECTOR_SIZE;
8916	pExtent = &pImage->pExtents[0];
8917
8918	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage, cbSectorSpaceNew,
8919	0 /* fFlags */, NULL,
8920	uPercentStart, uPercentSpan);
8921	if (RT_FAILURE(rc))
8922	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
8923
8924	rc = vmdkReplaceExtentSize(pImage, pExtent, pImage->Descriptor.uFirstExtent, cSectorsOld, cSectorsNew);
8925	if (RT_FAILURE(rc))
8926	return rc;
8927	}
8928
8929	/**
8930	* twoGbMaxExtentFlat. FIXED flag and SPLIT into 2 GB parts.
8931	*/
8932	if ((uImageFlags & VD_IMAGE_FLAGS_FIXED) && (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G))
8933	{
8934	/* Check to see how much space remains in last extent */
8935	bool fSpaceAvailible = false;
8936	uint64_t cLastExtentRemSectors = cSectorsOld % VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
8937	if (cLastExtentRemSectors)
8938	fSpaceAvailible = true;
8939
8940	uint64_t cSectorsNeeded = cSectorsNew - cSectorsOld;
8941
8942	/** Space remaining in current last extent file that we don't need to create another one. */
8943	if (fSpaceAvailible && cSectorsNeeded + cLastExtentRemSectors <= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE))
8944	{
8945	pExtent = &pImage->pExtents[cExtents - 1];
8946	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage,
8947	VMDK_SECTOR2BYTE(cSectorsNeeded + cLastExtentRemSectors),
8948	0 /* fFlags */, NULL, uPercentStart, uPercentSpan);
8949	if (RT_FAILURE(rc))
8950	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
8951
8952	rc = vmdkReplaceExtentSize(pImage, pExtent, pImage->Descriptor.uFirstExtent + cExtents - 1,
8953	pExtent->cNominalSectors, cSectorsNeeded + cLastExtentRemSectors);
8954	if (RT_FAILURE(rc))
8955	return rc;
8956	}
8957	//** Need more extent files to handle all the requested space. */
8958	else
8959	{
8960	if (fSpaceAvailible)
8961	{
8962	pExtent = &pImage->pExtents[cExtents - 1];
8963	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage, VMDK_2G_SPLIT_SIZE,
8964	0 /* fFlags */, NULL,
8965	uPercentStart, uPercentSpan);
8966	if (RT_FAILURE(rc))
8967	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
8968
8969	cSectorsNeeded = cSectorsNeeded - VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE) + cLastExtentRemSectors;
8970
8971	rc = vmdkReplaceExtentSize(pImage, pExtent, pImage->Descriptor.uFirstExtent + cExtents - 1,
8972	pExtent->cNominalSectors, VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE));
8973	if (RT_FAILURE(rc))
8974	return rc;
8975	}
8976
8977	unsigned cNewExtents = VMDK_SECTOR2BYTE(cSectorsNeeded) / VMDK_2G_SPLIT_SIZE;
8978	if (cNewExtents % VMDK_2G_SPLIT_SIZE \|\| cNewExtents < VMDK_2G_SPLIT_SIZE)
8979	cNewExtents++;
8980
8981	for (unsigned i = cExtents;
8982	i < cExtents + cNewExtents && cSectorsNeeded >= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
8983	i++)
8984	{
8985	rc = vmdkAddFileBackedExtent(pImage, VMDK_2G_SPLIT_SIZE);
8986	if (RT_FAILURE(rc))
8987	return rc;
8988
8989	pExtent = &pImage->pExtents[i];
8990
8991	pExtent->cSectors = VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
8992	cSectorsNeeded -= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
8993	}
8994
8995	if (cSectorsNeeded)
8996	{
8997	rc = vmdkAddFileBackedExtent(pImage, VMDK_SECTOR2BYTE(cSectorsNeeded));
8998	if (RT_FAILURE(rc))
8999	return rc;
9000	}
9001	}
9002	}
9003
9004	/**
9005	* monolithicSparse.
9006	*/
9007	if (pExtent->enmType == VMDKETYPE_HOSTED_SPARSE && !(uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G))
9008	{
9009	// 1. Calculate sectors needed for new overhead.
9010
9011	uint64_t cbNewOverhead = vmdkGetNewOverhead(pExtent, cSectorsNew);
9012	uint64_t cNewOverheadSectors = VMDK_BYTE2SECTOR(cbNewOverhead);
9013	uint64_t cOverheadSectorDiff = cNewOverheadSectors - pExtent->cOverheadSectors;
9014
9015	// 2. Relocate sectors to make room for new GD/GT, update entries in GD/GT
9016	if (cOverheadSectorDiff > 0)
9017	{
9018	if (pExtent->cSectors > 0)
9019	{
9020	/* Do the relocation. */
9021	LogFlow(("Relocating VMDK sectors\n"));
9022	rc = vmdkRelocateSectorsForSparseResize(pImage, pExtent, cSectorsNew);
9023	if (RT_FAILURE(rc))
9024	return rc;
9025
9026	rc = vmdkFlushImage(pImage, NULL);
9027	if (RT_FAILURE(rc))
9028	return rc;
9029	}
9030
9031	rc = vmdkResizeSparseMeta(pImage, pExtent, cSectorsNew);
9032	if (RT_FAILURE(rc))
9033	return rc;
9034	}
9035	}
9036
9037	/**
9038	* twoGbSparseExtent
9039	*/
9040	if (pExtent->enmType == VMDKETYPE_HOSTED_SPARSE && (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G))
9041	{
9042	/* Check to see how much space remains in last extent */
9043	bool fSpaceAvailible = false;
9044	uint64_t cLastExtentRemSectors = cSectorsOld % VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
9045	if (cLastExtentRemSectors)
9046	fSpaceAvailible = true;
9047
9048	uint64_t cSectorsNeeded = cSectorsNew - cSectorsOld;
9049
9050	if (fSpaceAvailible && cSectorsNeeded + cLastExtentRemSectors <= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE))
9051	{
9052	pExtent = &pImage->pExtents[cExtents - 1];
9053	rc = vmdkRelocateSectorsForSparseResize(pImage, pExtent, cSectorsNeeded + cLastExtentRemSectors);
9054	if (RT_FAILURE(rc))
9055	return rc;
9056
9057	rc = vmdkFlushImage(pImage, NULL);
9058	if (RT_FAILURE(rc))
9059	return rc;
9060
9061	rc = vmdkResizeSparseMeta(pImage, pExtent, cSectorsNeeded + cLastExtentRemSectors);
9062	if (RT_FAILURE(rc))
9063	return rc;
9064	}
9065	else
9066	{
9067	if (fSpaceAvailible)
9068	{
9069	pExtent = &pImage->pExtents[cExtents - 1];
9070	rc = vmdkRelocateSectorsForSparseResize(pImage, pExtent, VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE));
9071	if (RT_FAILURE(rc))
9072	return rc;
9073
9074	rc = vmdkFlushImage(pImage, NULL);
9075	if (RT_FAILURE(rc))
9076	return rc;
9077
9078	rc = vmdkResizeSparseMeta(pImage, pExtent, VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE));
9079	if (RT_FAILURE(rc))
9080	return rc;
9081
9082	cSectorsNeeded = cSectorsNeeded - VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE) + cLastExtentRemSectors;
9083	}
9084
9085	unsigned cNewExtents = VMDK_SECTOR2BYTE(cSectorsNeeded) / VMDK_2G_SPLIT_SIZE;
9086	if (cNewExtents % VMDK_2G_SPLIT_SIZE \|\| cNewExtents < VMDK_2G_SPLIT_SIZE)
9087	cNewExtents++;
9088
9089	for (unsigned i = cExtents;
9090	i < cExtents + cNewExtents && cSectorsNeeded >= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
9091	i++)
9092	{
9093	rc = vmdkAddFileBackedExtent(pImage, VMDK_2G_SPLIT_SIZE);
9094	if (RT_FAILURE(rc))
9095	return rc;
9096
9097	pExtent = &pImage->pExtents[i];
9098
9099	rc = vmdkFlushImage(pImage, NULL);
9100	if (RT_FAILURE(rc))
9101	return rc;
9102
9103	pExtent->cSectors = VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
9104	cSectorsNeeded -= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
9105	}
9106
9107	if (cSectorsNeeded)
9108	{
9109	rc = vmdkAddFileBackedExtent(pImage, VMDK_SECTOR2BYTE(cSectorsNeeded));
9110	if (RT_FAILURE(rc))
9111	return rc;
9112
9113	pExtent = &pImage->pExtents[pImage->cExtents];
9114
9115	rc = vmdkFlushImage(pImage, NULL);
9116	if (RT_FAILURE(rc))
9117	return rc;
9118	}
9119	}
9120	}
9121
9122	/* Successful resize. Update metadata */
9123	if (RT_SUCCESS(rc))
9124	{
9125	/* Update size and new block count. */
9126	pImage->cbSize = cbSize;
9127	pExtent->cNominalSectors = cSectorsNew;
9128	pExtent->cSectors = cSectorsNew;
9129
9130	/* Update geometry. */
9131	pImage->PCHSGeometry = *pPCHSGeometry;
9132	pImage->LCHSGeometry = *pLCHSGeometry;
9133	}
9134
9135	/* Update header information in base image file. */
9136	pImage->Descriptor.fDirty = true;
9137	rc = vmdkWriteDescriptor(pImage, NULL);
9138
9139	if (RT_SUCCESS(rc))
9140	rc = vmdkFlushImage(pImage, NULL);
9141	}
9142	/* Same size doesn't change the image at all. */
9143
9144	LogFlowFunc(("returns %Rrc\n", rc));
9145	return rc;
9146	}
9147
9148	const VDIMAGEBACKEND g_VmdkBackend =
9149	{
9150	/* u32Version */
9151	VD_IMGBACKEND_VERSION,
9152	/* pszBackendName */
9153	"VMDK",
9154	/* uBackendCaps */
9155	VD_CAP_UUID \| VD_CAP_CREATE_FIXED \| VD_CAP_CREATE_DYNAMIC
9156	\| VD_CAP_CREATE_SPLIT_2G \| VD_CAP_DIFF \| VD_CAP_FILE \| VD_CAP_ASYNC
9157	\| VD_CAP_VFS \| VD_CAP_PREFERRED,
9158	/* paFileExtensions */
9159	s_aVmdkFileExtensions,
9160	/* paConfigInfo */
9161	s_aVmdkConfigInfo,
9162	/* pfnProbe */
9163	vmdkProbe,
9164	/* pfnOpen */
9165	vmdkOpen,
9166	/* pfnCreate */
9167	vmdkCreate,
9168	/* pfnRename */
9169	vmdkRename,
9170	/* pfnClose */
9171	vmdkClose,
9172	/* pfnRead */
9173	vmdkRead,
9174	/* pfnWrite */
9175	vmdkWrite,
9176	/* pfnFlush */
9177	vmdkFlush,
9178	/* pfnDiscard */
9179	NULL,
9180	/* pfnGetVersion */
9181	vmdkGetVersion,
9182	/* pfnGetFileSize */
9183	vmdkGetFileSize,
9184	/* pfnGetPCHSGeometry */
9185	vmdkGetPCHSGeometry,
9186	/* pfnSetPCHSGeometry */
9187	vmdkSetPCHSGeometry,
9188	/* pfnGetLCHSGeometry */
9189	vmdkGetLCHSGeometry,
9190	/* pfnSetLCHSGeometry */
9191	vmdkSetLCHSGeometry,
9192	/* pfnQueryRegions */
9193	vmdkQueryRegions,
9194	/* pfnRegionListRelease */
9195	vmdkRegionListRelease,
9196	/* pfnGetImageFlags */
9197	vmdkGetImageFlags,
9198	/* pfnGetOpenFlags */
9199	vmdkGetOpenFlags,
9200	/* pfnSetOpenFlags */
9201	vmdkSetOpenFlags,
9202	/* pfnGetComment */
9203	vmdkGetComment,
9204	/* pfnSetComment */
9205	vmdkSetComment,
9206	/* pfnGetUuid */
9207	vmdkGetUuid,
9208	/* pfnSetUuid */
9209	vmdkSetUuid,
9210	/* pfnGetModificationUuid */
9211	vmdkGetModificationUuid,
9212	/* pfnSetModificationUuid */
9213	vmdkSetModificationUuid,
9214	/* pfnGetParentUuid */
9215	vmdkGetParentUuid,
9216	/* pfnSetParentUuid */
9217	vmdkSetParentUuid,
9218	/* pfnGetParentModificationUuid */
9219	vmdkGetParentModificationUuid,
9220	/* pfnSetParentModificationUuid */
9221	vmdkSetParentModificationUuid,
9222	/* pfnDump */
9223	vmdkDump,
9224	/* pfnGetTimestamp */
9225	NULL,
9226	/* pfnGetParentTimestamp */
9227	NULL,
9228	/* pfnSetParentTimestamp */
9229	NULL,
9230	/* pfnGetParentFilename */
9231	NULL,
9232	/* pfnSetParentFilename */
9233	NULL,
9234	/* pfnComposeLocation */
9235	genericFileComposeLocation,
9236	/* pfnComposeName */
9237	genericFileComposeName,
9238	/* pfnCompact */
9239	NULL,
9240	/* pfnResize */
9241	vmdkResize,
9242	/* pfnRepair */
9243	NULL,
9244	/* pfnTraverseMetadata */
9245	NULL,
9246	/* u32VersionEnd */
9247	VD_IMGBACKEND_VERSION
9248	};

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

source: vbox/trunk/src/VBox/Storage/VMDK.cpp@ 101696

Download in other formats: