VMDK.cpp@ 99960

Last change on this file since 99960 was 99739, checked in by vboxsync, 19 months ago
*: doxygen corrections (mostly about removing @returns from functions returning void).
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 99739 2023-05-11 01:01:08Z 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/string.h>
47	#include <iprt/sort.h>
48	#include <iprt/zip.h>
49	#include <iprt/asm.h>
50	#ifdef RT_OS_WINDOWS
51	# include <iprt/utf16.h>
52	# include <iprt/uni.h>
53	# include <iprt/uni.h>
54	# include <iprt/nt/nt-and-windows.h>
55	# include <winioctl.h>
56	#endif
57	#ifdef RT_OS_LINUX
58	# include <errno.h>
59	# include <sys/stat.h>
60	# include <iprt/dir.h>
61	# include <iprt/symlink.h>
62	# include <iprt/linux/sysfs.h>
63	#endif
64	#ifdef RT_OS_FREEBSD
65	#include <libgeom.h>
66	#include <sys/stat.h>
67	#include <stdlib.h>
68	#endif
69	#ifdef RT_OS_SOLARIS
70	#include <sys/dkio.h>
71	#include <sys/vtoc.h>
72	#include <sys/efi_partition.h>
73	#include <unistd.h>
74	#include <errno.h>
75	#endif
76	#ifdef RT_OS_DARWIN
77	# include <sys/stat.h>
78	# include <sys/disk.h>
79	# include <errno.h>
80	/* The following structure and IOCTLs are defined in znu bsd/sys/disk.h but
81	inside KERNEL ifdefs and thus stripped from the SDK edition of the header.
82	While we could try include the header from the Kernel.framework, it's a lot
83	easier to just add the structure and 4 defines here. */
84	typedef struct
85	{
86	uint64_t offset;
87	uint64_t length;
88	uint8_t reserved0128[12];
89	dev_t dev;
90	} dk_physical_extent_t;
91	# define DKIOCGETBASE _IOR( 'd', 73, uint64_t)
92	# define DKIOCLOCKPHYSICALEXTENTS _IO( 'd', 81)
93	# define DKIOCGETPHYSICALEXTENT _IOWR('d', 82, dk_physical_extent_t)
94	# define DKIOCUNLOCKPHYSICALEXTENTS _IO( 'd', 83)
95	#endif /* RT_OS_DARWIN */
96
97	#include "VDBackends.h"
98
99
100	/*********************************************************************************************************************************
101	* Constants And Macros, Structures and Typedefs *
102	*********************************************************************************************************************************/
103
104	/** Maximum encoded string size (including NUL) we allow for VMDK images.
105	* Deliberately not set high to avoid running out of descriptor space. */
106	#define VMDK_ENCODED_COMMENT_MAX 1024
107
108	/** VMDK descriptor DDB entry for PCHS cylinders. */
109	#define VMDK_DDB_GEO_PCHS_CYLINDERS "ddb.geometry.cylinders"
110
111	/** VMDK descriptor DDB entry for PCHS heads. */
112	#define VMDK_DDB_GEO_PCHS_HEADS "ddb.geometry.heads"
113
114	/** VMDK descriptor DDB entry for PCHS sectors. */
115	#define VMDK_DDB_GEO_PCHS_SECTORS "ddb.geometry.sectors"
116
117	/** VMDK descriptor DDB entry for LCHS cylinders. */
118	#define VMDK_DDB_GEO_LCHS_CYLINDERS "ddb.geometry.biosCylinders"
119
120	/** VMDK descriptor DDB entry for LCHS heads. */
121	#define VMDK_DDB_GEO_LCHS_HEADS "ddb.geometry.biosHeads"
122
123	/** VMDK descriptor DDB entry for LCHS sectors. */
124	#define VMDK_DDB_GEO_LCHS_SECTORS "ddb.geometry.biosSectors"
125
126	/** VMDK descriptor DDB entry for image UUID. */
127	#define VMDK_DDB_IMAGE_UUID "ddb.uuid.image"
128
129	/** VMDK descriptor DDB entry for image modification UUID. */
130	#define VMDK_DDB_MODIFICATION_UUID "ddb.uuid.modification"
131
132	/** VMDK descriptor DDB entry for parent image UUID. */
133	#define VMDK_DDB_PARENT_UUID "ddb.uuid.parent"
134
135	/** VMDK descriptor DDB entry for parent image modification UUID. */
136	#define VMDK_DDB_PARENT_MODIFICATION_UUID "ddb.uuid.parentmodification"
137
138	/** No compression for streamOptimized files. */
139	#define VMDK_COMPRESSION_NONE 0
140
141	/** Deflate compression for streamOptimized files. */
142	#define VMDK_COMPRESSION_DEFLATE 1
143
144	/** Marker that the actual GD value is stored in the footer. */
145	#define VMDK_GD_AT_END 0xffffffffffffffffULL
146
147	/** Marker for end-of-stream in streamOptimized images. */
148	#define VMDK_MARKER_EOS 0
149
150	/** Marker for grain table block in streamOptimized images. */
151	#define VMDK_MARKER_GT 1
152
153	/** Marker for grain directory block in streamOptimized images. */
154	#define VMDK_MARKER_GD 2
155
156	/** Marker for footer in streamOptimized images. */
157	#define VMDK_MARKER_FOOTER 3
158
159	/** Marker for unknown purpose in streamOptimized images.
160	* Shows up in very recent images created by vSphere, but only sporadically.
161	* They "forgot" to document that one in the VMDK specification. */
162	#define VMDK_MARKER_UNSPECIFIED 4
163
164	/** Dummy marker for "don't check the marker value". */
165	#define VMDK_MARKER_IGNORE 0xffffffffU
166
167	/**
168	* Magic number for hosted images created by VMware Workstation 4, VMware
169	* Workstation 5, VMware Server or VMware Player. Not necessarily sparse.
170	*/
171	#define VMDK_SPARSE_MAGICNUMBER 0x564d444b /* 'V' 'M' 'D' 'K' */
172
173	/** VMDK sector size in bytes. */
174	#define VMDK_SECTOR_SIZE 512
175	/** Max string buffer size for uint64_t with null term */
176	#define UINT64_MAX_BUFF_SIZE 21
177	/** Grain directory entry size in bytes */
178	#define VMDK_GRAIN_DIR_ENTRY_SIZE 4
179	/** Grain table size in bytes */
180	#define VMDK_GRAIN_TABLE_SIZE 2048
181
182	/**
183	* VMDK hosted binary extent header. The "Sparse" is a total misnomer, as
184	* this header is also used for monolithic flat images.
185	*/
186	#pragma pack(1)
187	typedef struct SparseExtentHeader
188	{
189	uint32_t magicNumber;
190	uint32_t version;
191	uint32_t flags;
192	uint64_t capacity;
193	uint64_t grainSize;
194	uint64_t descriptorOffset;
195	uint64_t descriptorSize;
196	uint32_t numGTEsPerGT;
197	uint64_t rgdOffset;
198	uint64_t gdOffset;
199	uint64_t overHead;
200	bool uncleanShutdown;
201	char singleEndLineChar;
202	char nonEndLineChar;
203	char doubleEndLineChar1;
204	char doubleEndLineChar2;
205	uint16_t compressAlgorithm;
206	uint8_t pad[433];
207	} SparseExtentHeader;
208	#pragma pack()
209
210	/** The maximum allowed descriptor size in the extent header in sectors. */
211	#define VMDK_SPARSE_DESCRIPTOR_SIZE_MAX UINT64_C(20480) /* 10MB */
212
213	/** VMDK capacity for a single chunk when 2G splitting is turned on. Should be
214	* divisible by the default grain size (64K) */
215	#define VMDK_2G_SPLIT_SIZE (2047 * 1024 * 1024)
216
217	/** VMDK streamOptimized file format marker. The type field may or may not
218	* be actually valid, but there's always data to read there. */
219	#pragma pack(1)
220	typedef struct VMDKMARKER
221	{
222	uint64_t uSector;
223	uint32_t cbSize;
224	uint32_t uType;
225	} VMDKMARKER, *PVMDKMARKER;
226	#pragma pack()
227
228
229	/** Convert sector number/size to byte offset/size. */
230	#define VMDK_SECTOR2BYTE(u) ((uint64_t)(u) << 9)
231
232	/** Convert byte offset/size to sector number/size. */
233	#define VMDK_BYTE2SECTOR(u) ((u) >> 9)
234
235	/**
236	* VMDK extent type.
237	*/
238	typedef enum VMDKETYPE
239	{
240	/** Hosted sparse extent. */
241	VMDKETYPE_HOSTED_SPARSE = 1,
242	/** Flat extent. */
243	VMDKETYPE_FLAT,
244	/** Zero extent. */
245	VMDKETYPE_ZERO,
246	/** VMFS extent, used by ESX. */
247	VMDKETYPE_VMFS
248	} VMDKETYPE, *PVMDKETYPE;
249
250	/**
251	* VMDK access type for a extent.
252	*/
253	typedef enum VMDKACCESS
254	{
255	/** No access allowed. */
256	VMDKACCESS_NOACCESS = 0,
257	/** Read-only access. */
258	VMDKACCESS_READONLY,
259	/** Read-write access. */
260	VMDKACCESS_READWRITE
261	} VMDKACCESS, *PVMDKACCESS;
262
263	/** Forward declaration for PVMDKIMAGE. */
264	typedef struct VMDKIMAGE *PVMDKIMAGE;
265
266	/**
267	* Extents files entry. Used for opening a particular file only once.
268	*/
269	typedef struct VMDKFILE
270	{
271	/** Pointer to file path. Local copy. */
272	const char *pszFilename;
273	/** Pointer to base name. Local copy. */
274	const char *pszBasename;
275	/** File open flags for consistency checking. */
276	unsigned fOpen;
277	/** Handle for sync/async file abstraction.*/
278	PVDIOSTORAGE pStorage;
279	/** Reference counter. */
280	unsigned uReferences;
281	/** Flag whether the file should be deleted on last close. */
282	bool fDelete;
283	/** Pointer to the image we belong to (for debugging purposes). */
284	PVMDKIMAGE pImage;
285	/** Pointer to next file descriptor. */
286	struct VMDKFILE *pNext;
287	/** Pointer to the previous file descriptor. */
288	struct VMDKFILE *pPrev;
289	} VMDKFILE, *PVMDKFILE;
290
291	/**
292	* VMDK extent data structure.
293	*/
294	typedef struct VMDKEXTENT
295	{
296	/** File handle. */
297	PVMDKFILE pFile;
298	/** Base name of the image extent. */
299	const char *pszBasename;
300	/** Full name of the image extent. */
301	const char *pszFullname;
302	/** Number of sectors in this extent. */
303	uint64_t cSectors;
304	/** Number of sectors per block (grain in VMDK speak). */
305	uint64_t cSectorsPerGrain;
306	/** Starting sector number of descriptor. */
307	uint64_t uDescriptorSector;
308	/** Size of descriptor in sectors. */
309	uint64_t cDescriptorSectors;
310	/** Starting sector number of grain directory. */
311	uint64_t uSectorGD;
312	/** Starting sector number of redundant grain directory. */
313	uint64_t uSectorRGD;
314	/** Total number of metadata sectors. */
315	uint64_t cOverheadSectors;
316	/** Nominal size (i.e. as described by the descriptor) of this extent. */
317	uint64_t cNominalSectors;
318	/** Sector offset (i.e. as described by the descriptor) of this extent. */
319	uint64_t uSectorOffset;
320	/** Number of entries in a grain table. */
321	uint32_t cGTEntries;
322	/** Number of sectors reachable via a grain directory entry. */
323	uint32_t cSectorsPerGDE;
324	/** Number of entries in the grain directory. */
325	uint32_t cGDEntries;
326	/** Pointer to the next free sector. Legacy information. Do not use. */
327	uint32_t uFreeSector;
328	/** Number of this extent in the list of images. */
329	uint32_t uExtent;
330	/** Pointer to the descriptor (NULL if no descriptor in this extent). */
331	char *pDescData;
332	/** Pointer to the grain directory. */
333	uint32_t *pGD;
334	/** Pointer to the redundant grain directory. */
335	uint32_t *pRGD;
336	/** VMDK version of this extent. 1=1.0/1.1 */
337	uint32_t uVersion;
338	/** Type of this extent. */
339	VMDKETYPE enmType;
340	/** Access to this extent. */
341	VMDKACCESS enmAccess;
342	/** Flag whether this extent is marked as unclean. */
343	bool fUncleanShutdown;
344	/** Flag whether the metadata in the extent header needs to be updated. */
345	bool fMetaDirty;
346	/** Flag whether there is a footer in this extent. */
347	bool fFooter;
348	/** Compression type for this extent. */
349	uint16_t uCompression;
350	/** Append position for writing new grain. Only for sparse extents. */
351	uint64_t uAppendPosition;
352	/** Last grain which was accessed. Only for streamOptimized extents. */
353	uint32_t uLastGrainAccess;
354	/** Starting sector corresponding to the grain buffer. */
355	uint32_t uGrainSectorAbs;
356	/** Grain number corresponding to the grain buffer. */
357	uint32_t uGrain;
358	/** Actual size of the compressed data, only valid for reading. */
359	uint32_t cbGrainStreamRead;
360	/** Size of compressed grain buffer for streamOptimized extents. */
361	size_t cbCompGrain;
362	/** Compressed grain buffer for streamOptimized extents, with marker. */
363	void *pvCompGrain;
364	/** Decompressed grain buffer for streamOptimized extents. */
365	void *pvGrain;
366	/** Reference to the image in which this extent is used. Do not use this
367	* on a regular basis to avoid passing pImage references to functions
368	* explicitly. */
369	struct VMDKIMAGE *pImage;
370	} VMDKEXTENT, *PVMDKEXTENT;
371
372	/**
373	* Grain table cache size. Allocated per image.
374	*/
375	#define VMDK_GT_CACHE_SIZE 256
376
377	/**
378	* Grain table block size. Smaller than an actual grain table block to allow
379	* more grain table blocks to be cached without having to allocate excessive
380	* amounts of memory for the cache.
381	*/
382	#define VMDK_GT_CACHELINE_SIZE 128
383
384
385	/**
386	* Maximum number of lines in a descriptor file. Not worth the effort of
387	* making it variable. Descriptor files are generally very short (~20 lines),
388	* with the exception of sparse files split in 2G chunks, which need for the
389	* maximum size (almost 2T) exactly 1025 lines for the disk database.
390	*/
391	#define VMDK_DESCRIPTOR_LINES_MAX 1100U
392
393	/**
394	* Parsed descriptor information. Allows easy access and update of the
395	* descriptor (whether separate file or not). Free form text files suck.
396	*/
397	typedef struct VMDKDESCRIPTOR
398	{
399	/** Line number of first entry of the disk descriptor. */
400	unsigned uFirstDesc;
401	/** Line number of first entry in the extent description. */
402	unsigned uFirstExtent;
403	/** Line number of first disk database entry. */
404	unsigned uFirstDDB;
405	/** Total number of lines. */
406	unsigned cLines;
407	/** Total amount of memory available for the descriptor. */
408	size_t cbDescAlloc;
409	/** Set if descriptor has been changed and not yet written to disk. */
410	bool fDirty;
411	/** Array of pointers to the data in the descriptor. */
412	char *aLines[VMDK_DESCRIPTOR_LINES_MAX];
413	/** Array of line indices pointing to the next non-comment line. */
414	unsigned aNextLines[VMDK_DESCRIPTOR_LINES_MAX];
415	} VMDKDESCRIPTOR, *PVMDKDESCRIPTOR;
416
417
418	/**
419	* Cache entry for translating extent/sector to a sector number in that
420	* extent.
421	*/
422	typedef struct VMDKGTCACHEENTRY
423	{
424	/** Extent number for which this entry is valid. */
425	uint32_t uExtent;
426	/** GT data block number. */
427	uint64_t uGTBlock;
428	/** Data part of the cache entry. */
429	uint32_t aGTData[VMDK_GT_CACHELINE_SIZE];
430	} VMDKGTCACHEENTRY, *PVMDKGTCACHEENTRY;
431
432	/**
433	* Cache data structure for blocks of grain table entries. For now this is a
434	* fixed size direct mapping cache, but this should be adapted to the size of
435	* the sparse image and maybe converted to a set-associative cache. The
436	* implementation below implements a write-through cache with write allocate.
437	*/
438	typedef struct VMDKGTCACHE
439	{
440	/** Cache entries. */
441	VMDKGTCACHEENTRY aGTCache[VMDK_GT_CACHE_SIZE];
442	/** Number of cache entries (currently unused). */
443	unsigned cEntries;
444	} VMDKGTCACHE, *PVMDKGTCACHE;
445
446	/**
447	* Complete VMDK image data structure. Mainly a collection of extents and a few
448	* extra global data fields.
449	*/
450	typedef struct VMDKIMAGE
451	{
452	/** Image name. */
453	const char *pszFilename;
454	/** Descriptor file if applicable. */
455	PVMDKFILE pFile;
456
457	/** Pointer to the per-disk VD interface list. */
458	PVDINTERFACE pVDIfsDisk;
459	/** Pointer to the per-image VD interface list. */
460	PVDINTERFACE pVDIfsImage;
461
462	/** Error interface. */
463	PVDINTERFACEERROR pIfError;
464	/** I/O interface. */
465	PVDINTERFACEIOINT pIfIo;
466
467
468	/** Pointer to the image extents. */
469	PVMDKEXTENT pExtents;
470	/** Number of image extents. */
471	unsigned cExtents;
472	/** Pointer to the files list, for opening a file referenced multiple
473	* times only once (happens mainly with raw partition access). */
474	PVMDKFILE pFiles;
475
476	/**
477	* Pointer to an array of segment entries for async I/O.
478	* This is an optimization because the task number to submit is not known
479	* and allocating/freeing an array in the read/write functions every time
480	* is too expensive.
481	*/
482	PPDMDATASEG paSegments;
483	/** Entries available in the segments array. */
484	unsigned cSegments;
485
486	/** Open flags passed by VBoxHD layer. */
487	unsigned uOpenFlags;
488	/** Image flags defined during creation or determined during open. */
489	unsigned uImageFlags;
490	/** Total size of the image. */
491	uint64_t cbSize;
492	/** Physical geometry of this image. */
493	VDGEOMETRY PCHSGeometry;
494	/** Logical geometry of this image. */
495	VDGEOMETRY LCHSGeometry;
496	/** Image UUID. */
497	RTUUID ImageUuid;
498	/** Image modification UUID. */
499	RTUUID ModificationUuid;
500	/** Parent image UUID. */
501	RTUUID ParentUuid;
502	/** Parent image modification UUID. */
503	RTUUID ParentModificationUuid;
504
505	/** Pointer to grain table cache, if this image contains sparse extents. */
506	PVMDKGTCACHE pGTCache;
507	/** Pointer to the descriptor (NULL if no separate descriptor file). */
508	char *pDescData;
509	/** Allocation size of the descriptor file. */
510	size_t cbDescAlloc;
511	/** Parsed descriptor file content. */
512	VMDKDESCRIPTOR Descriptor;
513	/** The static region list. */
514	VDREGIONLIST RegionList;
515	} VMDKIMAGE;
516
517
518	/** State for the input/output callout of the inflate reader/deflate writer. */
519	typedef struct VMDKCOMPRESSIO
520	{
521	/* Image this operation relates to. */
522	PVMDKIMAGE pImage;
523	/* Current read position. */
524	ssize_t iOffset;
525	/* Size of the compressed grain buffer (available data). */
526	size_t cbCompGrain;
527	/* Pointer to the compressed grain buffer. */
528	void *pvCompGrain;
529	} VMDKCOMPRESSIO;
530
531
532	/** Tracks async grain allocation. */
533	typedef struct VMDKGRAINALLOCASYNC
534	{
535	/** Flag whether the allocation failed. */
536	bool fIoErr;
537	/** Current number of transfers pending.
538	* If reached 0 and there is an error the old state is restored. */
539	unsigned cIoXfersPending;
540	/** Sector number */
541	uint64_t uSector;
542	/** Flag whether the grain table needs to be updated. */
543	bool fGTUpdateNeeded;
544	/** Extent the allocation happens. */
545	PVMDKEXTENT pExtent;
546	/** Position of the new grain, required for the grain table update. */
547	uint64_t uGrainOffset;
548	/** Grain table sector. */
549	uint64_t uGTSector;
550	/** Backup grain table sector. */
551	uint64_t uRGTSector;
552	} VMDKGRAINALLOCASYNC, *PVMDKGRAINALLOCASYNC;
553
554	/**
555	* State information for vmdkRename() and helpers.
556	*/
557	typedef struct VMDKRENAMESTATE
558	{
559	/** Array of old filenames. */
560	char **apszOldName;
561	/** Array of new filenames. */
562	char **apszNewName;
563	/** Array of new lines in the extent descriptor. */
564	char **apszNewLines;
565	/** Name of the old descriptor file if not a sparse image. */
566	char *pszOldDescName;
567	/** Flag whether we called vmdkFreeImage(). */
568	bool fImageFreed;
569	/** Flag whther the descriptor is embedded in the image (sparse) or
570	* in a separate file. */
571	bool fEmbeddedDesc;
572	/** Number of extents in the image. */
573	unsigned cExtents;
574	/** New base filename. */
575	char *pszNewBaseName;
576	/** The old base filename. */
577	char *pszOldBaseName;
578	/** New full filename. */
579	char *pszNewFullName;
580	/** Old full filename. */
581	char *pszOldFullName;
582	/** The old image name. */
583	const char *pszOldImageName;
584	/** Copy of the original VMDK descriptor. */
585	VMDKDESCRIPTOR DescriptorCopy;
586	/** Copy of the extent state for sparse images. */
587	VMDKEXTENT ExtentCopy;
588	} VMDKRENAMESTATE;
589	/** Pointer to a VMDK rename state. */
590	typedef VMDKRENAMESTATE *PVMDKRENAMESTATE;
591
592
593	/*********************************************************************************************************************************
594	* Static Variables *
595	*********************************************************************************************************************************/
596
597	/** NULL-terminated array of supported file extensions. */
598	static const VDFILEEXTENSION s_aVmdkFileExtensions[] =
599	{
600	{"vmdk", VDTYPE_HDD},
601	{NULL, VDTYPE_INVALID}
602	};
603
604	/** NULL-terminated array of configuration option. */
605	static const VDCONFIGINFO s_aVmdkConfigInfo[] =
606	{
607	/* Options for VMDK raw disks */
608	{ "RawDrive", NULL, VDCFGVALUETYPE_STRING, 0 },
609	{ "Partitions", NULL, VDCFGVALUETYPE_STRING, 0 },
610	{ "BootSector", NULL, VDCFGVALUETYPE_BYTES, 0 },
611	{ "Relative", NULL, VDCFGVALUETYPE_INTEGER, 0 },
612
613	/* End of options list */
614	{ NULL, NULL, VDCFGVALUETYPE_INTEGER, 0 }
615	};
616
617
618	/*********************************************************************************************************************************
619	* Internal Functions *
620	*********************************************************************************************************************************/
621
622	static void vmdkFreeStreamBuffers(PVMDKEXTENT pExtent);
623	static int vmdkFreeExtentData(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
624	bool fDelete);
625
626	static int vmdkCreateExtents(PVMDKIMAGE pImage, unsigned cExtents);
627	static int vmdkFlushImage(PVMDKIMAGE pImage, PVDIOCTX pIoCtx);
628	static int vmdkSetImageComment(PVMDKIMAGE pImage, const char *pszComment);
629	static int vmdkFreeImage(PVMDKIMAGE pImage, bool fDelete, bool fFlush);
630
631	static DECLCALLBACK(int) vmdkAllocGrainComplete(void *pBackendData, PVDIOCTX pIoCtx,
632	void *pvUser, int rcReq);
633
634	/**
635	* Internal: open a file (using a file descriptor cache to ensure each file
636	* is only opened once - anything else can cause locking problems).
637	*/
638	static int vmdkFileOpen(PVMDKIMAGE pImage, PVMDKFILE *ppVmdkFile,
639	const char pszBasename, const char pszFilename, uint32_t fOpen)
640	{
641	int rc = VINF_SUCCESS;
642	PVMDKFILE pVmdkFile;
643
644	for (pVmdkFile = pImage->pFiles;
645	pVmdkFile != NULL;
646	pVmdkFile = pVmdkFile->pNext)
647	{
648	if (!strcmp(pszFilename, pVmdkFile->pszFilename))
649	{
650	Assert(fOpen == pVmdkFile->fOpen);
651	pVmdkFile->uReferences++;
652
653	*ppVmdkFile = pVmdkFile;
654
655	return rc;
656	}
657	}
658
659	/* If we get here, there's no matching entry in the cache. */
660	pVmdkFile = (PVMDKFILE)RTMemAllocZ(sizeof(VMDKFILE));
661	if (!pVmdkFile)
662	{
663	*ppVmdkFile = NULL;
664	return VERR_NO_MEMORY;
665	}
666
667	pVmdkFile->pszFilename = RTStrDup(pszFilename);
668	if (!pVmdkFile->pszFilename)
669	{
670	RTMemFree(pVmdkFile);
671	*ppVmdkFile = NULL;
672	return VERR_NO_MEMORY;
673	}
674
675	if (pszBasename)
676	{
677	pVmdkFile->pszBasename = RTStrDup(pszBasename);
678	if (!pVmdkFile->pszBasename)
679	{
680	RTStrFree((char )(void )pVmdkFile->pszFilename);
681	RTMemFree(pVmdkFile);
682	*ppVmdkFile = NULL;
683	return VERR_NO_MEMORY;
684	}
685	}
686
687	pVmdkFile->fOpen = fOpen;
688
689	rc = vdIfIoIntFileOpen(pImage->pIfIo, pszFilename, fOpen,
690	&pVmdkFile->pStorage);
691	if (RT_SUCCESS(rc))
692	{
693	pVmdkFile->uReferences = 1;
694	pVmdkFile->pImage = pImage;
695	pVmdkFile->pNext = pImage->pFiles;
696	if (pImage->pFiles)
697	pImage->pFiles->pPrev = pVmdkFile;
698	pImage->pFiles = pVmdkFile;
699	*ppVmdkFile = pVmdkFile;
700	}
701	else
702	{
703	RTStrFree((char )(void )pVmdkFile->pszFilename);
704	RTMemFree(pVmdkFile);
705	*ppVmdkFile = NULL;
706	}
707
708	return rc;
709	}
710
711	/**
712	* Internal: close a file, updating the file descriptor cache.
713	*/
714	static int vmdkFileClose(PVMDKIMAGE pImage, PVMDKFILE *ppVmdkFile, bool fDelete)
715	{
716	int rc = VINF_SUCCESS;
717	PVMDKFILE pVmdkFile = *ppVmdkFile;
718
719	AssertPtr(pVmdkFile);
720
721	pVmdkFile->fDelete \|= fDelete;
722	Assert(pVmdkFile->uReferences);
723	pVmdkFile->uReferences--;
724	if (pVmdkFile->uReferences == 0)
725	{
726	PVMDKFILE pPrev;
727	PVMDKFILE pNext;
728
729	/* Unchain the element from the list. */
730	pPrev = pVmdkFile->pPrev;
731	pNext = pVmdkFile->pNext;
732
733	if (pNext)
734	pNext->pPrev = pPrev;
735	if (pPrev)
736	pPrev->pNext = pNext;
737	else
738	pImage->pFiles = pNext;
739
740	rc = vdIfIoIntFileClose(pImage->pIfIo, pVmdkFile->pStorage);
741
742	bool fFileDel = pVmdkFile->fDelete;
743	if ( pVmdkFile->pszBasename
744	&& fFileDel)
745	{
746	const char *pszSuffix = RTPathSuffix(pVmdkFile->pszBasename);
747	if ( RTPathHasPath(pVmdkFile->pszBasename)
748	\|\| !pszSuffix
749	\|\| ( strcmp(pszSuffix, ".vmdk")
750	&& strcmp(pszSuffix, ".bin")
751	&& strcmp(pszSuffix, ".img")))
752	fFileDel = false;
753	}
754
755	if (fFileDel)
756	{
757	int rc2 = vdIfIoIntFileDelete(pImage->pIfIo, pVmdkFile->pszFilename);
758	if (RT_SUCCESS(rc))
759	rc = rc2;
760	}
761	else if (pVmdkFile->fDelete)
762	LogRel(("VMDK: Denying deletion of %s\n", pVmdkFile->pszBasename));
763	RTStrFree((char )(void )pVmdkFile->pszFilename);
764	if (pVmdkFile->pszBasename)
765	RTStrFree((char )(void )pVmdkFile->pszBasename);
766	RTMemFree(pVmdkFile);
767	}
768
769	*ppVmdkFile = NULL;
770	return rc;
771	}
772
773	/#define VMDK_USE_BLOCK_DECOMP_API - test and enable /
774	#ifndef VMDK_USE_BLOCK_DECOMP_API
775	static DECLCALLBACK(int) vmdkFileInflateHelper(void pvUser, void pvBuf, size_t cbBuf, size_t *pcbBuf)
776	{
777	VMDKCOMPRESSIO pInflateState = (VMDKCOMPRESSIO )pvUser;
778	size_t cbInjected = 0;
779
780	Assert(cbBuf);
781	if (pInflateState->iOffset < 0)
782	{
783	(uint8_t )pvBuf = RTZIPTYPE_ZLIB;
784	pvBuf = (uint8_t *)pvBuf + 1;
785	cbBuf--;
786	cbInjected = 1;
787	pInflateState->iOffset = RT_UOFFSETOF(VMDKMARKER, uType);
788	}
789	if (!cbBuf)
790	{
791	if (pcbBuf)
792	*pcbBuf = cbInjected;
793	return VINF_SUCCESS;
794	}
795	cbBuf = RT_MIN(cbBuf, pInflateState->cbCompGrain - pInflateState->iOffset);
796	memcpy(pvBuf,
797	(uint8_t *)pInflateState->pvCompGrain + pInflateState->iOffset,
798	cbBuf);
799	pInflateState->iOffset += cbBuf;
800	Assert(pcbBuf);
801	*pcbBuf = cbBuf + cbInjected;
802	return VINF_SUCCESS;
803	}
804	#endif
805
806	/**
807	* Internal: read from a file and inflate the compressed data,
808	* distinguishing between async and normal operation
809	*/
810	DECLINLINE(int) vmdkFileInflateSync(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
811	uint64_t uOffset, void *pvBuf,
812	size_t cbToRead, const void *pcvMarker,
813	uint64_t puLBA, uint32_t pcbMarkerData)
814	{
815	int rc;
816	#ifndef VMDK_USE_BLOCK_DECOMP_API
817	PRTZIPDECOMP pZip = NULL;
818	#endif
819	VMDKMARKER pMarker = (VMDKMARKER )pExtent->pvCompGrain;
820	size_t cbCompSize, cbActuallyRead;
821
822	if (!pcvMarker)
823	{
824	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
825	uOffset, pMarker, RT_UOFFSETOF(VMDKMARKER, uType));
826	if (RT_FAILURE(rc))
827	return rc;
828	}
829	else
830	{
831	memcpy(pMarker, pcvMarker, RT_UOFFSETOF(VMDKMARKER, uType));
832	/* pcvMarker endianness has already been partially transformed, fix it */
833	pMarker->uSector = RT_H2LE_U64(pMarker->uSector);
834	pMarker->cbSize = RT_H2LE_U32(pMarker->cbSize);
835	}
836
837	cbCompSize = RT_LE2H_U32(pMarker->cbSize);
838	if (cbCompSize == 0)
839	{
840	AssertMsgFailed(("VMDK: corrupted marker\n"));
841	return VERR_VD_VMDK_INVALID_FORMAT;
842	}
843
844	/* Sanity check - the expansion ratio should be much less than 2. */
845	Assert(cbCompSize < 2 * cbToRead);
846	if (cbCompSize >= 2 * cbToRead)
847	return VERR_VD_VMDK_INVALID_FORMAT;
848
849	/* Compressed grain marker. Data follows immediately. */
850	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
851	uOffset + RT_UOFFSETOF(VMDKMARKER, uType),
852	(uint8_t *)pExtent->pvCompGrain
853	+ RT_UOFFSETOF(VMDKMARKER, uType),
854	RT_ALIGN_Z( cbCompSize
855	+ RT_UOFFSETOF(VMDKMARKER, uType),
856	512)
857	- RT_UOFFSETOF(VMDKMARKER, uType));
858
859	if (puLBA)
860	*puLBA = RT_LE2H_U64(pMarker->uSector);
861	if (pcbMarkerData)
862	*pcbMarkerData = RT_ALIGN( cbCompSize
863	+ RT_UOFFSETOF(VMDKMARKER, uType),
864	512);
865
866	#ifdef VMDK_USE_BLOCK_DECOMP_API
867	rc = RTZipBlockDecompress(RTZIPTYPE_ZLIB, 0 /fFlags/,
868	pExtent->pvCompGrain, cbCompSize + RT_UOFFSETOF(VMDKMARKER, uType), NULL,
869	pvBuf, cbToRead, &cbActuallyRead);
870	#else
871	VMDKCOMPRESSIO InflateState;
872	InflateState.pImage = pImage;
873	InflateState.iOffset = -1;
874	InflateState.cbCompGrain = cbCompSize + RT_UOFFSETOF(VMDKMARKER, uType);
875	InflateState.pvCompGrain = pExtent->pvCompGrain;
876
877	rc = RTZipDecompCreate(&pZip, &InflateState, vmdkFileInflateHelper);
878	if (RT_FAILURE(rc))
879	return rc;
880	rc = RTZipDecompress(pZip, pvBuf, cbToRead, &cbActuallyRead);
881	RTZipDecompDestroy(pZip);
882	#endif /* !VMDK_USE_BLOCK_DECOMP_API */
883	if (RT_FAILURE(rc))
884	{
885	if (rc == VERR_ZIP_CORRUPTED)
886	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: Compressed image is corrupted '%s'"), pExtent->pszFullname);
887	return rc;
888	}
889	if (cbActuallyRead != cbToRead)
890	rc = VERR_VD_VMDK_INVALID_FORMAT;
891	return rc;
892	}
893
894	static DECLCALLBACK(int) vmdkFileDeflateHelper(void pvUser, const void pvBuf, size_t cbBuf)
895	{
896	VMDKCOMPRESSIO pDeflateState = (VMDKCOMPRESSIO )pvUser;
897
898	Assert(cbBuf);
899	if (pDeflateState->iOffset < 0)
900	{
901	pvBuf = (const uint8_t *)pvBuf + 1;
902	cbBuf--;
903	pDeflateState->iOffset = RT_UOFFSETOF(VMDKMARKER, uType);
904	}
905	if (!cbBuf)
906	return VINF_SUCCESS;
907	if (pDeflateState->iOffset + cbBuf > pDeflateState->cbCompGrain)
908	return VERR_BUFFER_OVERFLOW;
909	memcpy((uint8_t *)pDeflateState->pvCompGrain + pDeflateState->iOffset,
910	pvBuf, cbBuf);
911	pDeflateState->iOffset += cbBuf;
912	return VINF_SUCCESS;
913	}
914
915	/**
916	* Internal: deflate the uncompressed data and write to a file,
917	* distinguishing between async and normal operation
918	*/
919	DECLINLINE(int) vmdkFileDeflateSync(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
920	uint64_t uOffset, const void *pvBuf,
921	size_t cbToWrite, uint64_t uLBA,
922	uint32_t *pcbMarkerData)
923	{
924	int rc;
925	PRTZIPCOMP pZip = NULL;
926	VMDKCOMPRESSIO DeflateState;
927
928	DeflateState.pImage = pImage;
929	DeflateState.iOffset = -1;
930	DeflateState.cbCompGrain = pExtent->cbCompGrain;
931	DeflateState.pvCompGrain = pExtent->pvCompGrain;
932
933	rc = RTZipCompCreate(&pZip, &DeflateState, vmdkFileDeflateHelper,
934	RTZIPTYPE_ZLIB, RTZIPLEVEL_DEFAULT);
935	if (RT_FAILURE(rc))
936	return rc;
937	rc = RTZipCompress(pZip, pvBuf, cbToWrite);
938	if (RT_SUCCESS(rc))
939	rc = RTZipCompFinish(pZip);
940	RTZipCompDestroy(pZip);
941	if (RT_SUCCESS(rc))
942	{
943	Assert( DeflateState.iOffset > 0
944	&& (size_t)DeflateState.iOffset <= DeflateState.cbCompGrain);
945
946	/* pad with zeroes to get to a full sector size */
947	uint32_t uSize = DeflateState.iOffset;
948	if (uSize % 512)
949	{
950	uint32_t uSizeAlign = RT_ALIGN(uSize, 512);
951	memset((uint8_t *)pExtent->pvCompGrain + uSize, '\0',
952	uSizeAlign - uSize);
953	uSize = uSizeAlign;
954	}
955
956	if (pcbMarkerData)
957	*pcbMarkerData = uSize;
958
959	/* Compressed grain marker. Data follows immediately. */
960	VMDKMARKER pMarker = (VMDKMARKER )pExtent->pvCompGrain;
961	pMarker->uSector = RT_H2LE_U64(uLBA);
962	pMarker->cbSize = RT_H2LE_U32( DeflateState.iOffset
963	- RT_UOFFSETOF(VMDKMARKER, uType));
964	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
965	uOffset, pMarker, uSize);
966	if (RT_FAILURE(rc))
967	return rc;
968	}
969	return rc;
970	}
971
972
973	/**
974	* Internal: check if all files are closed, prevent leaking resources.
975	*/
976	static int vmdkFileCheckAllClose(PVMDKIMAGE pImage)
977	{
978	int rc = VINF_SUCCESS, rc2;
979	PVMDKFILE pVmdkFile;
980
981	Assert(pImage->pFiles == NULL);
982	for (pVmdkFile = pImage->pFiles;
983	pVmdkFile != NULL;
984	pVmdkFile = pVmdkFile->pNext)
985	{
986	LogRel(("VMDK: leaking reference to file \"%s\"\n",
987	pVmdkFile->pszFilename));
988	pImage->pFiles = pVmdkFile->pNext;
989
990	rc2 = vmdkFileClose(pImage, &pVmdkFile, pVmdkFile->fDelete);
991
992	if (RT_SUCCESS(rc))
993	rc = rc2;
994	}
995	return rc;
996	}
997
998	/**
999	* Internal: truncate a string (at a UTF8 code point boundary) and encode the
1000	* critical non-ASCII characters.
1001	*/
1002	static char vmdkEncodeString(const char psz)
1003	{
1004	char szEnc[VMDK_ENCODED_COMMENT_MAX + 3];
1005	char *pszDst = szEnc;
1006
1007	AssertPtr(psz);
1008
1009	for (; *psz; psz = RTStrNextCp(psz))
1010	{
1011	char *pszDstPrev = pszDst;
1012	RTUNICP Cp = RTStrGetCp(psz);
1013	if (Cp == '\\')
1014	{
1015	pszDst = RTStrPutCp(pszDst, Cp);
1016	pszDst = RTStrPutCp(pszDst, Cp);
1017	}
1018	else if (Cp == '\n')
1019	{
1020	pszDst = RTStrPutCp(pszDst, '\\');
1021	pszDst = RTStrPutCp(pszDst, 'n');
1022	}
1023	else if (Cp == '\r')
1024	{
1025	pszDst = RTStrPutCp(pszDst, '\\');
1026	pszDst = RTStrPutCp(pszDst, 'r');
1027	}
1028	else
1029	pszDst = RTStrPutCp(pszDst, Cp);
1030	if (pszDst - szEnc >= VMDK_ENCODED_COMMENT_MAX - 1)
1031	{
1032	pszDst = pszDstPrev;
1033	break;
1034	}
1035	}
1036	*pszDst = '\0';
1037	return RTStrDup(szEnc);
1038	}
1039
1040	/**
1041	* Internal: decode a string and store it into the specified string.
1042	*/
1043	static int vmdkDecodeString(const char pszEncoded, char psz, size_t cb)
1044	{
1045	int rc = VINF_SUCCESS;
1046	char szBuf[4];
1047
1048	if (!cb)
1049	return VERR_BUFFER_OVERFLOW;
1050
1051	AssertPtr(psz);
1052
1053	for (; *pszEncoded; pszEncoded = RTStrNextCp(pszEncoded))
1054	{
1055	char *pszDst = szBuf;
1056	RTUNICP Cp = RTStrGetCp(pszEncoded);
1057	if (Cp == '\\')
1058	{
1059	pszEncoded = RTStrNextCp(pszEncoded);
1060	RTUNICP CpQ = RTStrGetCp(pszEncoded);
1061	if (CpQ == 'n')
1062	RTStrPutCp(pszDst, '\n');
1063	else if (CpQ == 'r')
1064	RTStrPutCp(pszDst, '\r');
1065	else if (CpQ == '\0')
1066	{
1067	rc = VERR_VD_VMDK_INVALID_HEADER;
1068	break;
1069	}
1070	else
1071	RTStrPutCp(pszDst, CpQ);
1072	}
1073	else
1074	pszDst = RTStrPutCp(pszDst, Cp);
1075
1076	/* Need to leave space for terminating NUL. */
1077	if ((size_t)(pszDst - szBuf) + 1 >= cb)
1078	{
1079	rc = VERR_BUFFER_OVERFLOW;
1080	break;
1081	}
1082	memcpy(psz, szBuf, pszDst - szBuf);
1083	psz += pszDst - szBuf;
1084	}
1085	*psz = '\0';
1086	return rc;
1087	}
1088
1089	/**
1090	* Internal: free all buffers associated with grain directories.
1091	*/
1092	static void vmdkFreeGrainDirectory(PVMDKEXTENT pExtent)
1093	{
1094	if (pExtent->pGD)
1095	{
1096	RTMemFree(pExtent->pGD);
1097	pExtent->pGD = NULL;
1098	}
1099	if (pExtent->pRGD)
1100	{
1101	RTMemFree(pExtent->pRGD);
1102	pExtent->pRGD = NULL;
1103	}
1104	}
1105
1106	/**
1107	* Internal: allocate the compressed/uncompressed buffers for streamOptimized
1108	* images.
1109	*/
1110	static int vmdkAllocStreamBuffers(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
1111	{
1112	int rc = VINF_SUCCESS;
1113
1114	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
1115	{
1116	/* streamOptimized extents need a compressed grain buffer, which must
1117	* be big enough to hold uncompressible data (which needs ~8 bytes
1118	* more than the uncompressed data), the marker and padding. */
1119	pExtent->cbCompGrain = RT_ALIGN_Z( VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain)
1120	+ 8 + sizeof(VMDKMARKER), 512);
1121	pExtent->pvCompGrain = RTMemAlloc(pExtent->cbCompGrain);
1122	if (RT_LIKELY(pExtent->pvCompGrain))
1123	{
1124	/* streamOptimized extents need a decompressed grain buffer. */
1125	pExtent->pvGrain = RTMemAlloc(VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
1126	if (!pExtent->pvGrain)
1127	rc = VERR_NO_MEMORY;
1128	}
1129	else
1130	rc = VERR_NO_MEMORY;
1131	}
1132
1133	if (RT_FAILURE(rc))
1134	vmdkFreeStreamBuffers(pExtent);
1135	return rc;
1136	}
1137
1138	/**
1139	* Internal: allocate all buffers associated with grain directories.
1140	*/
1141	static int vmdkAllocGrainDirectory(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
1142	{
1143	RT_NOREF1(pImage);
1144	int rc = VINF_SUCCESS;
1145	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
1146
1147	pExtent->pGD = (uint32_t *)RTMemAllocZ(cbGD);
1148	if (RT_LIKELY(pExtent->pGD))
1149	{
1150	if (pExtent->uSectorRGD)
1151	{
1152	pExtent->pRGD = (uint32_t *)RTMemAllocZ(cbGD);
1153	if (RT_UNLIKELY(!pExtent->pRGD))
1154	rc = VERR_NO_MEMORY;
1155	}
1156	}
1157	else
1158	rc = VERR_NO_MEMORY;
1159
1160	if (RT_FAILURE(rc))
1161	vmdkFreeGrainDirectory(pExtent);
1162	return rc;
1163	}
1164
1165	/**
1166	* Converts the grain directory from little to host endianess.
1167	*
1168	* @param pGD The grain directory.
1169	* @param cGDEntries Number of entries in the grain directory to convert.
1170	*/
1171	DECLINLINE(void) vmdkGrainDirectoryConvToHost(uint32_t *pGD, uint32_t cGDEntries)
1172	{
1173	uint32_t *pGDTmp = pGD;
1174
1175	for (uint32_t i = 0; i < cGDEntries; i++, pGDTmp++)
1176	pGDTmp = RT_LE2H_U32(pGDTmp);
1177	}
1178
1179	/**
1180	* Read the grain directory and allocated grain tables verifying them against
1181	* their back up copies if available.
1182	*
1183	* @returns VBox status code.
1184	* @param pImage Image instance data.
1185	* @param pExtent The VMDK extent.
1186	*/
1187	static int vmdkReadGrainDirectory(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
1188	{
1189	int rc = VINF_SUCCESS;
1190	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
1191
1192	AssertReturn(( pExtent->enmType == VMDKETYPE_HOSTED_SPARSE
1193	&& pExtent->uSectorGD != VMDK_GD_AT_END
1194	&& pExtent->uSectorRGD != VMDK_GD_AT_END), VERR_INTERNAL_ERROR);
1195
1196	rc = vmdkAllocGrainDirectory(pImage, pExtent);
1197	if (RT_SUCCESS(rc))
1198	{
1199	/* The VMDK 1.1 spec seems to talk about compressed grain directories,
1200	* but in reality they are not compressed. */
1201	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1202	VMDK_SECTOR2BYTE(pExtent->uSectorGD),
1203	pExtent->pGD, cbGD);
1204	if (RT_SUCCESS(rc))
1205	{
1206	vmdkGrainDirectoryConvToHost(pExtent->pGD, pExtent->cGDEntries);
1207
1208	if ( pExtent->uSectorRGD
1209	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_SKIP_CONSISTENCY_CHECKS))
1210	{
1211	/* The VMDK 1.1 spec seems to talk about compressed grain directories,
1212	* but in reality they are not compressed. */
1213	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1214	VMDK_SECTOR2BYTE(pExtent->uSectorRGD),
1215	pExtent->pRGD, cbGD);
1216	if (RT_SUCCESS(rc))
1217	{
1218	vmdkGrainDirectoryConvToHost(pExtent->pRGD, pExtent->cGDEntries);
1219
1220	/* Check grain table and redundant grain table for consistency. */
1221	size_t cbGT = pExtent->cGTEntries * sizeof(uint32_t);
1222	size_t cbGTBuffers = cbGT; /* Start with space for one GT. */
1223	size_t cbGTBuffersMax = _1M;
1224
1225	uint32_t pTmpGT1 = (uint32_t )RTMemAlloc(cbGTBuffers);
1226	uint32_t pTmpGT2 = (uint32_t )RTMemAlloc(cbGTBuffers);
1227
1228	if ( !pTmpGT1
1229	\|\| !pTmpGT2)
1230	rc = VERR_NO_MEMORY;
1231
1232	size_t i = 0;
1233	uint32_t *pGDTmp = pExtent->pGD;
1234	uint32_t *pRGDTmp = pExtent->pRGD;
1235
1236	/* Loop through all entries. */
1237	while (i < pExtent->cGDEntries)
1238	{
1239	uint32_t uGTStart = *pGDTmp;
1240	uint32_t uRGTStart = *pRGDTmp;
1241	size_t cbGTRead = cbGT;
1242
1243	/* If no grain table is allocated skip the entry. */
1244	if (pGDTmp == 0 && pRGDTmp == 0)
1245	{
1246	i++;
1247	continue;
1248	}
1249
1250	if (pGDTmp == 0 \|\| pRGDTmp == 0 \|\| pGDTmp == pRGDTmp)
1251	{
1252	/* Just one grain directory entry refers to a not yet allocated
1253	* grain table or both grain directory copies refer to the same
1254	* grain table. Not allowed. */
1255	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1256	N_("VMDK: inconsistent references to grain directory in '%s'"), pExtent->pszFullname);
1257	break;
1258	}
1259
1260	i++;
1261	pGDTmp++;
1262	pRGDTmp++;
1263
1264	/*
1265	* Read a few tables at once if adjacent to decrease the number
1266	* of I/O requests. Read at maximum 1MB at once.
1267	*/
1268	while ( i < pExtent->cGDEntries
1269	&& cbGTRead < cbGTBuffersMax)
1270	{
1271	/* If no grain table is allocated skip the entry. */
1272	if (pGDTmp == 0 && pRGDTmp == 0)
1273	{
1274	i++;
1275	continue;
1276	}
1277
1278	if (pGDTmp == 0 \|\| pRGDTmp == 0 \|\| pGDTmp == pRGDTmp)
1279	{
1280	/* Just one grain directory entry refers to a not yet allocated
1281	* grain table or both grain directory copies refer to the same
1282	* grain table. Not allowed. */
1283	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1284	N_("VMDK: inconsistent references to grain directory in '%s'"), pExtent->pszFullname);
1285	break;
1286	}
1287
1288	/* Check that the start offsets are adjacent.*/
1289	if ( VMDK_SECTOR2BYTE(uGTStart) + cbGTRead != VMDK_SECTOR2BYTE(*pGDTmp)
1290	\|\| VMDK_SECTOR2BYTE(uRGTStart) + cbGTRead != VMDK_SECTOR2BYTE(*pRGDTmp))
1291	break;
1292
1293	i++;
1294	pGDTmp++;
1295	pRGDTmp++;
1296	cbGTRead += cbGT;
1297	}
1298
1299	/* Increase buffers if required. */
1300	if ( RT_SUCCESS(rc)
1301	&& cbGTBuffers < cbGTRead)
1302	{
1303	uint32_t *pTmp;
1304	pTmp = (uint32_t *)RTMemRealloc(pTmpGT1, cbGTRead);
1305	if (pTmp)
1306	{
1307	pTmpGT1 = pTmp;
1308	pTmp = (uint32_t *)RTMemRealloc(pTmpGT2, cbGTRead);
1309	if (pTmp)
1310	pTmpGT2 = pTmp;
1311	else
1312	rc = VERR_NO_MEMORY;
1313	}
1314	else
1315	rc = VERR_NO_MEMORY;
1316
1317	if (rc == VERR_NO_MEMORY)
1318	{
1319	/* Reset to the old values. */
1320	rc = VINF_SUCCESS;
1321	i -= cbGTRead / cbGT;
1322	cbGTRead = cbGT;
1323
1324	/* Don't try to increase the buffer again in the next run. */
1325	cbGTBuffersMax = cbGTBuffers;
1326	}
1327	}
1328
1329	if (RT_SUCCESS(rc))
1330	{
1331	/* The VMDK 1.1 spec seems to talk about compressed grain tables,
1332	* but in reality they are not compressed. */
1333	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1334	VMDK_SECTOR2BYTE(uGTStart),
1335	pTmpGT1, cbGTRead);
1336	if (RT_FAILURE(rc))
1337	{
1338	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1339	N_("VMDK: error reading grain table in '%s'"), pExtent->pszFullname);
1340	break;
1341	}
1342	/* The VMDK 1.1 spec seems to talk about compressed grain tables,
1343	* but in reality they are not compressed. */
1344	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1345	VMDK_SECTOR2BYTE(uRGTStart),
1346	pTmpGT2, cbGTRead);
1347	if (RT_FAILURE(rc))
1348	{
1349	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1350	N_("VMDK: error reading backup grain table in '%s'"), pExtent->pszFullname);
1351	break;
1352	}
1353	if (memcmp(pTmpGT1, pTmpGT2, cbGTRead))
1354	{
1355	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1356	N_("VMDK: inconsistency between grain table and backup grain table in '%s'"), pExtent->pszFullname);
1357	break;
1358	}
1359	}
1360	} /* while (i < pExtent->cGDEntries) */
1361
1362	/** @todo figure out what to do for unclean VMDKs. */
1363	if (pTmpGT1)
1364	RTMemFree(pTmpGT1);
1365	if (pTmpGT2)
1366	RTMemFree(pTmpGT2);
1367	}
1368	else
1369	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1370	N_("VMDK: could not read redundant grain directory in '%s'"), pExtent->pszFullname);
1371	}
1372	}
1373	else
1374	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1375	N_("VMDK: could not read grain directory in '%s': %Rrc"), pExtent->pszFullname, rc);
1376	}
1377
1378	if (RT_FAILURE(rc))
1379	vmdkFreeGrainDirectory(pExtent);
1380	return rc;
1381	}
1382
1383	/**
1384	* Creates a new grain directory for the given extent at the given start sector.
1385	*
1386	* @returns VBox status code.
1387	* @param pImage Image instance data.
1388	* @param pExtent The VMDK extent.
1389	* @param uStartSector Where the grain directory should be stored in the image.
1390	* @param fPreAlloc Flag whether to pre allocate the grain tables at this point.
1391	*/
1392	static int vmdkCreateGrainDirectory(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
1393	uint64_t uStartSector, bool fPreAlloc)
1394	{
1395	int rc = VINF_SUCCESS;
1396	unsigned i;
1397	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
1398	size_t cbGDRounded = RT_ALIGN_64(cbGD, 512);
1399	size_t cbGTRounded;
1400	uint64_t cbOverhead;
1401
1402	if (fPreAlloc)
1403	{
1404	cbGTRounded = RT_ALIGN_64(pExtent->cGDEntries * pExtent->cGTEntries * sizeof(uint32_t), 512);
1405	cbOverhead = VMDK_SECTOR2BYTE(uStartSector) + cbGDRounded + cbGTRounded;
1406	}
1407	else
1408	{
1409	/* Use a dummy start sector for layout computation. */
1410	if (uStartSector == VMDK_GD_AT_END)
1411	uStartSector = 1;
1412	cbGTRounded = 0;
1413	cbOverhead = VMDK_SECTOR2BYTE(uStartSector) + cbGDRounded;
1414	}
1415
1416	/* For streamOptimized extents there is only one grain directory,
1417	* and for all others take redundant grain directory into account. */
1418	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
1419	{
1420	cbOverhead = RT_ALIGN_64(cbOverhead,
1421	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
1422	}
1423	else
1424	{
1425	cbOverhead += cbGDRounded + cbGTRounded;
1426	cbOverhead = RT_ALIGN_64(cbOverhead,
1427	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
1428	rc = vdIfIoIntFileSetSize(pImage->pIfIo, pExtent->pFile->pStorage, cbOverhead);
1429	}
1430
1431	if (RT_SUCCESS(rc))
1432	{
1433	pExtent->uAppendPosition = cbOverhead;
1434	pExtent->cOverheadSectors = VMDK_BYTE2SECTOR(cbOverhead);
1435
1436	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
1437	{
1438	pExtent->uSectorRGD = 0;
1439	pExtent->uSectorGD = uStartSector;
1440	}
1441	else
1442	{
1443	pExtent->uSectorRGD = uStartSector;
1444	pExtent->uSectorGD = uStartSector + VMDK_BYTE2SECTOR(cbGDRounded + cbGTRounded);
1445	}
1446
1447	rc = vmdkAllocStreamBuffers(pImage, pExtent);
1448	if (RT_SUCCESS(rc))
1449	{
1450	rc = vmdkAllocGrainDirectory(pImage, pExtent);
1451	if ( RT_SUCCESS(rc)
1452	&& fPreAlloc)
1453	{
1454	uint32_t uGTSectorLE;
1455	uint64_t uOffsetSectors;
1456
1457	if (pExtent->pRGD)
1458	{
1459	uOffsetSectors = pExtent->uSectorRGD + VMDK_BYTE2SECTOR(cbGDRounded);
1460	for (i = 0; i < pExtent->cGDEntries; i++)
1461	{
1462	pExtent->pRGD[i] = uOffsetSectors;
1463	uGTSectorLE = RT_H2LE_U64(uOffsetSectors);
1464	/* Write the redundant grain directory entry to disk. */
1465	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
1466	VMDK_SECTOR2BYTE(pExtent->uSectorRGD) + i * sizeof(uGTSectorLE),
1467	&uGTSectorLE, sizeof(uGTSectorLE));
1468	if (RT_FAILURE(rc))
1469	{
1470	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write new redundant grain directory entry in '%s'"), pExtent->pszFullname);
1471	break;
1472	}
1473	uOffsetSectors += VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
1474	}
1475	}
1476
1477	if (RT_SUCCESS(rc))
1478	{
1479	uOffsetSectors = pExtent->uSectorGD + VMDK_BYTE2SECTOR(cbGDRounded);
1480	for (i = 0; i < pExtent->cGDEntries; i++)
1481	{
1482	pExtent->pGD[i] = uOffsetSectors;
1483	uGTSectorLE = RT_H2LE_U64(uOffsetSectors);
1484	/* Write the grain directory entry to disk. */
1485	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
1486	VMDK_SECTOR2BYTE(pExtent->uSectorGD) + i * sizeof(uGTSectorLE),
1487	&uGTSectorLE, sizeof(uGTSectorLE));
1488	if (RT_FAILURE(rc))
1489	{
1490	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write new grain directory entry in '%s'"), pExtent->pszFullname);
1491	break;
1492	}
1493	uOffsetSectors += VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
1494	}
1495	}
1496	}
1497	}
1498	}
1499
1500	if (RT_FAILURE(rc))
1501	vmdkFreeGrainDirectory(pExtent);
1502	return rc;
1503	}
1504
1505	/**
1506	* Unquotes the given string returning the result in a separate buffer.
1507	*
1508	* @returns VBox status code.
1509	* @param pImage The VMDK image state.
1510	* @param pszStr The string to unquote.
1511	* @param ppszUnquoted Where to store the return value, use RTMemTmpFree to
1512	* free.
1513	* @param ppszNext Where to store the pointer to any character following
1514	* the quoted value, optional.
1515	*/
1516	static int vmdkStringUnquote(PVMDKIMAGE pImage, const char *pszStr,
1517	char ppszUnquoted, char ppszNext)
1518	{
1519	const char *pszStart = pszStr;
1520	char *pszQ;
1521	char *pszUnquoted;
1522
1523	/* Skip over whitespace. */
1524	while (pszStr == ' ' \|\| pszStr == '\t')
1525	pszStr++;
1526
1527	if (*pszStr != '"')
1528	{
1529	pszQ = (char *)pszStr;
1530	while (pszQ && pszQ != ' ' && *pszQ != '\t')
1531	pszQ++;
1532	}
1533	else
1534	{
1535	pszStr++;
1536	pszQ = (char *)strchr(pszStr, '"');
1537	if (pszQ == NULL)
1538	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrectly quoted value in descriptor in '%s' (raw value %s)"),
1539	pImage->pszFilename, pszStart);
1540	}
1541
1542	pszUnquoted = (char *)RTMemTmpAlloc(pszQ - pszStr + 1);
1543	if (!pszUnquoted)
1544	return VERR_NO_MEMORY;
1545	memcpy(pszUnquoted, pszStr, pszQ - pszStr);
1546	pszUnquoted[pszQ - pszStr] = '\0';
1547	*ppszUnquoted = pszUnquoted;
1548	if (ppszNext)
1549	*ppszNext = pszQ + 1;
1550	return VINF_SUCCESS;
1551	}
1552
1553	static int vmdkDescInitStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1554	const char *pszLine)
1555	{
1556	char *pEnd = pDescriptor->aLines[pDescriptor->cLines];
1557	ssize_t cbDiff = strlen(pszLine) + 1;
1558
1559	if ( pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1
1560	&& pEnd - pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff)
1561	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1562
1563	memcpy(pEnd, pszLine, cbDiff);
1564	pDescriptor->cLines++;
1565	pDescriptor->aLines[pDescriptor->cLines] = pEnd + cbDiff;
1566	pDescriptor->fDirty = true;
1567
1568	return VINF_SUCCESS;
1569	}
1570
1571	static bool vmdkDescGetStr(PVMDKDESCRIPTOR pDescriptor, unsigned uStart,
1572	const char pszKey, const char *ppszValue)
1573	{
1574	size_t cbKey = strlen(pszKey);
1575	const char *pszValue;
1576
1577	while (uStart != 0)
1578	{
1579	if (!strncmp(pDescriptor->aLines[uStart], pszKey, cbKey))
1580	{
1581	/* Key matches, check for a '=' (preceded by whitespace). */
1582	pszValue = pDescriptor->aLines[uStart] + cbKey;
1583	while (pszValue == ' ' \|\| pszValue == '\t')
1584	pszValue++;
1585	if (*pszValue == '=')
1586	{
1587	*ppszValue = pszValue + 1;
1588	break;
1589	}
1590	}
1591	uStart = pDescriptor->aNextLines[uStart];
1592	}
1593	return !!uStart;
1594	}
1595
1596	static int vmdkDescSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1597	unsigned uStart,
1598	const char pszKey, const char pszValue)
1599	{
1600	char pszTmp = NULL; / (MSC naturally cannot figure this isn't used uninitialized) */
1601	size_t cbKey = strlen(pszKey);
1602	unsigned uLast = 0;
1603
1604	while (uStart != 0)
1605	{
1606	if (!strncmp(pDescriptor->aLines[uStart], pszKey, cbKey))
1607	{
1608	/* Key matches, check for a '=' (preceded by whitespace). */
1609	pszTmp = pDescriptor->aLines[uStart] + cbKey;
1610	while (pszTmp == ' ' \|\| pszTmp == '\t')
1611	pszTmp++;
1612	if (*pszTmp == '=')
1613	{
1614	pszTmp++;
1615	/** @todo r=bird: Doesn't skipping trailing blanks here just cause unecessary
1616	* bloat and potentially out of space error? */
1617	while (pszTmp == ' ' \|\| pszTmp == '\t')
1618	pszTmp++;
1619	break;
1620	}
1621	}
1622	if (!pDescriptor->aNextLines[uStart])
1623	uLast = uStart;
1624	uStart = pDescriptor->aNextLines[uStart];
1625	}
1626	if (uStart)
1627	{
1628	if (pszValue)
1629	{
1630	/* Key already exists, replace existing value. */
1631	size_t cbOldVal = strlen(pszTmp);
1632	size_t cbNewVal = strlen(pszValue);
1633	ssize_t cbDiff = cbNewVal - cbOldVal;
1634	/* Check for buffer overflow. */
1635	if ( pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[0]
1636	> (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff)
1637	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1638
1639	memmove(pszTmp + cbNewVal, pszTmp + cbOldVal,
1640	pDescriptor->aLines[pDescriptor->cLines] - pszTmp - cbOldVal);
1641	memcpy(pszTmp, pszValue, cbNewVal + 1);
1642	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1643	pDescriptor->aLines[i] += cbDiff;
1644	}
1645	else
1646	{
1647	memmove(pDescriptor->aLines[uStart], pDescriptor->aLines[uStart+1],
1648	pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[uStart+1] + 1);
1649	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1650	{
1651	pDescriptor->aLines[i-1] = pDescriptor->aLines[i];
1652	if (pDescriptor->aNextLines[i])
1653	pDescriptor->aNextLines[i-1] = pDescriptor->aNextLines[i] - 1;
1654	else
1655	pDescriptor->aNextLines[i-1] = 0;
1656	}
1657	pDescriptor->cLines--;
1658	/* Adjust starting line numbers of following descriptor sections. */
1659	if (uStart < pDescriptor->uFirstExtent)
1660	pDescriptor->uFirstExtent--;
1661	if (uStart < pDescriptor->uFirstDDB)
1662	pDescriptor->uFirstDDB--;
1663	}
1664	}
1665	else
1666	{
1667	/* Key doesn't exist, append after the last entry in this category. */
1668	if (!pszValue)
1669	{
1670	/* Key doesn't exist, and it should be removed. Simply a no-op. */
1671	return VINF_SUCCESS;
1672	}
1673	cbKey = strlen(pszKey);
1674	size_t cbValue = strlen(pszValue);
1675	ssize_t cbDiff = cbKey + 1 + cbValue + 1;
1676	/* Check for buffer overflow. */
1677	if ( (pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1)
1678	\|\| ( pDescriptor->aLines[pDescriptor->cLines]
1679	- pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff))
1680	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1681	for (unsigned i = pDescriptor->cLines + 1; i > uLast + 1; i--)
1682	{
1683	pDescriptor->aLines[i] = pDescriptor->aLines[i - 1];
1684	if (pDescriptor->aNextLines[i - 1])
1685	pDescriptor->aNextLines[i] = pDescriptor->aNextLines[i - 1] + 1;
1686	else
1687	pDescriptor->aNextLines[i] = 0;
1688	}
1689	uStart = uLast + 1;
1690	pDescriptor->aNextLines[uLast] = uStart;
1691	pDescriptor->aNextLines[uStart] = 0;
1692	pDescriptor->cLines++;
1693	pszTmp = pDescriptor->aLines[uStart];
1694	memmove(pszTmp + cbDiff, pszTmp,
1695	pDescriptor->aLines[pDescriptor->cLines] - pszTmp);
1696	memcpy(pDescriptor->aLines[uStart], pszKey, cbKey);
1697	pDescriptor->aLines[uStart][cbKey] = '=';
1698	memcpy(pDescriptor->aLines[uStart] + cbKey + 1, pszValue, cbValue + 1);
1699	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1700	pDescriptor->aLines[i] += cbDiff;
1701
1702	/* Adjust starting line numbers of following descriptor sections. */
1703	if (uStart <= pDescriptor->uFirstExtent)
1704	pDescriptor->uFirstExtent++;
1705	if (uStart <= pDescriptor->uFirstDDB)
1706	pDescriptor->uFirstDDB++;
1707	}
1708	pDescriptor->fDirty = true;
1709	return VINF_SUCCESS;
1710	}
1711
1712	static int vmdkDescBaseGetU32(PVMDKDESCRIPTOR pDescriptor, const char *pszKey,
1713	uint32_t *puValue)
1714	{
1715	const char *pszValue;
1716
1717	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDesc, pszKey,
1718	&pszValue))
1719	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1720	return RTStrToUInt32Ex(pszValue, NULL, 10, puValue);
1721	}
1722
1723	/**
1724	* Returns the value of the given key as a string allocating the necessary memory.
1725	*
1726	* @returns VBox status code.
1727	* @retval VERR_VD_VMDK_VALUE_NOT_FOUND if the value could not be found.
1728	* @param pImage The VMDK image state.
1729	* @param pDescriptor The descriptor to fetch the value from.
1730	* @param pszKey The key to get the value from.
1731	* @param ppszValue Where to store the return value, use RTMemTmpFree to
1732	* free.
1733	*/
1734	static int vmdkDescBaseGetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1735	const char pszKey, char *ppszValue)
1736	{
1737	const char *pszValue;
1738	char *pszValueUnquoted;
1739
1740	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDesc, pszKey,
1741	&pszValue))
1742	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1743	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1744	if (RT_FAILURE(rc))
1745	return rc;
1746	*ppszValue = pszValueUnquoted;
1747	return rc;
1748	}
1749
1750	static int vmdkDescBaseSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1751	const char pszKey, const char pszValue)
1752	{
1753	char *pszValueQuoted;
1754
1755	RTStrAPrintf(&pszValueQuoted, "\"%s\"", pszValue);
1756	if (!pszValueQuoted)
1757	return VERR_NO_STR_MEMORY;
1758	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc, pszKey,
1759	pszValueQuoted);
1760	RTStrFree(pszValueQuoted);
1761	return rc;
1762	}
1763
1764	static void vmdkDescExtRemoveDummy(PVMDKIMAGE pImage,
1765	PVMDKDESCRIPTOR pDescriptor)
1766	{
1767	RT_NOREF1(pImage);
1768	unsigned uEntry = pDescriptor->uFirstExtent;
1769	ssize_t cbDiff;
1770
1771	if (!uEntry)
1772	return;
1773
1774	cbDiff = strlen(pDescriptor->aLines[uEntry]) + 1;
1775	/* Move everything including \0 in the entry marking the end of buffer. */
1776	memmove(pDescriptor->aLines[uEntry], pDescriptor->aLines[uEntry + 1],
1777	pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[uEntry + 1] + 1);
1778	for (unsigned i = uEntry + 1; i <= pDescriptor->cLines; i++)
1779	{
1780	pDescriptor->aLines[i - 1] = pDescriptor->aLines[i] - cbDiff;
1781	if (pDescriptor->aNextLines[i])
1782	pDescriptor->aNextLines[i - 1] = pDescriptor->aNextLines[i] - 1;
1783	else
1784	pDescriptor->aNextLines[i - 1] = 0;
1785	}
1786	pDescriptor->cLines--;
1787	if (pDescriptor->uFirstDDB)
1788	pDescriptor->uFirstDDB--;
1789
1790	return;
1791	}
1792
1793	static void vmdkDescExtRemoveByLine(PVMDKIMAGE pImage,
1794	PVMDKDESCRIPTOR pDescriptor, unsigned uLine)
1795	{
1796	RT_NOREF1(pImage);
1797	unsigned uEntry = uLine;
1798	ssize_t cbDiff;
1799	if (!uEntry)
1800	return;
1801	cbDiff = strlen(pDescriptor->aLines[uEntry]) + 1;
1802	/* Move everything including \0 in the entry marking the end of buffer. */
1803	memmove(pDescriptor->aLines[uEntry], pDescriptor->aLines[uEntry + 1],
1804	pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[uEntry + 1] + 1);
1805	for (unsigned i = uEntry; i <= pDescriptor->cLines; i++)
1806	{
1807	if (i != uEntry)
1808	pDescriptor->aLines[i - 1] = pDescriptor->aLines[i] - cbDiff;
1809	if (pDescriptor->aNextLines[i])
1810	pDescriptor->aNextLines[i - 1] = pDescriptor->aNextLines[i] - 1;
1811	else
1812	pDescriptor->aNextLines[i - 1] = 0;
1813	}
1814	pDescriptor->cLines--;
1815	if (pDescriptor->uFirstDDB)
1816	pDescriptor->uFirstDDB--;
1817	return;
1818	}
1819
1820	static int vmdkDescExtInsert(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1821	VMDKACCESS enmAccess, uint64_t cNominalSectors,
1822	VMDKETYPE enmType, const char *pszBasename,
1823	uint64_t uSectorOffset)
1824	{
1825	static const char *apszAccess[] = { "NOACCESS", "RDONLY", "RW" };
1826	static const char *apszType[] = { "", "SPARSE", "FLAT", "ZERO", "VMFS" };
1827	char *pszTmp;
1828	unsigned uStart = pDescriptor->uFirstExtent, uLast = 0;
1829	char szExt[1024];
1830	ssize_t cbDiff;
1831
1832	Assert((unsigned)enmAccess < RT_ELEMENTS(apszAccess));
1833	Assert((unsigned)enmType < RT_ELEMENTS(apszType));
1834
1835	/* Find last entry in extent description. */
1836	while (uStart)
1837	{
1838	if (!pDescriptor->aNextLines[uStart])
1839	uLast = uStart;
1840	uStart = pDescriptor->aNextLines[uStart];
1841	}
1842
1843	if (enmType == VMDKETYPE_ZERO)
1844	{
1845	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s ", apszAccess[enmAccess],
1846	cNominalSectors, apszType[enmType]);
1847	}
1848	else if (enmType == VMDKETYPE_FLAT)
1849	{
1850	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s \"%s\" %llu",
1851	apszAccess[enmAccess], cNominalSectors,
1852	apszType[enmType], pszBasename, uSectorOffset);
1853	}
1854	else
1855	{
1856	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s \"%s\"",
1857	apszAccess[enmAccess], cNominalSectors,
1858	apszType[enmType], pszBasename);
1859	}
1860	cbDiff = strlen(szExt) + 1;
1861
1862	/* Check for buffer overflow. */
1863	if ( (pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1)
1864	\|\| ( pDescriptor->aLines[pDescriptor->cLines]
1865	- pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff))
1866	{
1867	if ((pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
1868	&& !(pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1))
1869	{
1870	pImage->cbDescAlloc *= 2;
1871	pDescriptor->cbDescAlloc *= 2;
1872	}
1873	else
1874	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1875	}
1876
1877	for (unsigned i = pDescriptor->cLines + 1; i > uLast + 1; i--)
1878	{
1879	pDescriptor->aLines[i] = pDescriptor->aLines[i - 1];
1880	if (pDescriptor->aNextLines[i - 1])
1881	pDescriptor->aNextLines[i] = pDescriptor->aNextLines[i - 1] + 1;
1882	else
1883	pDescriptor->aNextLines[i] = 0;
1884	}
1885	uStart = uLast + 1;
1886	pDescriptor->aNextLines[uLast] = uStart;
1887	pDescriptor->aNextLines[uStart] = 0;
1888	pDescriptor->cLines++;
1889	pszTmp = pDescriptor->aLines[uStart];
1890	memmove(pszTmp + cbDiff, pszTmp,
1891	pDescriptor->aLines[pDescriptor->cLines] - pszTmp);
1892	memcpy(pDescriptor->aLines[uStart], szExt, cbDiff);
1893	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1894	pDescriptor->aLines[i] += cbDiff;
1895
1896	/* Adjust starting line numbers of following descriptor sections. */
1897	if (uStart <= pDescriptor->uFirstDDB)
1898	pDescriptor->uFirstDDB++;
1899
1900	pDescriptor->fDirty = true;
1901	return VINF_SUCCESS;
1902	}
1903
1904	/**
1905	* Returns the value of the given key from the DDB as a string allocating
1906	* the necessary memory.
1907	*
1908	* @returns VBox status code.
1909	* @retval VERR_VD_VMDK_VALUE_NOT_FOUND if the value could not be found.
1910	* @param pImage The VMDK image state.
1911	* @param pDescriptor The descriptor to fetch the value from.
1912	* @param pszKey The key to get the value from.
1913	* @param ppszValue Where to store the return value, use RTMemTmpFree to
1914	* free.
1915	*/
1916	static int vmdkDescDDBGetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1917	const char pszKey, char *ppszValue)
1918	{
1919	const char *pszValue;
1920	char *pszValueUnquoted;
1921
1922	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1923	&pszValue))
1924	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1925	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1926	if (RT_FAILURE(rc))
1927	return rc;
1928	*ppszValue = pszValueUnquoted;
1929	return rc;
1930	}
1931
1932	static int vmdkDescDDBGetU32(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1933	const char pszKey, uint32_t puValue)
1934	{
1935	const char *pszValue;
1936	char *pszValueUnquoted;
1937
1938	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1939	&pszValue))
1940	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1941	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1942	if (RT_FAILURE(rc))
1943	return rc;
1944	rc = RTStrToUInt32Ex(pszValueUnquoted, NULL, 10, puValue);
1945	RTMemTmpFree(pszValueUnquoted);
1946	return rc;
1947	}
1948
1949	static int vmdkDescDDBGetUuid(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1950	const char *pszKey, PRTUUID pUuid)
1951	{
1952	const char *pszValue;
1953	char *pszValueUnquoted;
1954
1955	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1956	&pszValue))
1957	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1958	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1959	if (RT_FAILURE(rc))
1960	return rc;
1961	rc = RTUuidFromStr(pUuid, pszValueUnquoted);
1962	RTMemTmpFree(pszValueUnquoted);
1963	return rc;
1964	}
1965
1966	static int vmdkDescDDBSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1967	const char pszKey, const char pszVal)
1968	{
1969	int rc;
1970	char *pszValQuoted;
1971
1972	if (pszVal)
1973	{
1974	RTStrAPrintf(&pszValQuoted, "\"%s\"", pszVal);
1975	if (!pszValQuoted)
1976	return VERR_NO_STR_MEMORY;
1977	}
1978	else
1979	pszValQuoted = NULL;
1980	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1981	pszValQuoted);
1982	if (pszValQuoted)
1983	RTStrFree(pszValQuoted);
1984	return rc;
1985	}
1986
1987	static int vmdkDescDDBSetUuid(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1988	const char *pszKey, PCRTUUID pUuid)
1989	{
1990	char *pszUuid;
1991
1992	RTStrAPrintf(&pszUuid, "\"%RTuuid\"", pUuid);
1993	if (!pszUuid)
1994	return VERR_NO_STR_MEMORY;
1995	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1996	pszUuid);
1997	RTStrFree(pszUuid);
1998	return rc;
1999	}
2000
2001	static int vmdkDescDDBSetU32(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
2002	const char *pszKey, uint32_t uValue)
2003	{
2004	char *pszValue;
2005
2006	RTStrAPrintf(&pszValue, "\"%d\"", uValue);
2007	if (!pszValue)
2008	return VERR_NO_STR_MEMORY;
2009	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
2010	pszValue);
2011	RTStrFree(pszValue);
2012	return rc;
2013	}
2014
2015	/**
2016	* Splits the descriptor data into individual lines checking for correct line
2017	* endings and descriptor size.
2018	*
2019	* @returns VBox status code.
2020	* @param pImage The image instance.
2021	* @param pDesc The descriptor.
2022	* @param pszTmp The raw descriptor data from the image.
2023	*/
2024	static int vmdkDescSplitLines(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDesc, char *pszTmp)
2025	{
2026	unsigned cLine = 0;
2027	int rc = VINF_SUCCESS;
2028
2029	while ( RT_SUCCESS(rc)
2030	&& *pszTmp != '\0')
2031	{
2032	pDesc->aLines[cLine++] = pszTmp;
2033	if (cLine >= VMDK_DESCRIPTOR_LINES_MAX)
2034	{
2035	vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
2036	rc = VERR_VD_VMDK_INVALID_HEADER;
2037	break;
2038	}
2039
2040	while (pszTmp != '\0' && pszTmp != '\n')
2041	{
2042	if (*pszTmp == '\r')
2043	{
2044	if (*(pszTmp + 1) != '\n')
2045	{
2046	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: unsupported end of line in descriptor in '%s'"), pImage->pszFilename);
2047	break;
2048	}
2049	else
2050	{
2051	/* Get rid of CR character. */
2052	*pszTmp = '\0';
2053	}
2054	}
2055	pszTmp++;
2056	}
2057
2058	if (RT_FAILURE(rc))
2059	break;
2060
2061	/* Get rid of LF character. */
2062	if (*pszTmp == '\n')
2063	{
2064	*pszTmp = '\0';
2065	pszTmp++;
2066	}
2067	}
2068
2069	if (RT_SUCCESS(rc))
2070	{
2071	pDesc->cLines = cLine;
2072	/* Pointer right after the end of the used part of the buffer. */
2073	pDesc->aLines[cLine] = pszTmp;
2074	}
2075
2076	return rc;
2077	}
2078
2079	static int vmdkPreprocessDescriptor(PVMDKIMAGE pImage, char *pDescData,
2080	size_t cbDescData, PVMDKDESCRIPTOR pDescriptor)
2081	{
2082	pDescriptor->cbDescAlloc = cbDescData;
2083	int rc = vmdkDescSplitLines(pImage, pDescriptor, pDescData);
2084	if (RT_SUCCESS(rc))
2085	{
2086	if ( strcmp(pDescriptor->aLines[0], "# Disk DescriptorFile")
2087	&& strcmp(pDescriptor->aLines[0], "# Disk Descriptor File")
2088	&& strcmp(pDescriptor->aLines[0], "#Disk Descriptor File")
2089	&& strcmp(pDescriptor->aLines[0], "#Disk DescriptorFile"))
2090	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2091	N_("VMDK: descriptor does not start as expected in '%s'"), pImage->pszFilename);
2092	else
2093	{
2094	unsigned uLastNonEmptyLine = 0;
2095
2096	/* Initialize those, because we need to be able to reopen an image. */
2097	pDescriptor->uFirstDesc = 0;
2098	pDescriptor->uFirstExtent = 0;
2099	pDescriptor->uFirstDDB = 0;
2100	for (unsigned i = 0; i < pDescriptor->cLines; i++)
2101	{
2102	if (pDescriptor->aLines[i] != '#' && pDescriptor->aLines[i] != '\0')
2103	{
2104	if ( !strncmp(pDescriptor->aLines[i], "RW", 2)
2105	\|\| !strncmp(pDescriptor->aLines[i], "RDONLY", 6)
2106	\|\| !strncmp(pDescriptor->aLines[i], "NOACCESS", 8) )
2107	{
2108	/* An extent descriptor. */
2109	if (!pDescriptor->uFirstDesc \|\| pDescriptor->uFirstDDB)
2110	{
2111	/* Incorrect ordering of entries. */
2112	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2113	N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
2114	break;
2115	}
2116	if (!pDescriptor->uFirstExtent)
2117	{
2118	pDescriptor->uFirstExtent = i;
2119	uLastNonEmptyLine = 0;
2120	}
2121	}
2122	else if (!strncmp(pDescriptor->aLines[i], "ddb.", 4))
2123	{
2124	/* A disk database entry. */
2125	if (!pDescriptor->uFirstDesc \|\| !pDescriptor->uFirstExtent)
2126	{
2127	/* Incorrect ordering of entries. */
2128	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2129	N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
2130	break;
2131	}
2132	if (!pDescriptor->uFirstDDB)
2133	{
2134	pDescriptor->uFirstDDB = i;
2135	uLastNonEmptyLine = 0;
2136	}
2137	}
2138	else
2139	{
2140	/* A normal entry. */
2141	if (pDescriptor->uFirstExtent \|\| pDescriptor->uFirstDDB)
2142	{
2143	/* Incorrect ordering of entries. */
2144	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2145	N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
2146	break;
2147	}
2148	if (!pDescriptor->uFirstDesc)
2149	{
2150	pDescriptor->uFirstDesc = i;
2151	uLastNonEmptyLine = 0;
2152	}
2153	}
2154	if (uLastNonEmptyLine)
2155	pDescriptor->aNextLines[uLastNonEmptyLine] = i;
2156	uLastNonEmptyLine = i;
2157	}
2158	}
2159	}
2160	}
2161
2162	return rc;
2163	}
2164
2165	static int vmdkDescSetPCHSGeometry(PVMDKIMAGE pImage,
2166	PCVDGEOMETRY pPCHSGeometry)
2167	{
2168	int rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2169	VMDK_DDB_GEO_PCHS_CYLINDERS,
2170	pPCHSGeometry->cCylinders);
2171	if (RT_FAILURE(rc))
2172	return rc;
2173	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2174	VMDK_DDB_GEO_PCHS_HEADS,
2175	pPCHSGeometry->cHeads);
2176	if (RT_FAILURE(rc))
2177	return rc;
2178	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2179	VMDK_DDB_GEO_PCHS_SECTORS,
2180	pPCHSGeometry->cSectors);
2181	return rc;
2182	}
2183
2184	static int vmdkDescSetLCHSGeometry(PVMDKIMAGE pImage,
2185	PCVDGEOMETRY pLCHSGeometry)
2186	{
2187	int rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2188	VMDK_DDB_GEO_LCHS_CYLINDERS,
2189	pLCHSGeometry->cCylinders);
2190	if (RT_FAILURE(rc))
2191	return rc;
2192	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2193	VMDK_DDB_GEO_LCHS_HEADS,
2194
2195	pLCHSGeometry->cHeads);
2196	if (RT_FAILURE(rc))
2197	return rc;
2198	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2199	VMDK_DDB_GEO_LCHS_SECTORS,
2200	pLCHSGeometry->cSectors);
2201	return rc;
2202	}
2203
2204	static int vmdkCreateDescriptor(PVMDKIMAGE pImage, char *pDescData,
2205	size_t cbDescData, PVMDKDESCRIPTOR pDescriptor)
2206	{
2207	pDescriptor->uFirstDesc = 0;
2208	pDescriptor->uFirstExtent = 0;
2209	pDescriptor->uFirstDDB = 0;
2210	pDescriptor->cLines = 0;
2211	pDescriptor->cbDescAlloc = cbDescData;
2212	pDescriptor->fDirty = false;
2213	pDescriptor->aLines[pDescriptor->cLines] = pDescData;
2214	memset(pDescriptor->aNextLines, '\0', sizeof(pDescriptor->aNextLines));
2215
2216	int rc = vmdkDescInitStr(pImage, pDescriptor, "# Disk DescriptorFile");
2217	if (RT_SUCCESS(rc))
2218	rc = vmdkDescInitStr(pImage, pDescriptor, "version=1");
2219	if (RT_SUCCESS(rc))
2220	{
2221	pDescriptor->uFirstDesc = pDescriptor->cLines - 1;
2222	rc = vmdkDescInitStr(pImage, pDescriptor, "");
2223	}
2224	if (RT_SUCCESS(rc))
2225	rc = vmdkDescInitStr(pImage, pDescriptor, "# Extent description");
2226	if (RT_SUCCESS(rc))
2227	rc = vmdkDescInitStr(pImage, pDescriptor, "NOACCESS 0 ZERO ");
2228	if (RT_SUCCESS(rc))
2229	{
2230	pDescriptor->uFirstExtent = pDescriptor->cLines - 1;
2231	rc = vmdkDescInitStr(pImage, pDescriptor, "");
2232	}
2233	if (RT_SUCCESS(rc))
2234	{
2235	/* The trailing space is created by VMware, too. */
2236	rc = vmdkDescInitStr(pImage, pDescriptor, "# The disk Data Base ");
2237	}
2238	if (RT_SUCCESS(rc))
2239	rc = vmdkDescInitStr(pImage, pDescriptor, "#DDB");
2240	if (RT_SUCCESS(rc))
2241	rc = vmdkDescInitStr(pImage, pDescriptor, "");
2242	if (RT_SUCCESS(rc))
2243	rc = vmdkDescInitStr(pImage, pDescriptor, "ddb.virtualHWVersion = \"4\"");
2244	if (RT_SUCCESS(rc))
2245	{
2246	pDescriptor->uFirstDDB = pDescriptor->cLines - 1;
2247
2248	/* Now that the framework is in place, use the normal functions to insert
2249	* the remaining keys. */
2250	char szBuf[9];
2251	RTStrPrintf(szBuf, sizeof(szBuf), "%08x", RTRandU32());
2252	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc,
2253	"CID", szBuf);
2254	}
2255	if (RT_SUCCESS(rc))
2256	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc,
2257	"parentCID", "ffffffff");
2258	if (RT_SUCCESS(rc))
2259	rc = vmdkDescDDBSetStr(pImage, pDescriptor, "ddb.adapterType", "ide");
2260
2261	return rc;
2262	}
2263
2264	static int vmdkParseDescriptor(PVMDKIMAGE pImage, char *pDescData, size_t cbDescData)
2265	{
2266	int rc;
2267	unsigned cExtents;
2268	unsigned uLine;
2269	unsigned i;
2270
2271	rc = vmdkPreprocessDescriptor(pImage, pDescData, cbDescData,
2272	&pImage->Descriptor);
2273	if (RT_FAILURE(rc))
2274	return rc;
2275
2276	/* Check version, must be 1. */
2277	uint32_t uVersion;
2278	rc = vmdkDescBaseGetU32(&pImage->Descriptor, "version", &uVersion);
2279	if (RT_FAILURE(rc))
2280	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error finding key 'version' in descriptor in '%s'"), pImage->pszFilename);
2281	if (uVersion != 1)
2282	return vdIfError(pImage->pIfError, VERR_VD_VMDK_UNSUPPORTED_VERSION, RT_SRC_POS, N_("VMDK: unsupported format version in descriptor in '%s'"), pImage->pszFilename);
2283
2284	/* Get image creation type and determine image flags. */
2285	char pszCreateType = NULL; / initialized to make gcc shut up */
2286	rc = vmdkDescBaseGetStr(pImage, &pImage->Descriptor, "createType",
2287	&pszCreateType);
2288	if (RT_FAILURE(rc))
2289	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot get image type from descriptor in '%s'"), pImage->pszFilename);
2290	if ( !strcmp(pszCreateType, "twoGbMaxExtentSparse")
2291	\|\| !strcmp(pszCreateType, "twoGbMaxExtentFlat"))
2292	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_SPLIT_2G;
2293	else if ( !strcmp(pszCreateType, "partitionedDevice")
2294	\|\| !strcmp(pszCreateType, "fullDevice"))
2295	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_RAWDISK;
2296	else if (!strcmp(pszCreateType, "streamOptimized"))
2297	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED;
2298	else if (!strcmp(pszCreateType, "vmfs"))
2299	pImage->uImageFlags \|= VD_IMAGE_FLAGS_FIXED \| VD_VMDK_IMAGE_FLAGS_ESX;
2300	RTMemTmpFree(pszCreateType);
2301
2302	/* Count the number of extent config entries. */
2303	for (uLine = pImage->Descriptor.uFirstExtent, cExtents = 0;
2304	uLine != 0;
2305	uLine = pImage->Descriptor.aNextLines[uLine], cExtents++)
2306	/* nothing */;
2307
2308	if (!pImage->pDescData && cExtents != 1)
2309	{
2310	/* Monolithic image, must have only one extent (already opened). */
2311	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);
2312	}
2313
2314	if (pImage->pDescData)
2315	{
2316	/* Non-monolithic image, extents need to be allocated. */
2317	rc = vmdkCreateExtents(pImage, cExtents);
2318	if (RT_FAILURE(rc))
2319	return rc;
2320	}
2321
2322	for (i = 0, uLine = pImage->Descriptor.uFirstExtent;
2323	i < cExtents; i++, uLine = pImage->Descriptor.aNextLines[uLine])
2324	{
2325	char *pszLine = pImage->Descriptor.aLines[uLine];
2326
2327	/* Access type of the extent. */
2328	if (!strncmp(pszLine, "RW", 2))
2329	{
2330	pImage->pExtents[i].enmAccess = VMDKACCESS_READWRITE;
2331	pszLine += 2;
2332	}
2333	else if (!strncmp(pszLine, "RDONLY", 6))
2334	{
2335	pImage->pExtents[i].enmAccess = VMDKACCESS_READONLY;
2336	pszLine += 6;
2337	}
2338	else if (!strncmp(pszLine, "NOACCESS", 8))
2339	{
2340	pImage->pExtents[i].enmAccess = VMDKACCESS_NOACCESS;
2341	pszLine += 8;
2342	}
2343	else
2344	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2345	if (*pszLine++ != ' ')
2346	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2347
2348	/* Nominal size of the extent. */
2349	rc = RTStrToUInt64Ex(pszLine, &pszLine, 10,
2350	&pImage->pExtents[i].cNominalSectors);
2351	if (RT_FAILURE(rc))
2352	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2353	if (*pszLine++ != ' ')
2354	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2355
2356	/* Type of the extent. */
2357	if (!strncmp(pszLine, "SPARSE", 6))
2358	{
2359	pImage->pExtents[i].enmType = VMDKETYPE_HOSTED_SPARSE;
2360	pszLine += 6;
2361	}
2362	else if (!strncmp(pszLine, "FLAT", 4))
2363	{
2364	pImage->pExtents[i].enmType = VMDKETYPE_FLAT;
2365	pszLine += 4;
2366	}
2367	else if (!strncmp(pszLine, "ZERO", 4))
2368	{
2369	pImage->pExtents[i].enmType = VMDKETYPE_ZERO;
2370	pszLine += 4;
2371	}
2372	else if (!strncmp(pszLine, "VMFS", 4))
2373	{
2374	pImage->pExtents[i].enmType = VMDKETYPE_VMFS;
2375	pszLine += 4;
2376	}
2377	else
2378	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2379
2380	if (pImage->pExtents[i].enmType == VMDKETYPE_ZERO)
2381	{
2382	/* This one has no basename or offset. */
2383	if (*pszLine == ' ')
2384	pszLine++;
2385	if (*pszLine != '\0')
2386	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2387	pImage->pExtents[i].pszBasename = NULL;
2388	}
2389	else
2390	{
2391	/* All other extent types have basename and optional offset. */
2392	if (*pszLine++ != ' ')
2393	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2394
2395	/* Basename of the image. Surrounded by quotes. */
2396	char *pszBasename;
2397	rc = vmdkStringUnquote(pImage, pszLine, &pszBasename, &pszLine);
2398	if (RT_FAILURE(rc))
2399	return rc;
2400	pImage->pExtents[i].pszBasename = pszBasename;
2401	if (*pszLine == ' ')
2402	{
2403	pszLine++;
2404	if (*pszLine != '\0')
2405	{
2406	/* Optional offset in extent specified. */
2407	rc = RTStrToUInt64Ex(pszLine, &pszLine, 10,
2408	&pImage->pExtents[i].uSectorOffset);
2409	if (RT_FAILURE(rc))
2410	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2411	}
2412	}
2413
2414	if (*pszLine != '\0')
2415	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2416	}
2417	}
2418
2419	/* Determine PCHS geometry (autogenerate if necessary). */
2420	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2421	VMDK_DDB_GEO_PCHS_CYLINDERS,
2422	&pImage->PCHSGeometry.cCylinders);
2423	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2424	pImage->PCHSGeometry.cCylinders = 0;
2425	else if (RT_FAILURE(rc))
2426	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2427	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2428	VMDK_DDB_GEO_PCHS_HEADS,
2429	&pImage->PCHSGeometry.cHeads);
2430	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2431	pImage->PCHSGeometry.cHeads = 0;
2432	else if (RT_FAILURE(rc))
2433	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2434	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2435	VMDK_DDB_GEO_PCHS_SECTORS,
2436	&pImage->PCHSGeometry.cSectors);
2437	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2438	pImage->PCHSGeometry.cSectors = 0;
2439	else if (RT_FAILURE(rc))
2440	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2441	if ( pImage->PCHSGeometry.cCylinders == 0
2442	\|\| pImage->PCHSGeometry.cHeads == 0
2443	\|\| pImage->PCHSGeometry.cHeads > 16
2444	\|\| pImage->PCHSGeometry.cSectors == 0
2445	\|\| pImage->PCHSGeometry.cSectors > 63)
2446	{
2447	/* Mark PCHS geometry as not yet valid (can't do the calculation here
2448	* as the total image size isn't known yet). */
2449	pImage->PCHSGeometry.cCylinders = 0;
2450	pImage->PCHSGeometry.cHeads = 16;
2451	pImage->PCHSGeometry.cSectors = 63;
2452	}
2453
2454	/* Determine LCHS geometry (set to 0 if not specified). */
2455	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2456	VMDK_DDB_GEO_LCHS_CYLINDERS,
2457	&pImage->LCHSGeometry.cCylinders);
2458	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2459	pImage->LCHSGeometry.cCylinders = 0;
2460	else if (RT_FAILURE(rc))
2461	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2462	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2463	VMDK_DDB_GEO_LCHS_HEADS,
2464	&pImage->LCHSGeometry.cHeads);
2465	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2466	pImage->LCHSGeometry.cHeads = 0;
2467	else if (RT_FAILURE(rc))
2468	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2469	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2470	VMDK_DDB_GEO_LCHS_SECTORS,
2471	&pImage->LCHSGeometry.cSectors);
2472	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2473	pImage->LCHSGeometry.cSectors = 0;
2474	else if (RT_FAILURE(rc))
2475	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2476	if ( pImage->LCHSGeometry.cCylinders == 0
2477	\|\| pImage->LCHSGeometry.cHeads == 0
2478	\|\| pImage->LCHSGeometry.cSectors == 0)
2479	{
2480	pImage->LCHSGeometry.cCylinders = 0;
2481	pImage->LCHSGeometry.cHeads = 0;
2482	pImage->LCHSGeometry.cSectors = 0;
2483	}
2484
2485	/* Get image UUID. */
2486	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor, VMDK_DDB_IMAGE_UUID,
2487	&pImage->ImageUuid);
2488	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2489	{
2490	/* Image without UUID. Probably created by VMware and not yet used
2491	* by VirtualBox. Can only be added for images opened in read/write
2492	* mode, so don't bother producing a sensible UUID otherwise. */
2493	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2494	RTUuidClear(&pImage->ImageUuid);
2495	else
2496	{
2497	rc = RTUuidCreate(&pImage->ImageUuid);
2498	if (RT_FAILURE(rc))
2499	return rc;
2500	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2501	VMDK_DDB_IMAGE_UUID, &pImage->ImageUuid);
2502	if (RT_FAILURE(rc))
2503	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image UUID in descriptor in '%s'"), pImage->pszFilename);
2504	}
2505	}
2506	else if (RT_FAILURE(rc))
2507	return rc;
2508
2509	/* Get image modification UUID. */
2510	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor,
2511	VMDK_DDB_MODIFICATION_UUID,
2512	&pImage->ModificationUuid);
2513	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2514	{
2515	/* Image without UUID. Probably created by VMware and not yet used
2516	* by VirtualBox. Can only be added for images opened in read/write
2517	* mode, so don't bother producing a sensible UUID otherwise. */
2518	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2519	RTUuidClear(&pImage->ModificationUuid);
2520	else
2521	{
2522	rc = RTUuidCreate(&pImage->ModificationUuid);
2523	if (RT_FAILURE(rc))
2524	return rc;
2525	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2526	VMDK_DDB_MODIFICATION_UUID,
2527	&pImage->ModificationUuid);
2528	if (RT_FAILURE(rc))
2529	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image modification UUID in descriptor in '%s'"), pImage->pszFilename);
2530	}
2531	}
2532	else if (RT_FAILURE(rc))
2533	return rc;
2534
2535	/* Get UUID of parent image. */
2536	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor, VMDK_DDB_PARENT_UUID,
2537	&pImage->ParentUuid);
2538	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2539	{
2540	/* Image without UUID. Probably created by VMware and not yet used
2541	* by VirtualBox. Can only be added for images opened in read/write
2542	* mode, so don't bother producing a sensible UUID otherwise. */
2543	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2544	RTUuidClear(&pImage->ParentUuid);
2545	else
2546	{
2547	rc = RTUuidClear(&pImage->ParentUuid);
2548	if (RT_FAILURE(rc))
2549	return rc;
2550	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2551	VMDK_DDB_PARENT_UUID, &pImage->ParentUuid);
2552	if (RT_FAILURE(rc))
2553	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent UUID in descriptor in '%s'"), pImage->pszFilename);
2554	}
2555	}
2556	else if (RT_FAILURE(rc))
2557	return rc;
2558
2559	/* Get parent image modification UUID. */
2560	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor,
2561	VMDK_DDB_PARENT_MODIFICATION_UUID,
2562	&pImage->ParentModificationUuid);
2563	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2564	{
2565	/* Image without UUID. Probably created by VMware and not yet used
2566	* by VirtualBox. Can only be added for images opened in read/write
2567	* mode, so don't bother producing a sensible UUID otherwise. */
2568	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2569	RTUuidClear(&pImage->ParentModificationUuid);
2570	else
2571	{
2572	RTUuidClear(&pImage->ParentModificationUuid);
2573	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2574	VMDK_DDB_PARENT_MODIFICATION_UUID,
2575	&pImage->ParentModificationUuid);
2576	if (RT_FAILURE(rc))
2577	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent modification UUID in descriptor in '%s'"), pImage->pszFilename);
2578	}
2579	}
2580	else if (RT_FAILURE(rc))
2581	return rc;
2582
2583	return VINF_SUCCESS;
2584	}
2585
2586	/**
2587	* Internal : Prepares the descriptor to write to the image.
2588	*/
2589	static int vmdkDescriptorPrepare(PVMDKIMAGE pImage, uint64_t cbLimit,
2590	void *ppvData, size_t pcbData)
2591	{
2592	int rc = VINF_SUCCESS;
2593
2594	/*
2595	* Allocate temporary descriptor buffer.
2596	* In case there is no limit allocate a default
2597	* and increase if required.
2598	*/
2599	size_t cbDescriptor = cbLimit ? cbLimit : 4 * _1K;
2600	char pszDescriptor = (char )RTMemAllocZ(cbDescriptor);
2601	size_t offDescriptor = 0;
2602
2603	if (!pszDescriptor)
2604	return VERR_NO_MEMORY;
2605
2606	for (unsigned i = 0; i < pImage->Descriptor.cLines; i++)
2607	{
2608	const char *psz = pImage->Descriptor.aLines[i];
2609	size_t cb = strlen(psz);
2610
2611	/*
2612	* Increase the descriptor if there is no limit and
2613	* there is not enough room left for this line.
2614	*/
2615	if (offDescriptor + cb + 1 > cbDescriptor)
2616	{
2617	if (cbLimit)
2618	{
2619	rc = vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too long in '%s'"), pImage->pszFilename);
2620	break;
2621	}
2622	else
2623	{
2624	char *pszDescriptorNew = NULL;
2625	LogFlow(("Increasing descriptor cache\n"));
2626
2627	pszDescriptorNew = (char )RTMemRealloc(pszDescriptor, cbDescriptor + cb + 4 _1K);
2628	if (!pszDescriptorNew)
2629	{
2630	rc = VERR_NO_MEMORY;
2631	break;
2632	}
2633	pszDescriptor = pszDescriptorNew;
2634	cbDescriptor += cb + 4 * _1K;
2635	}
2636	}
2637
2638	if (cb > 0)
2639	{
2640	memcpy(pszDescriptor + offDescriptor, psz, cb);
2641	offDescriptor += cb;
2642	}
2643
2644	memcpy(pszDescriptor + offDescriptor, "\n", 1);
2645	offDescriptor++;
2646	}
2647
2648	if (RT_SUCCESS(rc))
2649	{
2650	*ppvData = pszDescriptor;
2651	*pcbData = offDescriptor;
2652	}
2653	else if (pszDescriptor)
2654	RTMemFree(pszDescriptor);
2655
2656	return rc;
2657	}
2658
2659	/**
2660	* Internal: write/update the descriptor part of the image.
2661	*/
2662	static int vmdkWriteDescriptor(PVMDKIMAGE pImage, PVDIOCTX pIoCtx)
2663	{
2664	int rc = VINF_SUCCESS;
2665	uint64_t cbLimit;
2666	uint64_t uOffset;
2667	PVMDKFILE pDescFile;
2668	void *pvDescriptor = NULL;
2669	size_t cbDescriptor;
2670
2671	if (pImage->pDescData)
2672	{
2673	/* Separate descriptor file. */
2674	uOffset = 0;
2675	cbLimit = 0;
2676	pDescFile = pImage->pFile;
2677	}
2678	else
2679	{
2680	/* Embedded descriptor file. */
2681	uOffset = VMDK_SECTOR2BYTE(pImage->pExtents[0].uDescriptorSector);
2682	cbLimit = VMDK_SECTOR2BYTE(pImage->pExtents[0].cDescriptorSectors);
2683	pDescFile = pImage->pExtents[0].pFile;
2684	}
2685	/* Bail out if there is no file to write to. */
2686	if (pDescFile == NULL)
2687	return VERR_INVALID_PARAMETER;
2688
2689	rc = vmdkDescriptorPrepare(pImage, cbLimit, &pvDescriptor, &cbDescriptor);
2690	if (RT_SUCCESS(rc))
2691	{
2692	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pDescFile->pStorage,
2693	uOffset, pvDescriptor,
2694	cbLimit ? cbLimit : cbDescriptor,
2695	pIoCtx, NULL, NULL);
2696	if ( RT_FAILURE(rc)
2697	&& rc != VERR_VD_ASYNC_IO_IN_PROGRESS)
2698	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error writing descriptor in '%s'"), pImage->pszFilename);
2699	}
2700
2701	if (RT_SUCCESS(rc) && !cbLimit)
2702	{
2703	rc = vdIfIoIntFileSetSize(pImage->pIfIo, pDescFile->pStorage, cbDescriptor);
2704	if (RT_FAILURE(rc))
2705	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error truncating descriptor in '%s'"), pImage->pszFilename);
2706	}
2707
2708	if (RT_SUCCESS(rc))
2709	pImage->Descriptor.fDirty = false;
2710
2711	if (pvDescriptor)
2712	RTMemFree(pvDescriptor);
2713	return rc;
2714
2715	}
2716
2717	/**
2718	* Internal: validate the consistency check values in a binary header.
2719	*/
2720	static int vmdkValidateHeader(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, const SparseExtentHeader *pHeader)
2721	{
2722	int rc = VINF_SUCCESS;
2723	if (RT_LE2H_U32(pHeader->magicNumber) != VMDK_SPARSE_MAGICNUMBER)
2724	{
2725	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrect magic in sparse extent header in '%s'"), pExtent->pszFullname);
2726	return rc;
2727	}
2728	if (RT_LE2H_U32(pHeader->version) != 1 && RT_LE2H_U32(pHeader->version) != 3)
2729	{
2730	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);
2731	return rc;
2732	}
2733	if ( (RT_LE2H_U32(pHeader->flags) & 1)
2734	&& ( pHeader->singleEndLineChar != '\n'
2735	\|\| pHeader->nonEndLineChar != ' '
2736	\|\| pHeader->doubleEndLineChar1 != '\r'
2737	\|\| pHeader->doubleEndLineChar2 != '\n') )
2738	{
2739	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: corrupted by CR/LF translation in '%s'"), pExtent->pszFullname);
2740	return rc;
2741	}
2742	if (RT_LE2H_U64(pHeader->descriptorSize) > VMDK_SPARSE_DESCRIPTOR_SIZE_MAX)
2743	{
2744	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor size out of bounds (%llu vs %llu) '%s'"),
2745	pExtent->pszFullname, RT_LE2H_U64(pHeader->descriptorSize), VMDK_SPARSE_DESCRIPTOR_SIZE_MAX);
2746	return rc;
2747	}
2748	return rc;
2749	}
2750
2751	/**
2752	* Internal: read metadata belonging to an extent with binary header, i.e.
2753	* as found in monolithic files.
2754	*/
2755	static int vmdkReadBinaryMetaExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
2756	bool fMagicAlreadyRead)
2757	{
2758	SparseExtentHeader Header;
2759	int rc;
2760
2761	if (!fMagicAlreadyRead)
2762	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage, 0,
2763	&Header, sizeof(Header));
2764	else
2765	{
2766	Header.magicNumber = RT_H2LE_U32(VMDK_SPARSE_MAGICNUMBER);
2767	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
2768	RT_UOFFSETOF(SparseExtentHeader, version),
2769	&Header.version,
2770	sizeof(Header)
2771	- RT_UOFFSETOF(SparseExtentHeader, version));
2772	}
2773
2774	if (RT_SUCCESS(rc))
2775	{
2776	rc = vmdkValidateHeader(pImage, pExtent, &Header);
2777	if (RT_SUCCESS(rc))
2778	{
2779	uint64_t cbFile = 0;
2780
2781	if ( (RT_LE2H_U32(Header.flags) & RT_BIT(17))
2782	&& RT_LE2H_U64(Header.gdOffset) == VMDK_GD_AT_END)
2783	pExtent->fFooter = true;
2784
2785	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2786	\|\| ( pExtent->fFooter
2787	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2788	{
2789	rc = vdIfIoIntFileGetSize(pImage->pIfIo, pExtent->pFile->pStorage, &cbFile);
2790	if (RT_FAILURE(rc))
2791	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot get size of '%s'"), pExtent->pszFullname);
2792	}
2793
2794	if (RT_SUCCESS(rc))
2795	{
2796	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
2797	pExtent->uAppendPosition = RT_ALIGN_64(cbFile, 512);
2798
2799	if ( pExtent->fFooter
2800	&& ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2801	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2802	{
2803	/* Read the footer, which comes before the end-of-stream marker. */
2804	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
2805	cbFile - 2*512, &Header,
2806	sizeof(Header));
2807	if (RT_FAILURE(rc))
2808	{
2809	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading extent footer in '%s'"), pExtent->pszFullname);
2810	rc = VERR_VD_VMDK_INVALID_HEADER;
2811	}
2812
2813	if (RT_SUCCESS(rc))
2814	rc = vmdkValidateHeader(pImage, pExtent, &Header);
2815	/* Prohibit any writes to this extent. */
2816	pExtent->uAppendPosition = 0;
2817	}
2818
2819	if (RT_SUCCESS(rc))
2820	{
2821	pExtent->uVersion = RT_LE2H_U32(Header.version);
2822	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE; /* Just dummy value, changed later. */
2823	pExtent->cSectors = RT_LE2H_U64(Header.capacity);
2824	pExtent->cSectorsPerGrain = RT_LE2H_U64(Header.grainSize);
2825	pExtent->uDescriptorSector = RT_LE2H_U64(Header.descriptorOffset);
2826	pExtent->cDescriptorSectors = RT_LE2H_U64(Header.descriptorSize);
2827	pExtent->cGTEntries = RT_LE2H_U32(Header.numGTEsPerGT);
2828	pExtent->cOverheadSectors = RT_LE2H_U64(Header.overHead);
2829	pExtent->fUncleanShutdown = !!Header.uncleanShutdown;
2830	pExtent->uCompression = RT_LE2H_U16(Header.compressAlgorithm);
2831	if (RT_LE2H_U32(Header.flags) & RT_BIT(1))
2832	{
2833	pExtent->uSectorRGD = RT_LE2H_U64(Header.rgdOffset);
2834	pExtent->uSectorGD = RT_LE2H_U64(Header.gdOffset);
2835	}
2836	else
2837	{
2838	pExtent->uSectorGD = RT_LE2H_U64(Header.gdOffset);
2839	pExtent->uSectorRGD = 0;
2840	}
2841
2842	if (pExtent->uDescriptorSector && !pExtent->cDescriptorSectors)
2843	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2844	N_("VMDK: inconsistent embedded descriptor config in '%s'"), pExtent->pszFullname);
2845
2846	if ( RT_SUCCESS(rc)
2847	&& ( pExtent->uSectorGD == VMDK_GD_AT_END
2848	\|\| pExtent->uSectorRGD == VMDK_GD_AT_END)
2849	&& ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2850	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2851	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2852	N_("VMDK: cannot resolve grain directory offset in '%s'"), pExtent->pszFullname);
2853
2854	if (RT_SUCCESS(rc))
2855	{
2856	uint64_t cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
2857	if (!cSectorsPerGDE \|\| cSectorsPerGDE > UINT32_MAX)
2858	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2859	N_("VMDK: incorrect grain directory size in '%s'"), pExtent->pszFullname);
2860	else
2861	{
2862	pExtent->cSectorsPerGDE = cSectorsPerGDE;
2863	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
2864
2865	/* Fix up the number of descriptor sectors, as some flat images have
2866	* really just one, and this causes failures when inserting the UUID
2867	* values and other extra information. */
2868	if (pExtent->cDescriptorSectors != 0 && pExtent->cDescriptorSectors < 4)
2869	{
2870	/* Do it the easy way - just fix it for flat images which have no
2871	* other complicated metadata which needs space too. */
2872	if ( pExtent->uDescriptorSector + 4 < pExtent->cOverheadSectors
2873	&& pExtent->cGTEntries * pExtent->cGDEntries == 0)
2874	pExtent->cDescriptorSectors = 4;
2875	}
2876	}
2877	}
2878	}
2879	}
2880	}
2881	}
2882	else
2883	{
2884	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading extent header in '%s'"), pExtent->pszFullname);
2885	rc = VERR_VD_VMDK_INVALID_HEADER;
2886	}
2887
2888	if (RT_FAILURE(rc))
2889	vmdkFreeExtentData(pImage, pExtent, false);
2890
2891	return rc;
2892	}
2893
2894	/**
2895	* Internal: read additional metadata belonging to an extent. For those
2896	* extents which have no additional metadata just verify the information.
2897	*/
2898	static int vmdkReadMetaExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
2899	{
2900	int rc = VINF_SUCCESS;
2901
2902	/* disabled the check as there are too many truncated vmdk images out there */
2903	#ifdef VBOX_WITH_VMDK_STRICT_SIZE_CHECK
2904	uint64_t cbExtentSize;
2905	/* The image must be a multiple of a sector in size and contain the data
2906	* area (flat images only). If not, it means the image is at least
2907	* truncated, or even seriously garbled. */
2908	rc = vdIfIoIntFileGetSize(pImage->pIfIo, pExtent->pFile->pStorage, &cbExtentSize);
2909	if (RT_FAILURE(rc))
2910	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting size in '%s'"), pExtent->pszFullname);
2911	else if ( cbExtentSize != RT_ALIGN_64(cbExtentSize, 512)
2912	&& (pExtent->enmType != VMDKETYPE_FLAT \|\| pExtent->cNominalSectors + pExtent->uSectorOffset > VMDK_BYTE2SECTOR(cbExtentSize)))
2913	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2914	N_("VMDK: file size is not a multiple of 512 in '%s', file is truncated or otherwise garbled"), pExtent->pszFullname);
2915	#endif /* VBOX_WITH_VMDK_STRICT_SIZE_CHECK */
2916	if ( RT_SUCCESS(rc)
2917	&& pExtent->enmType == VMDKETYPE_HOSTED_SPARSE)
2918	{
2919	/* The spec says that this must be a power of two and greater than 8,
2920	* but probably they meant not less than 8. */
2921	if ( (pExtent->cSectorsPerGrain & (pExtent->cSectorsPerGrain - 1))
2922	\|\| pExtent->cSectorsPerGrain < 8)
2923	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2924	N_("VMDK: invalid extent grain size %u in '%s'"), pExtent->cSectorsPerGrain, pExtent->pszFullname);
2925	else
2926	{
2927	/* This code requires that a grain table must hold a power of two multiple
2928	* of the number of entries per GT cache entry. */
2929	if ( (pExtent->cGTEntries & (pExtent->cGTEntries - 1))
2930	\|\| pExtent->cGTEntries < VMDK_GT_CACHELINE_SIZE)
2931	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2932	N_("VMDK: grain table cache size problem in '%s'"), pExtent->pszFullname);
2933	else
2934	{
2935	rc = vmdkAllocStreamBuffers(pImage, pExtent);
2936	if (RT_SUCCESS(rc))
2937	{
2938	/* Prohibit any writes to this streamOptimized extent. */
2939	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2940	pExtent->uAppendPosition = 0;
2941
2942	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2943	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2944	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
2945	rc = vmdkReadGrainDirectory(pImage, pExtent);
2946	else
2947	{
2948	pExtent->uGrainSectorAbs = pExtent->cOverheadSectors;
2949	pExtent->cbGrainStreamRead = 0;
2950	}
2951	}
2952	}
2953	}
2954	}
2955
2956	if (RT_FAILURE(rc))
2957	vmdkFreeExtentData(pImage, pExtent, false);
2958
2959	return rc;
2960	}
2961
2962	/**
2963	* Internal: write/update the metadata for a sparse extent.
2964	*/
2965	static int vmdkWriteMetaSparseExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
2966	uint64_t uOffset, PVDIOCTX pIoCtx)
2967	{
2968	SparseExtentHeader Header;
2969
2970	memset(&Header, '\0', sizeof(Header));
2971	Header.magicNumber = RT_H2LE_U32(VMDK_SPARSE_MAGICNUMBER);
2972	Header.version = RT_H2LE_U32(pExtent->uVersion);
2973	Header.flags = RT_H2LE_U32(RT_BIT(0));
2974	if (pExtent->pRGD)
2975	Header.flags \|= RT_H2LE_U32(RT_BIT(1));
2976	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2977	Header.flags \|= RT_H2LE_U32(RT_BIT(16) \| RT_BIT(17));
2978	Header.capacity = RT_H2LE_U64(pExtent->cSectors);
2979	Header.grainSize = RT_H2LE_U64(pExtent->cSectorsPerGrain);
2980	Header.descriptorOffset = RT_H2LE_U64(pExtent->uDescriptorSector);
2981	Header.descriptorSize = RT_H2LE_U64(pExtent->cDescriptorSectors);
2982	Header.numGTEsPerGT = RT_H2LE_U32(pExtent->cGTEntries);
2983	if (pExtent->fFooter && uOffset == 0)
2984	{
2985	if (pExtent->pRGD)
2986	{
2987	Assert(pExtent->uSectorRGD);
2988	Header.rgdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2989	Header.gdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2990	}
2991	else
2992	Header.gdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2993	}
2994	else
2995	{
2996	if (pExtent->pRGD)
2997	{
2998	Assert(pExtent->uSectorRGD);
2999	Header.rgdOffset = RT_H2LE_U64(pExtent->uSectorRGD);
3000	Header.gdOffset = RT_H2LE_U64(pExtent->uSectorGD);
3001	}
3002	else
3003	Header.gdOffset = RT_H2LE_U64(pExtent->uSectorGD);
3004	}
3005	Header.overHead = RT_H2LE_U64(pExtent->cOverheadSectors);
3006	Header.uncleanShutdown = pExtent->fUncleanShutdown;
3007	Header.singleEndLineChar = '\n';
3008	Header.nonEndLineChar = ' ';
3009	Header.doubleEndLineChar1 = '\r';
3010	Header.doubleEndLineChar2 = '\n';
3011	Header.compressAlgorithm = RT_H2LE_U16(pExtent->uCompression);
3012
3013	int rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
3014	uOffset, &Header, sizeof(Header),
3015	pIoCtx, NULL, NULL);
3016	if (RT_FAILURE(rc) && (rc != VERR_VD_ASYNC_IO_IN_PROGRESS))
3017	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error writing extent header in '%s'"), pExtent->pszFullname);
3018	return rc;
3019	}
3020
3021	/**
3022	* Internal: free the buffers used for streamOptimized images.
3023	*/
3024	static void vmdkFreeStreamBuffers(PVMDKEXTENT pExtent)
3025	{
3026	if (pExtent->pvCompGrain)
3027	{
3028	RTMemFree(pExtent->pvCompGrain);
3029	pExtent->pvCompGrain = NULL;
3030	}
3031	if (pExtent->pvGrain)
3032	{
3033	RTMemFree(pExtent->pvGrain);
3034	pExtent->pvGrain = NULL;
3035	}
3036	}
3037
3038	/**
3039	* Internal: free the memory used by the extent data structure, optionally
3040	* deleting the referenced files.
3041	*
3042	* @returns VBox status code.
3043	* @param pImage Pointer to the image instance data.
3044	* @param pExtent The extent to free.
3045	* @param fDelete Flag whether to delete the backing storage.
3046	*/
3047	static int vmdkFreeExtentData(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
3048	bool fDelete)
3049	{
3050	int rc = VINF_SUCCESS;
3051
3052	vmdkFreeGrainDirectory(pExtent);
3053	if (pExtent->pDescData)
3054	{
3055	RTMemFree(pExtent->pDescData);
3056	pExtent->pDescData = NULL;
3057	}
3058	if (pExtent->pFile != NULL)
3059	{
3060	/* Do not delete raw extents, these have full and base names equal. */
3061	rc = vmdkFileClose(pImage, &pExtent->pFile,
3062	fDelete
3063	&& pExtent->pszFullname
3064	&& pExtent->pszBasename
3065	&& strcmp(pExtent->pszFullname, pExtent->pszBasename));
3066	}
3067	if (pExtent->pszBasename)
3068	{
3069	RTMemTmpFree((void *)pExtent->pszBasename);
3070	pExtent->pszBasename = NULL;
3071	}
3072	if (pExtent->pszFullname)
3073	{
3074	RTStrFree((char )(void )pExtent->pszFullname);
3075	pExtent->pszFullname = NULL;
3076	}
3077	vmdkFreeStreamBuffers(pExtent);
3078
3079	return rc;
3080	}
3081
3082	/**
3083	* Internal: allocate grain table cache if necessary for this image.
3084	*/
3085	static int vmdkAllocateGrainTableCache(PVMDKIMAGE pImage)
3086	{
3087	PVMDKEXTENT pExtent;
3088
3089	/* Allocate grain table cache if any sparse extent is present. */
3090	for (unsigned i = 0; i < pImage->cExtents; i++)
3091	{
3092	pExtent = &pImage->pExtents[i];
3093	if (pExtent->enmType == VMDKETYPE_HOSTED_SPARSE)
3094	{
3095	/* Allocate grain table cache. */
3096	pImage->pGTCache = (PVMDKGTCACHE)RTMemAllocZ(sizeof(VMDKGTCACHE));
3097	if (!pImage->pGTCache)
3098	return VERR_NO_MEMORY;
3099	for (unsigned j = 0; j < VMDK_GT_CACHE_SIZE; j++)
3100	{
3101	PVMDKGTCACHEENTRY pGCE = &pImage->pGTCache->aGTCache[j];
3102	pGCE->uExtent = UINT32_MAX;
3103	}
3104	pImage->pGTCache->cEntries = VMDK_GT_CACHE_SIZE;
3105	break;
3106	}
3107	}
3108
3109	return VINF_SUCCESS;
3110	}
3111
3112	/**
3113	* Internal: allocate the given number of extents.
3114	*/
3115	static int vmdkCreateExtents(PVMDKIMAGE pImage, unsigned cExtents)
3116	{
3117	int rc = VINF_SUCCESS;
3118	PVMDKEXTENT pExtents = (PVMDKEXTENT)RTMemAllocZ(cExtents * sizeof(VMDKEXTENT));
3119	if (pExtents)
3120	{
3121	for (unsigned i = 0; i < cExtents; i++)
3122	{
3123	pExtents[i].pFile = NULL;
3124	pExtents[i].pszBasename = NULL;
3125	pExtents[i].pszFullname = NULL;
3126	pExtents[i].pGD = NULL;
3127	pExtents[i].pRGD = NULL;
3128	pExtents[i].pDescData = NULL;
3129	pExtents[i].uVersion = 1;
3130	pExtents[i].uCompression = VMDK_COMPRESSION_NONE;
3131	pExtents[i].uExtent = i;
3132	pExtents[i].pImage = pImage;
3133	}
3134	pImage->pExtents = pExtents;
3135	pImage->cExtents = cExtents;
3136	}
3137	else
3138	rc = VERR_NO_MEMORY;
3139
3140	return rc;
3141	}
3142
3143	/**
3144	* Internal: Create an additional file backed extent in split images.
3145	* Supports split sparse and flat images.
3146	*
3147	* @returns VBox status code.
3148	* @param pImage VMDK image instance.
3149	* @param cbSize Desiried size in bytes of new extent.
3150	*/
3151	static int vmdkAddFileBackedExtent(PVMDKIMAGE pImage, uint64_t cbSize)
3152	{
3153	int rc = VINF_SUCCESS;
3154	unsigned uImageFlags = pImage->uImageFlags;
3155
3156	/* Check for unsupported image type. */
3157	if ((uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
3158	\|\| (uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
3159	\|\| (uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK))
3160	{
3161	return VERR_NOT_SUPPORTED;
3162	}
3163
3164	/* Allocate array of extents and copy existing extents to it. */
3165	PVMDKEXTENT pNewExtents = (PVMDKEXTENT)RTMemAllocZ((pImage->cExtents + 1) * sizeof(VMDKEXTENT));
3166	if (!pNewExtents)
3167	{
3168	return VERR_NO_MEMORY;
3169	}
3170
3171	memcpy(pNewExtents, pImage->pExtents, pImage->cExtents * sizeof(VMDKEXTENT));
3172
3173	/* Locate newly created extent and populate default metadata. */
3174	PVMDKEXTENT pExtent = &pNewExtents[pImage->cExtents];
3175
3176	pExtent->pFile = NULL;
3177	pExtent->pszBasename = NULL;
3178	pExtent->pszFullname = NULL;
3179	pExtent->pGD = NULL;
3180	pExtent->pRGD = NULL;
3181	pExtent->pDescData = NULL;
3182	pExtent->uVersion = 1;
3183	pExtent->uCompression = VMDK_COMPRESSION_NONE;
3184	pExtent->uExtent = pImage->cExtents;
3185	pExtent->pImage = pImage;
3186	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
3187	pExtent->enmAccess = VMDKACCESS_READWRITE;
3188	pExtent->uSectorOffset = 0;
3189	pExtent->fMetaDirty = true;
3190
3191	/* Apply image type specific meta data. */
3192	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
3193	{
3194	pExtent->enmType = VMDKETYPE_FLAT;
3195	}
3196	else
3197	{
3198	uint64_t cSectorsPerGDE, cSectorsPerGD;
3199	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE;
3200	pExtent->cSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64(cbSize, _64K));
3201	pExtent->cSectorsPerGrain = VMDK_BYTE2SECTOR(_64K);
3202	pExtent->cGTEntries = 512;
3203	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
3204	pExtent->cSectorsPerGDE = cSectorsPerGDE;
3205	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
3206	cSectorsPerGD = (pExtent->cGDEntries + (512 / sizeof(uint32_t) - 1)) / (512 / sizeof(uint32_t));
3207	}
3208
3209	/* Allocate and set file name for extent. */
3210	char *pszBasenameSubstr = RTPathFilename(pImage->pszFilename);
3211	AssertPtr(pszBasenameSubstr);
3212
3213	char *pszBasenameSuff = RTPathSuffix(pszBasenameSubstr);
3214	char *pszBasenameBase = RTStrDup(pszBasenameSubstr);
3215	RTPathStripSuffix(pszBasenameBase);
3216	char *pszTmp;
3217	size_t cbTmp;
3218
3219	if (pImage->uImageFlags & VD_IMAGE_FLAGS_FIXED)
3220	RTStrAPrintf(&pszTmp, "%s-f%03d%s", pszBasenameBase,
3221	pExtent->uExtent + 1, pszBasenameSuff);
3222	else
3223	RTStrAPrintf(&pszTmp, "%s-s%03d%s", pszBasenameBase, pExtent->uExtent + 1,
3224	pszBasenameSuff);
3225
3226	RTStrFree(pszBasenameBase);
3227	if (!pszTmp)
3228	return VERR_NO_STR_MEMORY;
3229	cbTmp = strlen(pszTmp) + 1;
3230	char pszBasename = (char )RTMemTmpAlloc(cbTmp);
3231	if (!pszBasename)
3232	{
3233	RTStrFree(pszTmp);
3234	return VERR_NO_MEMORY;
3235	}
3236
3237	memcpy(pszBasename, pszTmp, cbTmp);
3238	RTStrFree(pszTmp);
3239
3240	pExtent->pszBasename = pszBasename;
3241
3242	char *pszBasedirectory = RTStrDup(pImage->pszFilename);
3243	if (!pszBasedirectory)
3244	return VERR_NO_STR_MEMORY;
3245	RTPathStripFilename(pszBasedirectory);
3246	char *pszFullname = RTPathJoinA(pszBasedirectory, pExtent->pszBasename);
3247	RTStrFree(pszBasedirectory);
3248	if (!pszFullname)
3249	return VERR_NO_STR_MEMORY;
3250	pExtent->pszFullname = pszFullname;
3251
3252	/* Create file for extent. */
3253	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
3254	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
3255	true /* fCreate */));
3256	if (RT_FAILURE(rc))
3257	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pExtent->pszFullname);
3258
3259	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
3260	{
3261	/* For flat images: Pre allocate file space. */
3262	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage, cbSize,
3263	0 /* fFlags */, NULL, 0, 0);
3264	if (RT_FAILURE(rc))
3265	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
3266	}
3267	else
3268	{
3269	/* For sparse images: Allocate new grain directories/tables. */
3270	/* fPreAlloc should never be false because VMware can't use such images. */
3271	rc = vmdkCreateGrainDirectory(pImage, pExtent,
3272	RT_MAX( pExtent->uDescriptorSector
3273	+ pExtent->cDescriptorSectors,
3274	1),
3275	true /* fPreAlloc */);
3276	if (RT_FAILURE(rc))
3277	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new grain directory in '%s'"), pExtent->pszFullname);
3278	}
3279
3280	/* Insert new extent into descriptor file. */
3281	rc = vmdkDescExtInsert(pImage, &pImage->Descriptor, pExtent->enmAccess,
3282	pExtent->cNominalSectors, pExtent->enmType,
3283	pExtent->pszBasename, pExtent->uSectorOffset);
3284	if (RT_FAILURE(rc))
3285	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not insert the extent list into descriptor in '%s'"), pImage->pszFilename);
3286
3287	pImage->pExtents = pNewExtents;
3288	pImage->cExtents++;
3289
3290	return rc;
3291	}
3292
3293	/**
3294	* Reads and processes the descriptor embedded in sparse images.
3295	*
3296	* @returns VBox status code.
3297	* @param pImage VMDK image instance.
3298	* @param pFile The sparse file handle.
3299	*/
3300	static int vmdkDescriptorReadSparse(PVMDKIMAGE pImage, PVMDKFILE pFile)
3301	{
3302	/* It's a hosted single-extent image. */
3303	int rc = vmdkCreateExtents(pImage, 1);
3304	if (RT_SUCCESS(rc))
3305	{
3306	/* The opened file is passed to the extent. No separate descriptor
3307	* file, so no need to keep anything open for the image. */
3308	PVMDKEXTENT pExtent = &pImage->pExtents[0];
3309	pExtent->pFile = pFile;
3310	pImage->pFile = NULL;
3311	pExtent->pszFullname = RTPathAbsDup(pImage->pszFilename);
3312	if (RT_LIKELY(pExtent->pszFullname))
3313	{
3314	/* As we're dealing with a monolithic image here, there must
3315	* be a descriptor embedded in the image file. */
3316	rc = vmdkReadBinaryMetaExtent(pImage, pExtent, true /* fMagicAlreadyRead */);
3317	if ( RT_SUCCESS(rc)
3318	&& pExtent->uDescriptorSector
3319	&& pExtent->cDescriptorSectors)
3320	{
3321	/* HACK: extend the descriptor if it is unusually small and it fits in
3322	* the unused space after the image header. Allows opening VMDK files
3323	* with extremely small descriptor in read/write mode.
3324	*
3325	* The previous version introduced a possible regression for VMDK stream
3326	* optimized images from VMware which tend to have only a single sector sized
3327	* descriptor. Increasing the descriptor size resulted in adding the various uuid
3328	* entries required to make it work with VBox but for stream optimized images
3329	* the updated binary header wasn't written to the disk creating a mismatch
3330	* between advertised and real descriptor size.
3331	*
3332	* The descriptor size will be increased even if opened readonly now if there
3333	* enough room but the new value will not be written back to the image.
3334	*/
3335	if ( pExtent->cDescriptorSectors < 3
3336	&& (int64_t)pExtent->uSectorGD - pExtent->uDescriptorSector >= 4
3337	&& (!pExtent->uSectorRGD \|\| (int64_t)pExtent->uSectorRGD - pExtent->uDescriptorSector >= 4))
3338	{
3339	uint64_t cDescriptorSectorsOld = pExtent->cDescriptorSectors;
3340
3341	pExtent->cDescriptorSectors = 4;
3342	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
3343	{
3344	/*
3345	* Update the on disk number now to make sure we don't introduce inconsistencies
3346	* in case of stream optimized images from VMware where the descriptor is just
3347	* one sector big (the binary header is not written to disk for complete
3348	* stream optimized images in vmdkFlushImage()).
3349	*/
3350	uint64_t u64DescSizeNew = RT_H2LE_U64(pExtent->cDescriptorSectors);
3351	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pFile->pStorage,
3352	RT_UOFFSETOF(SparseExtentHeader, descriptorSize),
3353	&u64DescSizeNew, sizeof(u64DescSizeNew));
3354	if (RT_FAILURE(rc))
3355	{
3356	LogFlowFunc(("Increasing the descriptor size failed with %Rrc\n", rc));
3357	/* Restore the old size and carry on. */
3358	pExtent->cDescriptorSectors = cDescriptorSectorsOld;
3359	}
3360	}
3361	}
3362	/* Read the descriptor from the extent. */
3363	pExtent->pDescData = (char *)RTMemAllocZ(VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
3364	if (RT_LIKELY(pExtent->pDescData))
3365	{
3366	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
3367	VMDK_SECTOR2BYTE(pExtent->uDescriptorSector),
3368	pExtent->pDescData,
3369	VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
3370	if (RT_SUCCESS(rc))
3371	{
3372	rc = vmdkParseDescriptor(pImage, pExtent->pDescData,
3373	VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
3374	if ( RT_SUCCESS(rc)
3375	&& ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
3376	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_ASYNC_IO)))
3377	{
3378	rc = vmdkReadMetaExtent(pImage, pExtent);
3379	if (RT_SUCCESS(rc))
3380	{
3381	/* Mark the extent as unclean if opened in read-write mode. */
3382	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3383	&& !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
3384	{
3385	pExtent->fUncleanShutdown = true;
3386	pExtent->fMetaDirty = true;
3387	}
3388	}
3389	}
3390	else if (RT_SUCCESS(rc))
3391	rc = VERR_NOT_SUPPORTED;
3392	}
3393	else
3394	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: read error for descriptor in '%s'"), pExtent->pszFullname);
3395	}
3396	else
3397	rc = VERR_NO_MEMORY;
3398	}
3399	else if (RT_SUCCESS(rc))
3400	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: monolithic image without descriptor in '%s'"), pImage->pszFilename);
3401	}
3402	else
3403	rc = VERR_NO_MEMORY;
3404	}
3405
3406	return rc;
3407	}
3408
3409	/**
3410	* Reads the descriptor from a pure text file.
3411	*
3412	* @returns VBox status code.
3413	* @param pImage VMDK image instance.
3414	* @param pFile The descriptor file handle.
3415	*/
3416	static int vmdkDescriptorReadAscii(PVMDKIMAGE pImage, PVMDKFILE pFile)
3417	{
3418	/* Allocate at least 10K, and make sure that there is 5K free space
3419	* in case new entries need to be added to the descriptor. Never
3420	* allocate more than 128K, because that's no valid descriptor file
3421	* and will result in the correct "truncated read" error handling. */
3422	uint64_t cbFileSize;
3423	int rc = vdIfIoIntFileGetSize(pImage->pIfIo, pFile->pStorage, &cbFileSize);
3424	if ( RT_SUCCESS(rc)
3425	&& cbFileSize >= 50)
3426	{
3427	uint64_t cbSize = cbFileSize;
3428	if (cbSize % VMDK_SECTOR2BYTE(10))
3429	cbSize += VMDK_SECTOR2BYTE(20) - cbSize % VMDK_SECTOR2BYTE(10);
3430	else
3431	cbSize += VMDK_SECTOR2BYTE(10);
3432	cbSize = RT_MIN(cbSize, _128K);
3433	pImage->cbDescAlloc = RT_MAX(VMDK_SECTOR2BYTE(20), cbSize);
3434	pImage->pDescData = (char *)RTMemAllocZ(pImage->cbDescAlloc);
3435	if (RT_LIKELY(pImage->pDescData))
3436	{
3437	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pFile->pStorage, 0, pImage->pDescData,
3438	RT_MIN(pImage->cbDescAlloc, cbFileSize));
3439	if (RT_SUCCESS(rc))
3440	{
3441	#if 0 /** @todo Revisit */
3442	cbRead += sizeof(u32Magic);
3443	if (cbRead == pImage->cbDescAlloc)
3444	{
3445	/* Likely the read is truncated. Better fail a bit too early
3446	* (normally the descriptor is much smaller than our buffer). */
3447	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: cannot read descriptor in '%s'"), pImage->pszFilename);
3448	goto out;
3449	}
3450	#endif
3451	rc = vmdkParseDescriptor(pImage, pImage->pDescData,
3452	pImage->cbDescAlloc);
3453	if (RT_SUCCESS(rc))
3454	{
3455	for (unsigned i = 0; i < pImage->cExtents && RT_SUCCESS(rc); i++)
3456	{
3457	PVMDKEXTENT pExtent = &pImage->pExtents[i];
3458	if (pExtent->pszBasename)
3459	{
3460	/* Hack to figure out whether the specified name in the
3461	* extent descriptor is absolute. Doesn't always work, but
3462	* should be good enough for now. */
3463	char *pszFullname;
3464	/** @todo implement proper path absolute check. */
3465	if (pExtent->pszBasename[0] == RTPATH_SLASH)
3466	{
3467	pszFullname = RTStrDup(pExtent->pszBasename);
3468	if (!pszFullname)
3469	{
3470	rc = VERR_NO_MEMORY;
3471	break;
3472	}
3473	}
3474	else
3475	{
3476	char *pszDirname = RTStrDup(pImage->pszFilename);
3477	if (!pszDirname)
3478	{
3479	rc = VERR_NO_MEMORY;
3480	break;
3481	}
3482	RTPathStripFilename(pszDirname);
3483	pszFullname = RTPathJoinA(pszDirname, pExtent->pszBasename);
3484	RTStrFree(pszDirname);
3485	if (!pszFullname)
3486	{
3487	rc = VERR_NO_STR_MEMORY;
3488	break;
3489	}
3490	}
3491	pExtent->pszFullname = pszFullname;
3492	}
3493	else
3494	pExtent->pszFullname = NULL;
3495
3496	unsigned uOpenFlags = pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0);
3497	switch (pExtent->enmType)
3498	{
3499	case VMDKETYPE_HOSTED_SPARSE:
3500	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
3501	VDOpenFlagsToFileOpenFlags(uOpenFlags, false /* fCreate */));
3502	if (RT_FAILURE(rc))
3503	{
3504	/* Do NOT signal an appropriate error here, as the VD
3505	* layer has the choice of retrying the open if it
3506	* failed. */
3507	break;
3508	}
3509	rc = vmdkReadBinaryMetaExtent(pImage, pExtent,
3510	false /* fMagicAlreadyRead */);
3511	if (RT_FAILURE(rc))
3512	break;
3513	rc = vmdkReadMetaExtent(pImage, pExtent);
3514	if (RT_FAILURE(rc))
3515	break;
3516
3517	/* Mark extent as unclean if opened in read-write mode. */
3518	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
3519	{
3520	pExtent->fUncleanShutdown = true;
3521	pExtent->fMetaDirty = true;
3522	}
3523	break;
3524	case VMDKETYPE_VMFS:
3525	case VMDKETYPE_FLAT:
3526	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
3527	VDOpenFlagsToFileOpenFlags(uOpenFlags, false /* fCreate */));
3528	if (RT_FAILURE(rc))
3529	{
3530	/* Do NOT signal an appropriate error here, as the VD
3531	* layer has the choice of retrying the open if it
3532	* failed. */
3533	break;
3534	}
3535	break;
3536	case VMDKETYPE_ZERO:
3537	/* Nothing to do. */
3538	break;
3539	default:
3540	AssertMsgFailed(("unknown vmdk extent type %d\n", pExtent->enmType));
3541	}
3542	}
3543	}
3544	}
3545	else
3546	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: read error for descriptor in '%s'"), pImage->pszFilename);
3547	}
3548	else
3549	rc = VERR_NO_MEMORY;
3550	}
3551	else if (RT_SUCCESS(rc))
3552	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor in '%s' is too short"), pImage->pszFilename);
3553
3554	return rc;
3555	}
3556
3557	/**
3558	* Read and process the descriptor based on the image type.
3559	*
3560	* @returns VBox status code.
3561	* @param pImage VMDK image instance.
3562	* @param pFile VMDK file handle.
3563	*/
3564	static int vmdkDescriptorRead(PVMDKIMAGE pImage, PVMDKFILE pFile)
3565	{
3566	uint32_t u32Magic;
3567
3568	/* Read magic (if present). */
3569	int rc = vdIfIoIntFileReadSync(pImage->pIfIo, pFile->pStorage, 0,
3570	&u32Magic, sizeof(u32Magic));
3571	if (RT_SUCCESS(rc))
3572	{
3573	/* Handle the file according to its magic number. */
3574	if (RT_LE2H_U32(u32Magic) == VMDK_SPARSE_MAGICNUMBER)
3575	rc = vmdkDescriptorReadSparse(pImage, pFile);
3576	else
3577	rc = vmdkDescriptorReadAscii(pImage, pFile);
3578	}
3579	else
3580	{
3581	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading the magic number in '%s'"), pImage->pszFilename);
3582	rc = VERR_VD_VMDK_INVALID_HEADER;
3583	}
3584
3585	return rc;
3586	}
3587
3588	/**
3589	* Internal: Open an image, constructing all necessary data structures.
3590	*/
3591	static int vmdkOpenImage(PVMDKIMAGE pImage, unsigned uOpenFlags)
3592	{
3593	pImage->uOpenFlags = uOpenFlags;
3594	pImage->pIfError = VDIfErrorGet(pImage->pVDIfsDisk);
3595	pImage->pIfIo = VDIfIoIntGet(pImage->pVDIfsImage);
3596	AssertPtrReturn(pImage->pIfIo, VERR_INVALID_PARAMETER);
3597
3598	/*
3599	* Open the image.
3600	* We don't have to check for asynchronous access because
3601	* we only support raw access and the opened file is a description
3602	* file were no data is stored.
3603	*/
3604	PVMDKFILE pFile;
3605	int rc = vmdkFileOpen(pImage, &pFile, NULL, pImage->pszFilename,
3606	VDOpenFlagsToFileOpenFlags(uOpenFlags, false /* fCreate */));
3607	if (RT_SUCCESS(rc))
3608	{
3609	pImage->pFile = pFile;
3610
3611	rc = vmdkDescriptorRead(pImage, pFile);
3612	if (RT_SUCCESS(rc))
3613	{
3614	/* Determine PCHS geometry if not set. */
3615	if (pImage->PCHSGeometry.cCylinders == 0)
3616	{
3617	uint64_t cCylinders = VMDK_BYTE2SECTOR(pImage->cbSize)
3618	/ pImage->PCHSGeometry.cHeads
3619	/ pImage->PCHSGeometry.cSectors;
3620	pImage->PCHSGeometry.cCylinders = (unsigned)RT_MIN(cCylinders, 16383);
3621	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3622	&& !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
3623	{
3624	rc = vmdkDescSetPCHSGeometry(pImage, &pImage->PCHSGeometry);
3625	AssertRC(rc);
3626	}
3627	}
3628
3629	/* Update the image metadata now in case has changed. */
3630	rc = vmdkFlushImage(pImage, NULL);
3631	if (RT_SUCCESS(rc))
3632	{
3633	/* Figure out a few per-image constants from the extents. */
3634	pImage->cbSize = 0;
3635	for (unsigned i = 0; i < pImage->cExtents; i++)
3636	{
3637	PVMDKEXTENT pExtent = &pImage->pExtents[i];
3638	if (pExtent->enmType == VMDKETYPE_HOSTED_SPARSE)
3639	{
3640	/* Here used to be a check whether the nominal size of an extent
3641	* is a multiple of the grain size. The spec says that this is
3642	* always the case, but unfortunately some files out there in the
3643	* wild violate the spec (e.g. ReactOS 0.3.1). */
3644	}
3645	else if ( pExtent->enmType == VMDKETYPE_FLAT
3646	\|\| pExtent->enmType == VMDKETYPE_ZERO)
3647	pImage->uImageFlags \|= VD_IMAGE_FLAGS_FIXED;
3648
3649	pImage->cbSize += VMDK_SECTOR2BYTE(pExtent->cNominalSectors);
3650	}
3651
3652	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
3653	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3654	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
3655	rc = vmdkAllocateGrainTableCache(pImage);
3656	}
3657	}
3658	}
3659	/* else: Do NOT signal an appropriate error here, as the VD layer has the
3660	* choice of retrying the open if it failed. */
3661
3662	if (RT_SUCCESS(rc))
3663	{
3664	PVDREGIONDESC pRegion = &pImage->RegionList.aRegions[0];
3665	pImage->RegionList.fFlags = 0;
3666	pImage->RegionList.cRegions = 1;
3667
3668	pRegion->offRegion = 0; /* Disk start. */
3669	pRegion->cbBlock = 512;
3670	pRegion->enmDataForm = VDREGIONDATAFORM_RAW;
3671	pRegion->enmMetadataForm = VDREGIONMETADATAFORM_NONE;
3672	pRegion->cbData = 512;
3673	pRegion->cbMetadata = 0;
3674	pRegion->cRegionBlocksOrBytes = pImage->cbSize;
3675	}
3676	else
3677	vmdkFreeImage(pImage, false, false /fFlush/); /* Don't try to flush anything if opening failed. */
3678	return rc;
3679	}
3680
3681	/**
3682	* Frees a raw descriptor.
3683	* @internal
3684	*/
3685	static int vmdkRawDescFree(PVDISKRAW pRawDesc)
3686	{
3687	if (!pRawDesc)
3688	return VINF_SUCCESS;
3689
3690	RTStrFree(pRawDesc->pszRawDisk);
3691	pRawDesc->pszRawDisk = NULL;
3692
3693	/* Partitions: */
3694	for (unsigned i = 0; i < pRawDesc->cPartDescs; i++)
3695	{
3696	RTStrFree(pRawDesc->pPartDescs[i].pszRawDevice);
3697	pRawDesc->pPartDescs[i].pszRawDevice = NULL;
3698
3699	RTMemFree(pRawDesc->pPartDescs[i].pvPartitionData);
3700	pRawDesc->pPartDescs[i].pvPartitionData = NULL;
3701	}
3702
3703	RTMemFree(pRawDesc->pPartDescs);
3704	pRawDesc->pPartDescs = NULL;
3705
3706	RTMemFree(pRawDesc);
3707	return VINF_SUCCESS;
3708	}
3709
3710	/**
3711	* Helper that grows the raw partition descriptor table by @a cToAdd entries,
3712	* returning the pointer to the first new entry.
3713	* @internal
3714	*/
3715	static int vmdkRawDescAppendPartDesc(PVMDKIMAGE pImage, PVDISKRAW pRawDesc, uint32_t cToAdd, PVDISKRAWPARTDESC *ppRet)
3716	{
3717	uint32_t const cOld = pRawDesc->cPartDescs;
3718	uint32_t const cNew = cOld + cToAdd;
3719	PVDISKRAWPARTDESC paNew = (PVDISKRAWPARTDESC)RTMemReallocZ(pRawDesc->pPartDescs,
3720	cOld * sizeof(pRawDesc->pPartDescs[0]),
3721	cNew * sizeof(pRawDesc->pPartDescs[0]));
3722	if (paNew)
3723	{
3724	pRawDesc->cPartDescs = cNew;
3725	pRawDesc->pPartDescs = paNew;
3726
3727	*ppRet = &paNew[cOld];
3728	return VINF_SUCCESS;
3729	}
3730	*ppRet = NULL;
3731	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS,
3732	N_("VMDK: Image path: '%s'. Out of memory growing the partition descriptors (%u -> %u)."),
3733	pImage->pszFilename, cOld, cNew);
3734	}
3735
3736	/**
3737	* @callback_method_impl{FNRTSORTCMP}
3738	*/
3739	static DECLCALLBACK(int) vmdkRawDescPartComp(void const pvElement1, void const pvElement2, void *pvUser)
3740	{
3741	RT_NOREF(pvUser);
3742	int64_t const iDelta = ((PVDISKRAWPARTDESC)pvElement1)->offStartInVDisk - ((PVDISKRAWPARTDESC)pvElement2)->offStartInVDisk;
3743	return iDelta < 0 ? -1 : iDelta > 0 ? 1 : 0;
3744	}
3745
3746	/**
3747	* Post processes the partition descriptors.
3748	*
3749	* Sorts them and check that they don't overlap.
3750	*/
3751	static int vmdkRawDescPostProcessPartitions(PVMDKIMAGE pImage, PVDISKRAW pRawDesc, uint64_t cbSize)
3752	{
3753	/*
3754	* Sort data areas in ascending order of start.
3755	*/
3756	RTSortShell(pRawDesc->pPartDescs, pRawDesc->cPartDescs, sizeof(pRawDesc->pPartDescs[0]), vmdkRawDescPartComp, NULL);
3757
3758	/*
3759	* Check that we don't have overlapping descriptors. If we do, that's an
3760	* indication that the drive is corrupt or that the RTDvm code is buggy.
3761	*/
3762	VDISKRAWPARTDESC const *paPartDescs = pRawDesc->pPartDescs;
3763	for (uint32_t i = 0; i < pRawDesc->cPartDescs; i++)
3764	{
3765	uint64_t offLast = paPartDescs[i].offStartInVDisk + paPartDescs[i].cbData;
3766	if (offLast <= paPartDescs[i].offStartInVDisk)
3767	return vdIfError(pImage->pIfError, VERR_FILESYSTEM_CORRUPT /?/, RT_SRC_POS,
3768	N_("VMDK: Image path: '%s'. Bogus partition descriptor #%u (%#RX64 LB %#RX64%s): Wrap around or zero"),
3769	pImage->pszFilename, i, paPartDescs[i].offStartInVDisk, paPartDescs[i].cbData,
3770	paPartDescs[i].pvPartitionData ? " (data)" : "");
3771	offLast -= 1;
3772
3773	if (i + 1 < pRawDesc->cPartDescs && offLast >= paPartDescs[i + 1].offStartInVDisk)
3774	return vdIfError(pImage->pIfError, VERR_FILESYSTEM_CORRUPT /?/, RT_SRC_POS,
3775	N_("VMDK: Image path: '%s'. Partition descriptor #%u (%#RX64 LB %#RX64%s) overlaps with the next (%#RX64 LB %#RX64%s)"),
3776	pImage->pszFilename, i, paPartDescs[i].offStartInVDisk, paPartDescs[i].cbData,
3777	paPartDescs[i].pvPartitionData ? " (data)" : "", paPartDescs[i + 1].offStartInVDisk,
3778	paPartDescs[i + 1].cbData, paPartDescs[i + 1].pvPartitionData ? " (data)" : "");
3779	if (offLast >= cbSize)
3780	return vdIfError(pImage->pIfError, VERR_FILESYSTEM_CORRUPT /?/, RT_SRC_POS,
3781	N_("VMDK: Image path: '%s'. Partition descriptor #%u (%#RX64 LB %#RX64%s) goes beyond the end of the drive (%#RX64)"),
3782	pImage->pszFilename, i, paPartDescs[i].offStartInVDisk, paPartDescs[i].cbData,
3783	paPartDescs[i].pvPartitionData ? " (data)" : "", cbSize);
3784	}
3785
3786	return VINF_SUCCESS;
3787	}
3788
3789
3790	#ifdef RT_OS_LINUX
3791	/**
3792	* Searches the dir specified in @a pszBlockDevDir for subdirectories with a
3793	* 'dev' file matching @a uDevToLocate.
3794	*
3795	* This is used both
3796	*
3797	* @returns IPRT status code, errors have been reported properly.
3798	* @param pImage For error reporting.
3799	* @param pszBlockDevDir Input: Path to the directory search under.
3800	* Output: Path to the directory containing information
3801	* for @a uDevToLocate.
3802	* @param cbBlockDevDir The size of the buffer @a pszBlockDevDir points to.
3803	* @param uDevToLocate The device number of the block device info dir to
3804	* locate.
3805	* @param pszDevToLocate For error reporting.
3806	*/
3807	static int vmdkFindSysBlockDevPath(PVMDKIMAGE pImage, char *pszBlockDevDir, size_t cbBlockDevDir,
3808	dev_t uDevToLocate, const char *pszDevToLocate)
3809	{
3810	size_t const cchDir = RTPathEnsureTrailingSeparator(pszBlockDevDir, cbBlockDevDir);
3811	AssertReturn(cchDir > 0, VERR_BUFFER_OVERFLOW);
3812
3813	RTDIR hDir = NIL_RTDIR;
3814	int rc = RTDirOpen(&hDir, pszBlockDevDir);
3815	if (RT_SUCCESS(rc))
3816	{
3817	for (;;)
3818	{
3819	RTDIRENTRY Entry;
3820	rc = RTDirRead(hDir, &Entry, NULL);
3821	if (RT_SUCCESS(rc))
3822	{
3823	/* We're interested in directories and symlinks. */
3824	if ( Entry.enmType == RTDIRENTRYTYPE_DIRECTORY
3825	\|\| Entry.enmType == RTDIRENTRYTYPE_SYMLINK
3826	\|\| Entry.enmType == RTDIRENTRYTYPE_UNKNOWN)
3827	{
3828	rc = RTStrCopy(&pszBlockDevDir[cchDir], cbBlockDevDir - cchDir, Entry.szName);
3829	AssertContinue(RT_SUCCESS(rc)); /* should not happen! */
3830
3831	dev_t uThisDevNo = ~uDevToLocate;
3832	rc = RTLinuxSysFsReadDevNumFile(&uThisDevNo, "%s/dev", pszBlockDevDir);
3833	if (RT_SUCCESS(rc) && uThisDevNo == uDevToLocate)
3834	break;
3835	}
3836	}
3837	else
3838	{
3839	pszBlockDevDir[cchDir] = '\0';
3840	if (rc == VERR_NO_MORE_FILES)
3841	rc = vdIfError(pImage->pIfError, VERR_NOT_FOUND, RT_SRC_POS,
3842	N_("VMDK: Image path: '%s'. Failed to locate device corresponding to '%s' under '%s'"),
3843	pImage->pszFilename, pszDevToLocate, pszBlockDevDir);
3844	else
3845	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
3846	N_("VMDK: Image path: '%s'. RTDirRead failed enumerating '%s': %Rrc"),
3847	pImage->pszFilename, pszBlockDevDir, rc);
3848	break;
3849	}
3850	}
3851	RTDirClose(hDir);
3852	}
3853	else
3854	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
3855	N_("VMDK: Image path: '%s'. Failed to open dir '%s' for listing: %Rrc"),
3856	pImage->pszFilename, pszBlockDevDir, rc);
3857	return rc;
3858	}
3859	#endif /* RT_OS_LINUX */
3860
3861	#ifdef RT_OS_FREEBSD
3862
3863
3864	/**
3865	* Reads the config data from the provider and returns offset and size
3866	*
3867	* @return IPRT status code
3868	* @param pProvider GEOM provider representing partition
3869	* @param pcbOffset Placeholder for the offset of the partition
3870	* @param pcbSize Placeholder for the size of the partition
3871	*/
3872	static int vmdkReadPartitionsParamsFromProvider(gprovider pProvider, uint64_t pcbOffset, uint64_t *pcbSize)
3873	{
3874	gconfig *pConfEntry;
3875	int rc = VERR_NOT_FOUND;
3876
3877	/*
3878	* Required parameters are located in the list containing key/value pairs.
3879	* Both key and value are in text form. Manuals tells nothing about the fact
3880	* that the both parameters should be present in the list. Thus, there are
3881	* cases when only one parameter is presented. To handle such cases we treat
3882	* absent params as zero allowing the caller decide the case is either correct
3883	* or an error.
3884	*/
3885	uint64_t cbOffset = 0;
3886	uint64_t cbSize = 0;
3887	LIST_FOREACH(pConfEntry, &pProvider->lg_config, lg_config)
3888	{
3889	if (RTStrCmp(pConfEntry->lg_name, "offset") == 0)
3890	{
3891	cbOffset = RTStrToUInt64(pConfEntry->lg_val);
3892	rc = VINF_SUCCESS;
3893	}
3894	else if (RTStrCmp(pConfEntry->lg_name, "length") == 0)
3895	{
3896	cbSize = RTStrToUInt64(pConfEntry->lg_val);
3897	rc = VINF_SUCCESS;
3898	}
3899	}
3900	if (RT_SUCCESS(rc))
3901	{
3902	*pcbOffset = cbOffset;
3903	*pcbSize = cbSize;
3904	}
3905	return rc;
3906	}
3907
3908
3909	/**
3910	* Searches the partition specified by name and calculates its size and absolute offset.
3911	*
3912	* @return IPRT status code.
3913	* @param pParentClass Class containing pParentGeom
3914	* @param pszParentGeomName Name of the parent geom where we are looking for provider
3915	* @param pszProviderName Name of the provider we are looking for
3916	* @param pcbAbsoluteOffset Placeholder for the absolute offset of the partition, i.e. offset from the beginning of the disk
3917	* @param psbSize Placeholder for the size of the partition.
3918	*/
3919	static int vmdkFindPartitionParamsByName(gclass pParentClass, const char pszParentGeomName, const char *pszProviderName,
3920	uint64_t pcbAbsoluteOffset, uint64_t pcbSize)
3921	{
3922	AssertReturn(pParentClass, VERR_INVALID_PARAMETER);
3923	AssertReturn(pszParentGeomName, VERR_INVALID_PARAMETER);
3924	AssertReturn(pszProviderName, VERR_INVALID_PARAMETER);
3925	AssertReturn(pcbAbsoluteOffset, VERR_INVALID_PARAMETER);
3926	AssertReturn(pcbSize, VERR_INVALID_PARAMETER);
3927
3928	ggeom *pParentGeom;
3929	int rc = VERR_NOT_FOUND;
3930	LIST_FOREACH(pParentGeom, &pParentClass->lg_geom, lg_geom)
3931	{
3932	if (RTStrCmp(pParentGeom->lg_name, pszParentGeomName) == 0)
3933	{
3934	rc = VINF_SUCCESS;
3935	break;
3936	}
3937	}
3938	if (RT_FAILURE(rc))
3939	return rc;
3940
3941	gprovider *pProvider;
3942	/*
3943	* First, go over providers without handling EBR or BSDLabel
3944	* partitions for case when looking provider is child
3945	* of the givng geom, to reduce searching time
3946	*/
3947	LIST_FOREACH(pProvider, &pParentGeom->lg_provider, lg_provider)
3948	{
3949	if (RTStrCmp(pProvider->lg_name, pszProviderName) == 0)
3950	return vmdkReadPartitionsParamsFromProvider(pProvider, pcbAbsoluteOffset, pcbSize);
3951	}
3952
3953	/*
3954	* No provider found. Go over the parent geom again
3955	* and make recursions if geom represents EBR or BSDLabel.
3956	* In this case given parent geom contains only EBR or BSDLabel
3957	* partition itself and their own partitions are in the separate
3958	* geoms. Also, partition offsets are relative to geom, so
3959	* we have to add offset from child provider with parent geoms
3960	* provider
3961	*/
3962
3963	LIST_FOREACH(pProvider, &pParentGeom->lg_provider, lg_provider)
3964	{
3965	uint64_t cbOffset = 0;
3966	uint64_t cbSize = 0;
3967	rc = vmdkReadPartitionsParamsFromProvider(pProvider, &cbOffset, &cbSize);
3968	if (RT_FAILURE(rc))
3969	return rc;
3970
3971	uint64_t cbProviderOffset = 0;
3972	uint64_t cbProviderSize = 0;
3973	rc = vmdkFindPartitionParamsByName(pParentClass, pProvider->lg_name, pszProviderName, &cbProviderOffset, &cbProviderSize);
3974	if (RT_SUCCESS(rc))
3975	{
3976	*pcbAbsoluteOffset = cbOffset + cbProviderOffset;
3977	*pcbSize = cbProviderSize;
3978	return rc;
3979	}
3980	}
3981
3982	return VERR_NOT_FOUND;
3983	}
3984	#endif
3985
3986
3987	/**
3988	* Attempts to verify the raw partition path.
3989	*
3990	* We don't want to trust RTDvm and the partition device node morphing blindly.
3991	*/
3992	static int vmdkRawDescVerifyPartitionPath(PVMDKIMAGE pImage, PVDISKRAWPARTDESC pPartDesc, uint32_t idxPartition,
3993	const char *pszRawDrive, RTFILE hRawDrive, uint32_t cbSector, RTDVMVOLUME hVol)
3994	{
3995	RT_NOREF(pImage, pPartDesc, idxPartition, pszRawDrive, hRawDrive, cbSector, hVol);
3996
3997	/*
3998	* Try open the raw partition device.
3999	*/
4000	RTFILE hRawPart = NIL_RTFILE;
4001	int rc = RTFileOpen(&hRawPart, pPartDesc->pszRawDevice, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE);
4002	if (RT_FAILURE(rc))
4003	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4004	N_("VMDK: Image path: '%s'. Failed to open partition #%u on '%s' via '%s' (%Rrc)"),
4005	pImage->pszFilename, idxPartition, pszRawDrive, pPartDesc->pszRawDevice, rc);
4006
4007	/*
4008	* Compare the partition UUID if we can get it.
4009	*/
4010	#ifdef RT_OS_WINDOWS
4011	DWORD cbReturned;
4012
4013	/* 1. Get the device numbers for both handles, they should have the same disk. */
4014	STORAGE_DEVICE_NUMBER DevNum1;
4015	RT_ZERO(DevNum1);
4016	if (!DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_STORAGE_GET_DEVICE_NUMBER,
4017	NULL /pvInBuffer/, 0 /cbInBuffer/, &DevNum1, sizeof(DevNum1), &cbReturned, NULL /pOverlapped/))
4018	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
4019	N_("VMDK: Image path: '%s'. IOCTL_STORAGE_GET_DEVICE_NUMBER failed on '%s': %u"),
4020	pImage->pszFilename, pszRawDrive, GetLastError());
4021
4022	STORAGE_DEVICE_NUMBER DevNum2;
4023	RT_ZERO(DevNum2);
4024	if (!DeviceIoControl((HANDLE)RTFileToNative(hRawPart), IOCTL_STORAGE_GET_DEVICE_NUMBER,
4025	NULL /pvInBuffer/, 0 /cbInBuffer/, &DevNum2, sizeof(DevNum2), &cbReturned, NULL /pOverlapped/))
4026	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
4027	N_("VMDK: Image path: '%s'. IOCTL_STORAGE_GET_DEVICE_NUMBER failed on '%s': %u"),
4028	pImage->pszFilename, pPartDesc->pszRawDevice, GetLastError());
4029	if ( RT_SUCCESS(rc)
4030	&& ( DevNum1.DeviceNumber != DevNum2.DeviceNumber
4031	\|\| DevNum1.DeviceType != DevNum2.DeviceType))
4032	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4033	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s' (%#x != %#x \|\| %#x != %#x)"),
4034	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4035	DevNum1.DeviceNumber, DevNum2.DeviceNumber, DevNum1.DeviceType, DevNum2.DeviceType);
4036	if (RT_SUCCESS(rc))
4037	{
4038	/* Get the partitions from the raw drive and match up with the volume info
4039	from RTDvm. The partition number is found in DevNum2. */
4040	DWORD cbNeeded = 0;
4041	if ( DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_DISK_GET_DRIVE_LAYOUT_EX,
4042	NULL /pvInBuffer/, 0 /cbInBuffer/, NULL, 0, &cbNeeded, NULL /pOverlapped/)
4043	\|\| cbNeeded < RT_UOFFSETOF_DYN(DRIVE_LAYOUT_INFORMATION_EX, PartitionEntry[1]))
4044	cbNeeded = RT_UOFFSETOF_DYN(DRIVE_LAYOUT_INFORMATION_EX, PartitionEntry[64]);
4045	cbNeeded += sizeof(PARTITION_INFORMATION_EX) * 2; /* just in case */
4046	DRIVE_LAYOUT_INFORMATION_EX pLayout = (DRIVE_LAYOUT_INFORMATION_EX )RTMemTmpAllocZ(cbNeeded);
4047	if (pLayout)
4048	{
4049	cbReturned = 0;
4050	if (DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_DISK_GET_DRIVE_LAYOUT_EX,
4051	NULL /pvInBuffer/, 0 /cbInBuffer/, pLayout, cbNeeded, &cbReturned, NULL /pOverlapped/))
4052	{
4053	/* Find the entry with the given partition number (it's not an index, array contains empty MBR entries ++). */
4054	unsigned iEntry = 0;
4055	while ( iEntry < pLayout->PartitionCount
4056	&& pLayout->PartitionEntry[iEntry].PartitionNumber != DevNum2.PartitionNumber)
4057	iEntry++;
4058	if (iEntry < pLayout->PartitionCount)
4059	{
4060	/* Compare the basics */
4061	PARTITION_INFORMATION_EX const * const pLayoutEntry = &pLayout->PartitionEntry[iEntry];
4062	if (pLayoutEntry->StartingOffset.QuadPart != (int64_t)pPartDesc->offStartInVDisk)
4063	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4064	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': StartingOffset %RU64, expected %RU64"),
4065	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4066	pLayoutEntry->StartingOffset.QuadPart, pPartDesc->offStartInVDisk);
4067	else if (pLayoutEntry->PartitionLength.QuadPart != (int64_t)pPartDesc->cbData)
4068	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4069	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': PartitionLength %RU64, expected %RU64"),
4070	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4071	pLayoutEntry->PartitionLength.QuadPart, pPartDesc->cbData);
4072	/** @todo We could compare the MBR type, GPT type and ID. */
4073	RT_NOREF(hVol);
4074	}
4075	else
4076	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4077	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': PartitionCount (%#x vs %#x)"),
4078	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4079	DevNum2.PartitionNumber, pLayout->PartitionCount);
4080	# ifndef LOG_ENABLED
4081	if (RT_FAILURE(rc))
4082	# endif
4083	{
4084	LogRel(("VMDK: Windows reports %u partitions for '%s':\n", pLayout->PartitionCount, pszRawDrive));
4085	PARTITION_INFORMATION_EX const *pEntry = &pLayout->PartitionEntry[0];
4086	for (DWORD i = 0; i < pLayout->PartitionCount; i++, pEntry++)
4087	{
4088	LogRel(("VMDK: #%u/%u: %016RU64 LB %016RU64 style=%d rewrite=%d",
4089	i, pEntry->PartitionNumber, pEntry->StartingOffset.QuadPart, pEntry->PartitionLength.QuadPart,
4090	pEntry->PartitionStyle, pEntry->RewritePartition));
4091	if (pEntry->PartitionStyle == PARTITION_STYLE_MBR)
4092	LogRel((" type=%#x boot=%d rec=%d hidden=%u\n", pEntry->Mbr.PartitionType, pEntry->Mbr.BootIndicator,
4093	pEntry->Mbr.RecognizedPartition, pEntry->Mbr.HiddenSectors));
4094	else if (pEntry->PartitionStyle == PARTITION_STYLE_GPT)
4095	LogRel((" type=%RTuuid id=%RTuuid aatrib=%RX64 name=%.36ls\n", &pEntry->Gpt.PartitionType,
4096	&pEntry->Gpt.PartitionId, pEntry->Gpt.Attributes, &pEntry->Gpt.Name[0]));
4097	else
4098	LogRel(("\n"));
4099	}
4100	LogRel(("VMDK: Looked for partition #%u (%u, '%s') at %RU64 LB %RU64\n", DevNum2.PartitionNumber,
4101	idxPartition, pPartDesc->pszRawDevice, pPartDesc->offStartInVDisk, pPartDesc->cbData));
4102	}
4103	}
4104	else
4105	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
4106	N_("VMDK: Image path: '%s'. IOCTL_DISK_GET_DRIVE_LAYOUT_EX failed on '%s': %u (cb %u, cbRet %u)"),
4107	pImage->pszFilename, pPartDesc->pszRawDevice, GetLastError(), cbNeeded, cbReturned);
4108	RTMemTmpFree(pLayout);
4109	}
4110	else
4111	rc = VERR_NO_TMP_MEMORY;
4112	}
4113
4114	#elif defined(RT_OS_LINUX)
4115	RT_NOREF(hVol);
4116
4117	/* Stat the two devices first to get their device numbers. (We probably
4118	could make some assumptions here about the major & minor number assignments
4119	for legacy nodes, but it doesn't hold up for nvme, so we'll skip that.) */
4120	struct stat StDrive, StPart;
4121	if (fstat((int)RTFileToNative(hRawDrive), &StDrive) != 0)
4122	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4123	N_("VMDK: Image path: '%s'. fstat failed on '%s': %d"), pImage->pszFilename, pszRawDrive, errno);
4124	else if (fstat((int)RTFileToNative(hRawPart), &StPart) != 0)
4125	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4126	N_("VMDK: Image path: '%s'. fstat failed on '%s': %d"), pImage->pszFilename, pPartDesc->pszRawDevice, errno);
4127	else
4128	{
4129	/* Scan the directories immediately under /sys/block/ for one with a
4130	'dev' file matching the drive's device number: */
4131	char szSysPath[RTPATH_MAX];
4132	rc = RTLinuxConstructPath(szSysPath, sizeof(szSysPath), "block/");
4133	AssertRCReturn(rc, rc); /* this shall not fail */
4134	if (RTDirExists(szSysPath))
4135	{
4136	rc = vmdkFindSysBlockDevPath(pImage, szSysPath, sizeof(szSysPath), StDrive.st_rdev, pszRawDrive);
4137
4138	/* Now, scan the directories under that again for a partition device
4139	matching the hRawPart device's number: */
4140	if (RT_SUCCESS(rc))
4141	rc = vmdkFindSysBlockDevPath(pImage, szSysPath, sizeof(szSysPath), StPart.st_rdev, pPartDesc->pszRawDevice);
4142
4143	/* Having found the /sys/block/device/partition/ path, we can finally
4144	read the partition attributes and compare with hVol. */
4145	if (RT_SUCCESS(rc))
4146	{
4147	/* partition number: */
4148	int64_t iLnxPartition = 0;
4149	rc = RTLinuxSysFsReadIntFile(10, &iLnxPartition, "%s/partition", szSysPath);
4150	if (RT_SUCCESS(rc) && iLnxPartition != idxPartition)
4151	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4152	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Partition number %RI64, expected %RU32"),
4153	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, iLnxPartition, idxPartition);
4154	/* else: ignore failure? */
4155
4156	/* start offset: */
4157	uint32_t const cbLnxSector = 512; /* It's hardcoded in the Linux kernel */
4158	if (RT_SUCCESS(rc))
4159	{
4160	int64_t offLnxStart = -1;
4161	rc = RTLinuxSysFsReadIntFile(10, &offLnxStart, "%s/start", szSysPath);
4162	offLnxStart *= cbLnxSector;
4163	if (RT_SUCCESS(rc) && offLnxStart != (int64_t)pPartDesc->offStartInVDisk)
4164	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4165	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RI64, expected %RU64"),
4166	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, offLnxStart, pPartDesc->offStartInVDisk);
4167	/* else: ignore failure? */
4168	}
4169
4170	/* the size: */
4171	if (RT_SUCCESS(rc))
4172	{
4173	int64_t cbLnxData = -1;
4174	rc = RTLinuxSysFsReadIntFile(10, &cbLnxData, "%s/size", szSysPath);
4175	cbLnxData *= cbLnxSector;
4176	if (RT_SUCCESS(rc) && cbLnxData != (int64_t)pPartDesc->cbData)
4177	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4178	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RI64, expected %RU64"),
4179	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbLnxData, pPartDesc->cbData);
4180	/* else: ignore failure? */
4181	}
4182	}
4183	}
4184	/* else: We've got nothing to work on, so only do content comparison. */
4185	}
4186
4187	#elif defined(RT_OS_FREEBSD)
4188	char szDriveDevName[256];
4189	char* pszDevName = fdevname_r(RTFileToNative(hRawDrive), szDriveDevName, 256);
4190	if (pszDevName == NULL)
4191	rc = vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4192	N_("VMDK: Image path: '%s'. '%s' is not a drive path"), pImage->pszFilename, pszRawDrive);
4193	char szPartDevName[256];
4194	if (RT_SUCCESS(rc))
4195	{
4196	pszDevName = fdevname_r(RTFileToNative(hRawPart), szPartDevName, 256);
4197	if (pszDevName == NULL)
4198	rc = vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4199	N_("VMDK: Image path: '%s'. '%s' is not a partition path"), pImage->pszFilename, pPartDesc->pszRawDevice);
4200	}
4201	if (RT_SUCCESS(rc))
4202	{
4203	gmesh geomMesh;
4204	int err = geom_gettree(&geomMesh);
4205	if (err == 0)
4206	{
4207	/* Find root class containg partitions info */
4208	gclass* pPartClass;
4209	LIST_FOREACH(pPartClass, &geomMesh.lg_class, lg_class)
4210	{
4211	if (RTStrCmp(pPartClass->lg_name, "PART") == 0)
4212	break;
4213	}
4214	if (pPartClass == NULL \|\| RTStrCmp(pPartClass->lg_name, "PART") != 0)
4215	rc = vdIfError(pImage->pIfError, VERR_GENERAL_FAILURE, RT_SRC_POS,
4216	N_("VMDK: Image path: '%s'. 'PART' class not found in the GEOM tree"), pImage->pszFilename);
4217
4218
4219	if (RT_SUCCESS(rc))
4220	{
4221	/* Find provider representing partition device */
4222	uint64_t cbOffset;
4223	uint64_t cbSize;
4224	rc = vmdkFindPartitionParamsByName(pPartClass, szDriveDevName, szPartDevName, &cbOffset, &cbSize);
4225	if (RT_SUCCESS(rc))
4226	{
4227	if (cbOffset != pPartDesc->offStartInVDisk)
4228	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4229	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RU64, expected %RU64"),
4230	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbOffset, pPartDesc->offStartInVDisk);
4231	if (cbSize != pPartDesc->cbData)
4232	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4233	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RU64, expected %RU64"),
4234	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbSize, pPartDesc->cbData);
4235	}
4236	else
4237	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4238	N_("VMDK: Image path: '%s'. Error getting geom provider for the partition '%s' of the drive '%s' in the GEOM tree: %Rrc"),
4239	pImage->pszFilename, pPartDesc->pszRawDevice, pszRawDrive, rc);
4240	}
4241
4242	geom_deletetree(&geomMesh);
4243	}
4244	else
4245	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(err), RT_SRC_POS,
4246	N_("VMDK: Image path: '%s'. geom_gettree failed: %d"), pImage->pszFilename, err);
4247	}
4248
4249	#elif defined(RT_OS_SOLARIS)
4250	RT_NOREF(hVol);
4251
4252	dk_cinfo dkiDriveInfo;
4253	dk_cinfo dkiPartInfo;
4254	if (ioctl(RTFileToNative(hRawDrive), DKIOCINFO, (caddr_t)&dkiDriveInfo) == -1)
4255	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4256	N_("VMDK: Image path: '%s'. DKIOCINFO failed on '%s': %d"), pImage->pszFilename, pszRawDrive, errno);
4257	else if (ioctl(RTFileToNative(hRawPart), DKIOCINFO, (caddr_t)&dkiPartInfo) == -1)
4258	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4259	N_("VMDK: Image path: '%s'. DKIOCINFO failed on '%s': %d"), pImage->pszFilename, pszRawDrive, errno);
4260	else if ( dkiDriveInfo.dki_ctype != dkiPartInfo.dki_ctype
4261	\|\| dkiDriveInfo.dki_cnum != dkiPartInfo.dki_cnum
4262	\|\| dkiDriveInfo.dki_addr != dkiPartInfo.dki_addr
4263	\|\| dkiDriveInfo.dki_unit != dkiPartInfo.dki_unit
4264	\|\| dkiDriveInfo.dki_slave != dkiPartInfo.dki_slave)
4265	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4266	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s' (%#x != %#x \|\| %#x != %#x \|\| %#x != %#x \|\| %#x != %#x \|\| %#x != %#x)"),
4267	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4268	dkiDriveInfo.dki_ctype, dkiPartInfo.dki_ctype, dkiDriveInfo.dki_cnum, dkiPartInfo.dki_cnum,
4269	dkiDriveInfo.dki_addr, dkiPartInfo.dki_addr, dkiDriveInfo.dki_unit, dkiPartInfo.dki_unit,
4270	dkiDriveInfo.dki_slave, dkiPartInfo.dki_slave);
4271	else
4272	{
4273	uint64_t cbOffset = 0;
4274	uint64_t cbSize = 0;
4275	dk_gpt *pEfi = NULL;
4276	int idxEfiPart = efi_alloc_and_read(RTFileToNative(hRawPart), &pEfi);
4277	if (idxEfiPart >= 0)
4278	{
4279	if ((uint32_t)dkiPartInfo.dki_partition + 1 == idxPartition)
4280	{
4281	cbOffset = pEfi->efi_parts[idxEfiPart].p_start * pEfi->efi_lbasize;
4282	cbSize = pEfi->efi_parts[idxEfiPart].p_size * pEfi->efi_lbasize;
4283	}
4284	else
4285	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4286	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s' (%#x != %#x)"),
4287	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4288	idxPartition, (uint32_t)dkiPartInfo.dki_partition + 1);
4289	efi_free(pEfi);
4290	}
4291	else
4292	{
4293	/*
4294	* Manual says the efi_alloc_and_read returns VT_EINVAL if no EFI partition table found.
4295	* Actually, the function returns any error, e.g. VT_ERROR. Thus, we are not sure, is it
4296	* real error or just no EFI table found. Therefore, let's try to obtain partition info
4297	* using another way. If there is an error, it returns errno which will be handled below.
4298	*/
4299
4300	uint32_t numPartition = (uint32_t)dkiPartInfo.dki_partition;
4301	if (numPartition > NDKMAP)
4302	numPartition -= NDKMAP;
4303	if (numPartition != idxPartition)
4304	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4305	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s' (%#x != %#x)"),
4306	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4307	idxPartition, numPartition);
4308	else
4309	{
4310	dk_minfo_ext mediaInfo;
4311	if (ioctl(RTFileToNative(hRawPart), DKIOCGMEDIAINFOEXT, (caddr_t)&mediaInfo) == -1)
4312	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4313	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s'. Can not obtain partition info: %d"),
4314	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4315	else
4316	{
4317	extpart_info extPartInfo;
4318	if (ioctl(RTFileToNative(hRawPart), DKIOCEXTPARTINFO, (caddr_t)&extPartInfo) != -1)
4319	{
4320	cbOffset = (uint64_t)extPartInfo.p_start * mediaInfo.dki_lbsize;
4321	cbSize = (uint64_t)extPartInfo.p_length * mediaInfo.dki_lbsize;
4322	}
4323	else
4324	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4325	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s'. Can not obtain partition info: %d"),
4326	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4327	}
4328	}
4329	}
4330	if (RT_SUCCESS(rc) && cbOffset != pPartDesc->offStartInVDisk)
4331	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4332	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RI64, expected %RU64"),
4333	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbOffset, pPartDesc->offStartInVDisk);
4334
4335	if (RT_SUCCESS(rc) && cbSize != pPartDesc->cbData)
4336	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4337	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RI64, expected %RU64"),
4338	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbSize, pPartDesc->cbData);
4339	}
4340
4341	#elif defined(RT_OS_DARWIN)
4342	/* Stat the drive get its device number. */
4343	struct stat StDrive;
4344	if (fstat((int)RTFileToNative(hRawDrive), &StDrive) != 0)
4345	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4346	N_("VMDK: Image path: '%s'. fstat failed on '%s' (errno=%d)"), pImage->pszFilename, pszRawDrive, errno);
4347	else
4348	{
4349	if (ioctl(RTFileToNative(hRawPart), DKIOCLOCKPHYSICALEXTENTS, NULL) == -1)
4350	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4351	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to lock the partition (errno=%d)"),
4352	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4353	else
4354	{
4355	uint32_t cbBlockSize = 0;
4356	uint64_t cbOffset = 0;
4357	uint64_t cbSize = 0;
4358	if (ioctl(RTFileToNative(hRawPart), DKIOCGETBLOCKSIZE, (caddr_t)&cbBlockSize) == -1)
4359	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4360	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain the sector size of the partition (errno=%d)"),
4361	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4362	else if (ioctl(RTFileToNative(hRawPart), DKIOCGETBASE, (caddr_t)&cbOffset) == -1)
4363	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4364	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain the start offset of the partition (errno=%d)"),
4365	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4366	else if (ioctl(RTFileToNative(hRawPart), DKIOCGETBLOCKCOUNT, (caddr_t)&cbSize) == -1)
4367	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4368	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain the size of the partition (errno=%d)"),
4369	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4370	else
4371	{
4372	cbSize *= (uint64_t)cbBlockSize;
4373	dk_physical_extent_t dkPartExtent = {0};
4374	dkPartExtent.offset = 0;
4375	dkPartExtent.length = cbSize;
4376	if (ioctl(RTFileToNative(hRawPart), DKIOCGETPHYSICALEXTENT, (caddr_t)&dkPartExtent) == -1)
4377	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4378	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain partition info (errno=%d)"),
4379	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4380	else
4381	{
4382	if (dkPartExtent.dev != StDrive.st_rdev)
4383	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4384	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Drive does not contain the partition"),
4385	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive);
4386	else if (cbOffset != pPartDesc->offStartInVDisk)
4387	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4388	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RU64, expected %RU64"),
4389	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbOffset, pPartDesc->offStartInVDisk);
4390	else if (cbSize != pPartDesc->cbData)
4391	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4392	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RU64, expected %RU64"),
4393	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbSize, pPartDesc->cbData);
4394	}
4395	}
4396
4397	if (ioctl(RTFileToNative(hRawPart), DKIOCUNLOCKPHYSICALEXTENTS, NULL) == -1)
4398	{
4399	int rc2 = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4400	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to unlock the partition (errno=%d)"),
4401	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4402	if (RT_SUCCESS(rc))
4403	rc = rc2;
4404	}
4405	}
4406	}
4407
4408	#else
4409	RT_NOREF(hVol); /* PORTME */
4410	rc = VERR_NOT_SUPPORTED;
4411	#endif
4412	if (RT_SUCCESS(rc))
4413	{
4414	/*
4415	* Compare the first 32 sectors of the partition.
4416	*
4417	* This might not be conclusive, but for partitions formatted with the more
4418	* common file systems it should be as they have a superblock copy at or near
4419	* the start of the partition (fat, fat32, ntfs, and ext4 does at least).
4420	*/
4421	size_t const cbToCompare = (size_t)RT_MIN(pPartDesc->cbData / cbSector, 32) * cbSector;
4422	uint8_t pbSector1 = (uint8_t )RTMemTmpAlloc(cbToCompare * 2);
4423	if (pbSector1 != NULL)
4424	{
4425	uint8_t *pbSector2 = pbSector1 + cbToCompare;
4426
4427	/* Do the comparing, we repeat if it fails and the data might be volatile. */
4428	uint64_t uPrevCrc1 = 0;
4429	uint64_t uPrevCrc2 = 0;
4430	uint32_t cStable = 0;
4431	for (unsigned iTry = 0; iTry < 256; iTry++)
4432	{
4433	rc = RTFileReadAt(hRawDrive, pPartDesc->offStartInVDisk, pbSector1, cbToCompare, NULL);
4434	if (RT_SUCCESS(rc))
4435	{
4436	rc = RTFileReadAt(hRawPart, pPartDesc->offStartInDevice, pbSector2, cbToCompare, NULL);
4437	if (RT_SUCCESS(rc))
4438	{
4439	if (memcmp(pbSector1, pbSector2, cbToCompare) != 0)
4440	{
4441	rc = VERR_MISMATCH;
4442
4443	/* Do data stability checks before repeating: */
4444	uint64_t const uCrc1 = RTCrc64(pbSector1, cbToCompare);
4445	uint64_t const uCrc2 = RTCrc64(pbSector2, cbToCompare);
4446	if ( uPrevCrc1 != uCrc1
4447	\|\| uPrevCrc2 != uCrc2)
4448	cStable = 0;
4449	else if (++cStable > 4)
4450	break;
4451	uPrevCrc1 = uCrc1;
4452	uPrevCrc2 = uCrc2;
4453	continue;
4454	}
4455	rc = VINF_SUCCESS;
4456	}
4457	else
4458	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4459	N_("VMDK: Image path: '%s'. Error reading %zu bytes from '%s' at offset %RU64 (%Rrc)"),
4460	pImage->pszFilename, cbToCompare, pPartDesc->pszRawDevice, pPartDesc->offStartInDevice, rc);
4461	}
4462	else
4463	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4464	N_("VMDK: Image path: '%s'. Error reading %zu bytes from '%s' at offset %RU64 (%Rrc)"),
4465	pImage->pszFilename, cbToCompare, pszRawDrive, pPartDesc->offStartInVDisk, rc);
4466	break;
4467	}
4468	if (rc == VERR_MISMATCH)
4469	{
4470	/* Find the first mismatching bytes: */
4471	size_t offMissmatch = 0;
4472	while (offMissmatch < cbToCompare && pbSector1[offMissmatch] == pbSector2[offMissmatch])
4473	offMissmatch++;
4474	int cbSample = (int)RT_MIN(cbToCompare - offMissmatch, 16);
4475
4476	if (cStable > 0)
4477	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4478	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s' (cStable=%d @%#zx: %.Rhxs vs %.Rhxs)"),
4479	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cStable,
4480	offMissmatch, cbSample, &pbSector1[offMissmatch], cbSample, &pbSector2[offMissmatch]);
4481	else
4482	{
4483	LogRel(("VMDK: Image path: '%s'. Partition #%u path ('%s') verification undecided on '%s' because of unstable data! (@%#zx: %.Rhxs vs %.Rhxs)\n",
4484	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4485	offMissmatch, cbSample, &pbSector1[offMissmatch], cbSample, &pbSector2[offMissmatch]));
4486	rc = -rc;
4487	}
4488	}
4489
4490	RTMemTmpFree(pbSector1);
4491	}
4492	else
4493	rc = vdIfError(pImage->pIfError, VERR_NO_TMP_MEMORY, RT_SRC_POS,
4494	N_("VMDK: Image path: '%s'. Failed to allocate %zu bytes for a temporary read buffer\n"),
4495	pImage->pszFilename, cbToCompare * 2);
4496	}
4497	RTFileClose(hRawPart);
4498	return rc;
4499	}
4500
4501	#ifdef RT_OS_WINDOWS
4502	/**
4503	* Construct the device name for the given partition number.
4504	*/
4505	static int vmdkRawDescWinMakePartitionName(PVMDKIMAGE pImage, const char *pszRawDrive, RTFILE hRawDrive, uint32_t idxPartition,
4506	char **ppszRawPartition)
4507	{
4508	int rc = VINF_SUCCESS;
4509	DWORD cbReturned = 0;
4510	STORAGE_DEVICE_NUMBER DevNum;
4511	RT_ZERO(DevNum);
4512	if (DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_STORAGE_GET_DEVICE_NUMBER,
4513	NULL /pvInBuffer/, 0 /cbInBuffer/, &DevNum, sizeof(DevNum), &cbReturned, NULL /pOverlapped/))
4514	RTStrAPrintf(ppszRawPartition, "\\\\.\\Harddisk%uPartition%u", DevNum.DeviceNumber, idxPartition);
4515	else
4516	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
4517	N_("VMDK: Image path: '%s'. IOCTL_STORAGE_GET_DEVICE_NUMBER failed on '%s': %u"),
4518	pImage->pszFilename, pszRawDrive, GetLastError());
4519	return rc;
4520	}
4521	#endif /* RT_OS_WINDOWS */
4522
4523	/**
4524	* Worker for vmdkMakeRawDescriptor that adds partition descriptors when the
4525	* 'Partitions' configuration value is present.
4526	*
4527	* @returns VBox status code, error message has been set on failure.
4528	*
4529	* @note Caller is assumed to clean up @a pRawDesc and release
4530	* @a *phVolToRelease.
4531	* @internal
4532	*/
4533	static int vmdkRawDescDoPartitions(PVMDKIMAGE pImage, RTDVM hVolMgr, PVDISKRAW pRawDesc,
4534	RTFILE hRawDrive, const char *pszRawDrive, uint32_t cbSector,
4535	uint32_t fPartitions, uint32_t fPartitionsReadOnly, bool fRelative,
4536	PRTDVMVOLUME phVolToRelease)
4537	{
4538	*phVolToRelease = NIL_RTDVMVOLUME;
4539
4540	/* Check sanity/understanding. */
4541	Assert(fPartitions);
4542	Assert((fPartitions & fPartitionsReadOnly) == fPartitionsReadOnly); /* RO should be a sub-set */
4543
4544	/*
4545	* Allocate on descriptor for each volume up front.
4546	*/
4547	uint32_t const cVolumes = RTDvmMapGetValidVolumes(hVolMgr);
4548
4549	PVDISKRAWPARTDESC paPartDescs = NULL;
4550	int rc = vmdkRawDescAppendPartDesc(pImage, pRawDesc, cVolumes, &paPartDescs);
4551	AssertRCReturn(rc, rc);
4552
4553	/*
4554	* Enumerate the partitions (volumes) on the disk and create descriptors for each of them.
4555	*/
4556	uint32_t fPartitionsLeft = fPartitions;
4557	RTDVMVOLUME hVol = NIL_RTDVMVOLUME; /* the current volume, needed for getting the next. */
4558	for (uint32_t i = 0; i < cVolumes; i++)
4559	{
4560	/*
4561	* Get the next/first volume and release the current.
4562	*/
4563	RTDVMVOLUME hVolNext = NIL_RTDVMVOLUME;
4564	if (i == 0)
4565	rc = RTDvmMapQueryFirstVolume(hVolMgr, &hVolNext);
4566	else
4567	rc = RTDvmMapQueryNextVolume(hVolMgr, hVol, &hVolNext);
4568	if (RT_FAILURE(rc))
4569	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4570	N_("VMDK: Image path: '%s'. Volume enumeration failed at volume #%u on '%s' (%Rrc)"),
4571	pImage->pszFilename, i, pszRawDrive, rc);
4572	uint32_t cRefs = RTDvmVolumeRelease(hVol);
4573	Assert(cRefs != UINT32_MAX); RT_NOREF(cRefs);
4574	*phVolToRelease = hVol = hVolNext;
4575
4576	/*
4577	* Depending on the fPartitions selector and associated read-only mask,
4578	* the guest either gets read-write or read-only access (bits set)
4579	* or no access (selector bit clear, access directed to the VMDK).
4580	*/
4581	paPartDescs[i].cbData = RTDvmVolumeGetSize(hVol);
4582
4583	uint64_t offVolumeEndIgnored = 0;
4584	rc = RTDvmVolumeQueryRange(hVol, &paPartDescs[i].offStartInVDisk, &offVolumeEndIgnored);
4585	if (RT_FAILURE(rc))
4586	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4587	N_("VMDK: Image path: '%s'. Failed to get location of volume #%u on '%s' (%Rrc)"),
4588	pImage->pszFilename, i, pszRawDrive, rc);
4589	Assert(paPartDescs[i].cbData == offVolumeEndIgnored + 1 - paPartDescs[i].offStartInVDisk);
4590
4591	/* Note! The index must match IHostDrivePartition::number. */
4592	uint32_t idxPartition = RTDvmVolumeGetIndex(hVol, RTDVMVOLIDX_HOST);
4593	if ( idxPartition < 32
4594	&& (fPartitions & RT_BIT_32(idxPartition)))
4595	{
4596	fPartitionsLeft &= ~RT_BIT_32(idxPartition);
4597	if (fPartitionsReadOnly & RT_BIT_32(idxPartition))
4598	paPartDescs[i].uFlags \|= VDISKRAW_READONLY;
4599
4600	if (!fRelative)
4601	{
4602	/*
4603	* Accessing the drive thru the main device node (pRawDesc->pszRawDisk).
4604	*/
4605	paPartDescs[i].offStartInDevice = paPartDescs[i].offStartInVDisk;
4606	paPartDescs[i].pszRawDevice = RTStrDup(pszRawDrive);
4607	AssertPtrReturn(paPartDescs[i].pszRawDevice, VERR_NO_STR_MEMORY);
4608	}
4609	else
4610	{
4611	/*
4612	* Relative means access the partition data via the device node for that
4613	* partition, allowing the sysadmin/OS to allow a user access to individual
4614	* partitions without necessarily being able to compromise the host OS.
4615	* Obviously, the creation of the VMDK requires read access to the main
4616	* device node for the drive, but that's a one-time thing and can be done
4617	* by the sysadmin. Here data starts at offset zero in the device node.
4618	*/
4619	paPartDescs[i].offStartInDevice = 0;
4620
4621	#if defined(RT_OS_DARWIN) \|\| defined(RT_OS_FREEBSD)
4622	/* /dev/rdisk1 -> /dev/rdisk1s2 (s=slice) */
4623	RTStrAPrintf(&paPartDescs[i].pszRawDevice, "%ss%u", pszRawDrive, idxPartition);
4624	#elif defined(RT_OS_LINUX)
4625	/* Two naming schemes here: /dev/nvme0n1 -> /dev/nvme0n1p1; /dev/sda -> /dev/sda1 */
4626	RTStrAPrintf(&paPartDescs[i].pszRawDevice,
4627	RT_C_IS_DIGIT(pszRawDrive[strlen(pszRawDrive) - 1]) ? "%sp%u" : "%s%u", pszRawDrive, idxPartition);
4628	#elif defined(RT_OS_WINDOWS)
4629	rc = vmdkRawDescWinMakePartitionName(pImage, pszRawDrive, hRawDrive, idxPartition, &paPartDescs[i].pszRawDevice);
4630	AssertRCReturn(rc, rc);
4631	#elif defined(RT_OS_SOLARIS)
4632	if (pRawDesc->enmPartitioningType == VDISKPARTTYPE_MBR)
4633	{
4634	/*
4635	* MBR partitions have device nodes in form /dev/(r)dsk/cXtYdZpK
4636	* where X is the controller,
4637	* Y is target (SCSI device number),
4638	* Z is disk number,
4639	* K is partition number,
4640	* where p0 is the whole disk
4641	* p1-pN are the partitions of the disk
4642	*/
4643	const char *pszRawDrivePath = pszRawDrive;
4644	char szDrivePath[RTPATH_MAX];
4645	size_t cbRawDrive = strlen(pszRawDrive);
4646	if ( cbRawDrive > 1 && strcmp(&pszRawDrive[cbRawDrive - 2], "p0") == 0)
4647	{
4648	memcpy(szDrivePath, pszRawDrive, cbRawDrive - 2);
4649	szDrivePath[cbRawDrive - 2] = '\0';
4650	pszRawDrivePath = szDrivePath;
4651	}
4652	RTStrAPrintf(&paPartDescs[i].pszRawDevice, "%sp%u", pszRawDrivePath, idxPartition);
4653	}
4654	else /* GPT */
4655	{
4656	/*
4657	* GPT partitions have device nodes in form /dev/(r)dsk/cXtYdZsK
4658	* where X is the controller,
4659	* Y is target (SCSI device number),
4660	* Z is disk number,
4661	* K is partition number, zero based. Can be only from 0 to 6.
4662	* Thus, only partitions numbered 0 through 6 have device nodes.
4663	*/
4664	if (idxPartition > 7)
4665	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4666	N_("VMDK: Image path: '%s'. the partition #%u on '%s' has no device node and can not be specified with 'Relative' property"),
4667	pImage->pszFilename, idxPartition, pszRawDrive);
4668	RTStrAPrintf(&paPartDescs[i].pszRawDevice, "%ss%u", pszRawDrive, idxPartition - 1);
4669	}
4670	#else
4671	AssertFailedReturn(VERR_INTERNAL_ERROR_4); /* The option parsing code should have prevented this - PORTME */
4672	#endif
4673	AssertPtrReturn(paPartDescs[i].pszRawDevice, VERR_NO_STR_MEMORY);
4674
4675	rc = vmdkRawDescVerifyPartitionPath(pImage, &paPartDescs[i], idxPartition, pszRawDrive, hRawDrive, cbSector, hVol);
4676	AssertRCReturn(rc, rc);
4677	}
4678	}
4679	else
4680	{
4681	/* Not accessible to the guest. */
4682	paPartDescs[i].offStartInDevice = 0;
4683	paPartDescs[i].pszRawDevice = NULL;
4684	}
4685	} /* for each volume */
4686
4687	RTDvmVolumeRelease(hVol);
4688	*phVolToRelease = NIL_RTDVMVOLUME;
4689
4690	/*
4691	* Check that we found all the partitions the user selected.
4692	*/
4693	if (fPartitionsLeft)
4694	{
4695	char szLeft[3 * sizeof(fPartitions) * 8];
4696	size_t cchLeft = 0;
4697	for (unsigned i = 0; i < sizeof(fPartitions) * 8; i++)
4698	if (fPartitionsLeft & RT_BIT_32(i))
4699	cchLeft += RTStrPrintf(&szLeft[cchLeft], sizeof(szLeft) - cchLeft, cchLeft ? "%u" : ",%u", i);
4700	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4701	N_("VMDK: Image path: '%s'. Not all the specified partitions for drive '%s' was found: %s"),
4702	pImage->pszFilename, pszRawDrive, szLeft);
4703	}
4704
4705	return VINF_SUCCESS;
4706	}
4707
4708	/**
4709	* Worker for vmdkMakeRawDescriptor that adds partition descriptors with copies
4710	* of the partition tables and associated padding areas when the 'Partitions'
4711	* configuration value is present.
4712	*
4713	* The guest is not allowed access to the partition tables, however it needs
4714	* them to be able to access the drive. So, create descriptors for each of the
4715	* tables and attach the current disk content. vmdkCreateRawImage() will later
4716	* write the content to the VMDK. Any changes the guest later makes to the
4717	* partition tables will then go to the VMDK copy, rather than the host drive.
4718	*
4719	* @returns VBox status code, error message has been set on failure.
4720	*
4721	* @note Caller is assumed to clean up @a pRawDesc
4722	* @internal
4723	*/
4724	static int vmdkRawDescDoCopyPartitionTables(PVMDKIMAGE pImage, RTDVM hVolMgr, PVDISKRAW pRawDesc,
4725	const char pszRawDrive, RTFILE hRawDrive, void pvBootSector, size_t cbBootSector)
4726	{
4727	/*
4728	* Query the locations.
4729	*/
4730	/* Determin how many locations there are: */
4731	size_t cLocations = 0;
4732	int rc = RTDvmMapQueryTableLocations(hVolMgr, RTDVMMAPQTABLOC_F_INCLUDE_LEGACY, NULL, 0, &cLocations);
4733	if (rc != VERR_BUFFER_OVERFLOW)
4734	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4735	N_("VMDK: Image path: '%s'. RTDvmMapQueryTableLocations failed on '%s' (%Rrc)"),
4736	pImage->pszFilename, pszRawDrive, rc);
4737	AssertReturn(cLocations > 0 && cLocations < _16M, VERR_INTERNAL_ERROR_5);
4738
4739	/* We can allocate the partition descriptors here to save an intentation level. */
4740	PVDISKRAWPARTDESC paPartDescs = NULL;
4741	rc = vmdkRawDescAppendPartDesc(pImage, pRawDesc, (uint32_t)cLocations, &paPartDescs);
4742	AssertRCReturn(rc, rc);
4743
4744	/* Allocate the result table and repeat the location table query: */
4745	PRTDVMTABLELOCATION paLocations = (PRTDVMTABLELOCATION)RTMemAllocZ(sizeof(paLocations[0]) * cLocations);
4746	if (!paLocations)
4747	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS, N_("VMDK: Image path: '%s'. Failed to allocate %zu bytes"),
4748	pImage->pszFilename, sizeof(paLocations[0]) * cLocations);
4749	rc = RTDvmMapQueryTableLocations(hVolMgr, RTDVMMAPQTABLOC_F_INCLUDE_LEGACY, paLocations, cLocations, NULL);
4750	if (RT_SUCCESS(rc))
4751	{
4752	/*
4753	* Translate them into descriptors.
4754	*
4755	* We restrict the amount of partition alignment padding to 4MiB as more
4756	* will just be a waste of space. The use case for including the padding
4757	* are older boot loaders and boot manager (including one by a team member)
4758	* that put data and code in the 62 sectors between the MBR and the first
4759	* partition (total of 63). Later CHS was abandond and partition started
4760	* being aligned on power of two sector boundraries (typically 64KiB or
4761	* 1MiB depending on the media size).
4762	*/
4763	for (size_t i = 0; i < cLocations && RT_SUCCESS(rc); i++)
4764	{
4765	Assert(paLocations[i].cb > 0);
4766	if (paLocations[i].cb <= _64M)
4767	{
4768	/* Create the partition descriptor entry: */
4769	//paPartDescs[i].pszRawDevice = NULL;
4770	//paPartDescs[i].offStartInDevice = 0;
4771	//paPartDescs[i].uFlags = 0;
4772	paPartDescs[i].offStartInVDisk = paLocations[i].off;
4773	paPartDescs[i].cbData = paLocations[i].cb;
4774	if (paPartDescs[i].cbData < _4M)
4775	paPartDescs[i].cbData = RT_MIN(paPartDescs[i].cbData + paLocations[i].cbPadding, _4M);
4776	paPartDescs[i].pvPartitionData = RTMemAllocZ((size_t)paPartDescs[i].cbData);
4777	if (paPartDescs[i].pvPartitionData)
4778	{
4779	/* Read the content from the drive: */
4780	rc = RTFileReadAt(hRawDrive, paPartDescs[i].offStartInVDisk, paPartDescs[i].pvPartitionData,
4781	(size_t)paPartDescs[i].cbData, NULL);
4782	if (RT_SUCCESS(rc))
4783	{
4784	/* Do we have custom boot sector code? */
4785	if (pvBootSector && cbBootSector && paPartDescs[i].offStartInVDisk == 0)
4786	{
4787	/* Note! Old code used to quietly drop the bootsector if it was considered too big.
4788	Instead we fail as we weren't able to do what the user requested us to do.
4789	Better if the user knows than starts questioning why the guest isn't
4790	booting as expected. */
4791	if (cbBootSector <= paPartDescs[i].cbData)
4792	memcpy(paPartDescs[i].pvPartitionData, pvBootSector, cbBootSector);
4793	else
4794	rc = vdIfError(pImage->pIfError, VERR_TOO_MUCH_DATA, RT_SRC_POS,
4795	N_("VMDK: Image path: '%s'. The custom boot sector is too big: %zu bytes, %RU64 bytes available"),
4796	pImage->pszFilename, cbBootSector, paPartDescs[i].cbData);
4797	}
4798	}
4799	else
4800	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4801	N_("VMDK: Image path: '%s'. Failed to read partition at off %RU64 length %zu from '%s' (%Rrc)"),
4802	pImage->pszFilename, paPartDescs[i].offStartInVDisk,
4803	(size_t)paPartDescs[i].cbData, pszRawDrive, rc);
4804	}
4805	else
4806	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4807	N_("VMDK: Image path: '%s'. Failed to allocate %zu bytes for copying the partition table at off %RU64"),
4808	pImage->pszFilename, (size_t)paPartDescs[i].cbData, paPartDescs[i].offStartInVDisk);
4809	}
4810	else
4811	rc = vdIfError(pImage->pIfError, VERR_TOO_MUCH_DATA, RT_SRC_POS,
4812	N_("VMDK: Image path: '%s'. Partition table #%u at offset %RU64 in '%s' is to big: %RU64 bytes"),
4813	pImage->pszFilename, i, paLocations[i].off, pszRawDrive, paLocations[i].cb);
4814	}
4815	}
4816	else
4817	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4818	N_("VMDK: Image path: '%s'. RTDvmMapQueryTableLocations failed on '%s' (%Rrc)"),
4819	pImage->pszFilename, pszRawDrive, rc);
4820	RTMemFree(paLocations);
4821	return rc;
4822	}
4823
4824	/**
4825	* Opens the volume manager for the raw drive when in selected-partition mode.
4826	*
4827	* @param pImage The VMDK image (for errors).
4828	* @param hRawDrive The raw drive handle.
4829	* @param pszRawDrive The raw drive device path (for errors).
4830	* @param cbSector The sector size.
4831	* @param phVolMgr Where to return the handle to the volume manager on
4832	* success.
4833	* @returns VBox status code, errors have been reported.
4834	* @internal
4835	*/
4836	static int vmdkRawDescOpenVolMgr(PVMDKIMAGE pImage, RTFILE hRawDrive, const char *pszRawDrive, uint32_t cbSector, PRTDVM phVolMgr)
4837	{
4838	*phVolMgr = NIL_RTDVM;
4839
4840	RTVFSFILE hVfsFile = NIL_RTVFSFILE;
4841	int rc = RTVfsFileFromRTFile(hRawDrive, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE, true /fLeaveOpen/, &hVfsFile);
4842	if (RT_FAILURE(rc))
4843	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4844	N_("VMDK: Image path: '%s'. RTVfsFileFromRTFile failed for '%s' handle (%Rrc)"),
4845	pImage->pszFilename, pszRawDrive, rc);
4846
4847	RTDVM hVolMgr = NIL_RTDVM;
4848	rc = RTDvmCreate(&hVolMgr, hVfsFile, cbSector, 0 /fFlags/);
4849
4850	RTVfsFileRelease(hVfsFile);
4851
4852	if (RT_FAILURE(rc))
4853	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4854	N_("VMDK: Image path: '%s'. Failed to create volume manager instance for '%s' (%Rrc)"),
4855	pImage->pszFilename, pszRawDrive, rc);
4856
4857	rc = RTDvmMapOpen(hVolMgr);
4858	if (RT_SUCCESS(rc))
4859	{
4860	*phVolMgr = hVolMgr;
4861	return VINF_SUCCESS;
4862	}
4863	RTDvmRelease(hVolMgr);
4864	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: Image path: '%s'. RTDvmMapOpen failed for '%s' (%Rrc)"),
4865	pImage->pszFilename, pszRawDrive, rc);
4866	}
4867
4868	/**
4869	* Opens the raw drive device and get the sizes for it.
4870	*
4871	* @param pImage The image (for error reporting).
4872	* @param pszRawDrive The device/whatever to open.
4873	* @param phRawDrive Where to return the file handle.
4874	* @param pcbRawDrive Where to return the size.
4875	* @param pcbSector Where to return the sector size.
4876	* @returns IPRT status code, errors have been reported.
4877	* @internal
4878	*/
4879	static int vmkdRawDescOpenDevice(PVMDKIMAGE pImage, const char *pszRawDrive,
4880	PRTFILE phRawDrive, uint64_t pcbRawDrive, uint32_t pcbSector)
4881	{
4882	/*
4883	* Open the device for the raw drive.
4884	*/
4885	RTFILE hRawDrive = NIL_RTFILE;
4886	int rc = RTFileOpen(&hRawDrive, pszRawDrive, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE);
4887	if (RT_FAILURE(rc))
4888	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4889	N_("VMDK: Image path: '%s'. Failed to open the raw drive '%s' for reading (%Rrc)"),
4890	pImage->pszFilename, pszRawDrive, rc);
4891
4892	/*
4893	* Get the sector size.
4894	*/
4895	uint32_t cbSector = 0;
4896	rc = RTFileQuerySectorSize(hRawDrive, &cbSector);
4897	if (RT_SUCCESS(rc))
4898	{
4899	/* sanity checks */
4900	if ( cbSector >= 512
4901	&& cbSector <= _64K
4902	&& RT_IS_POWER_OF_TWO(cbSector))
4903	{
4904	/*
4905	* Get the size.
4906	*/
4907	uint64_t cbRawDrive = 0;
4908	rc = RTFileQuerySize(hRawDrive, &cbRawDrive);
4909	if (RT_SUCCESS(rc))
4910	{
4911	/* Check whether cbSize is actually sensible. */
4912	if (cbRawDrive > cbSector && (cbRawDrive % cbSector) == 0)
4913	{
4914	*phRawDrive = hRawDrive;
4915	*pcbRawDrive = cbRawDrive;
4916	*pcbSector = cbSector;
4917	return VINF_SUCCESS;
4918	}
4919	rc = vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4920	N_("VMDK: Image path: '%s'. Got a bogus size for the raw drive '%s': %RU64 (sector size %u)"),
4921	pImage->pszFilename, pszRawDrive, cbRawDrive, cbSector);
4922	}
4923	else
4924	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4925	N_("VMDK: Image path: '%s'. Failed to query size of the drive '%s' (%Rrc)"),
4926	pImage->pszFilename, pszRawDrive, rc);
4927	}
4928	else
4929	rc = vdIfError(pImage->pIfError, VERR_OUT_OF_RANGE, RT_SRC_POS,
4930	N_("VMDK: Image path: '%s'. Unsupported sector size for '%s': %u (%#x)"),
4931	pImage->pszFilename, pszRawDrive, cbSector, cbSector);
4932	}
4933	else
4934	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4935	N_("VMDK: Image path: '%s'. Failed to get the sector size for '%s' (%Rrc)"),
4936	pImage->pszFilename, pszRawDrive, rc);
4937	RTFileClose(hRawDrive);
4938	return rc;
4939	}
4940
4941	/**
4942	* Reads the raw disk configuration, leaving initalization and cleanup to the
4943	* caller (regardless of return status).
4944	*
4945	* @returns VBox status code, errors properly reported.
4946	* @internal
4947	*/
4948	static int vmdkRawDescParseConfig(PVMDKIMAGE pImage, char **ppszRawDrive,
4949	uint32_t pfPartitions, uint32_t pfPartitionsReadOnly,
4950	void *ppvBootSector, size_t pcbBootSector, bool *pfRelative,
4951	char **ppszFreeMe)
4952	{
4953	PVDINTERFACECONFIG pImgCfg = VDIfConfigGet(pImage->pVDIfsImage);
4954	if (!pImgCfg)
4955	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4956	N_("VMDK: Image path: '%s'. Getting config interface failed"), pImage->pszFilename);
4957
4958	/*
4959	* RawDrive = path
4960	*/
4961	int rc = VDCFGQueryStringAlloc(pImgCfg, "RawDrive", ppszRawDrive);
4962	if (RT_FAILURE(rc))
4963	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4964	N_("VMDK: Image path: '%s'. Getting 'RawDrive' configuration failed (%Rrc)"), pImage->pszFilename, rc);
4965	AssertPtrReturn(*ppszRawDrive, VERR_INTERNAL_ERROR_3);
4966
4967	/*
4968	* Partitions=n[r][,...]
4969	*/
4970	uint32_t const cMaxPartitionBits = sizeof(pfPartitions) 8 /* ASSUMES 8 bits per char */;
4971	pfPartitions = pfPartitionsReadOnly = 0;
4972
4973	rc = VDCFGQueryStringAlloc(pImgCfg, "Partitions", ppszFreeMe);
4974	if (RT_SUCCESS(rc))
4975	{
4976	char psz = ppszFreeMe;
4977	while (*psz != '\0')
4978	{
4979	char *pszNext;
4980	uint32_t u32;
4981	rc = RTStrToUInt32Ex(psz, &pszNext, 0, &u32);
4982	if (rc == VWRN_NUMBER_TOO_BIG \|\| rc == VWRN_NEGATIVE_UNSIGNED)
4983	rc = -rc;
4984	if (RT_FAILURE(rc))
4985	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4986	N_("VMDK: Image path: '%s'. Parsing 'Partitions' config value failed. Incorrect value (%Rrc): %s"),
4987	pImage->pszFilename, rc, psz);
4988	if (u32 >= cMaxPartitionBits)
4989	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4990	N_("VMDK: Image path: '%s'. 'Partitions' config sub-value out of range: %RU32, max %RU32"),
4991	pImage->pszFilename, u32, cMaxPartitionBits);
4992	*pfPartitions \|= RT_BIT_32(u32);
4993	psz = pszNext;
4994	if (*psz == 'r')
4995	{
4996	*pfPartitionsReadOnly \|= RT_BIT_32(u32);
4997	psz++;
4998	}
4999	if (*psz == ',')
5000	psz++;
5001	else if (*psz != '\0')
5002	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5003	N_("VMDK: Image path: '%s'. Malformed 'Partitions' config value, expected separator: %s"),
5004	pImage->pszFilename, psz);
5005	}
5006
5007	RTStrFree(*ppszFreeMe);
5008	*ppszFreeMe = NULL;
5009	}
5010	else if (rc != VERR_CFGM_VALUE_NOT_FOUND)
5011	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5012	N_("VMDK: Image path: '%s'. Getting 'Partitions' configuration failed (%Rrc)"), pImage->pszFilename, rc);
5013
5014	/*
5015	* BootSector=base64
5016	*/
5017	rc = VDCFGQueryStringAlloc(pImgCfg, "BootSector", ppszFreeMe);
5018	if (RT_SUCCESS(rc))
5019	{
5020	ssize_t cbBootSector = RTBase64DecodedSize(*ppszFreeMe, NULL);
5021	if (cbBootSector < 0)
5022	return vdIfError(pImage->pIfError, VERR_INVALID_BASE64_ENCODING, RT_SRC_POS,
5023	N_("VMDK: Image path: '%s'. BASE64 decoding failed on the custom bootsector for '%s'"),
5024	pImage->pszFilename, *ppszRawDrive);
5025	if (cbBootSector == 0)
5026	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5027	N_("VMDK: Image path: '%s'. Custom bootsector for '%s' is zero bytes big"),
5028	pImage->pszFilename, *ppszRawDrive);
5029	if (cbBootSector > _4M) /* this is just a preliminary max */
5030	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5031	N_("VMDK: Image path: '%s'. Custom bootsector for '%s' is way too big: %zu bytes, max 4MB"),
5032	pImage->pszFilename, *ppszRawDrive, cbBootSector);
5033
5034	/* Refuse the boot sector if whole-drive. This used to be done quietly,
5035	however, bird disagrees and thinks the user should be told that what
5036	he/she/it tries to do isn't possible. There should be less head
5037	scratching this way when the guest doesn't do the expected thing. */
5038	if (!*pfPartitions)
5039	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5040	N_("VMDK: Image path: '%s'. Custom bootsector for '%s' is not supported for whole-drive configurations, only when selecting partitions"),
5041	pImage->pszFilename, *ppszRawDrive);
5042
5043	*pcbBootSector = (size_t)cbBootSector;
5044	*ppvBootSector = RTMemAlloc((size_t)cbBootSector);
5045	if (!*ppvBootSector)
5046	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS,
5047	N_("VMDK: Image path: '%s'. Failed to allocate %zd bytes for the custom bootsector for '%s'"),
5048	pImage->pszFilename, cbBootSector, *ppszRawDrive);
5049
5050	rc = RTBase64Decode(ppszFreeMe, ppvBootSector, cbBootSector, NULL /pcbActual/, NULL /ppszEnd/);
5051	if (RT_FAILURE(rc))
5052	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS,
5053	N_("VMDK: Image path: '%s'. Base64 decoding of the custom boot sector for '%s' failed (%Rrc)"),
5054	pImage->pszFilename, *ppszRawDrive, rc);
5055
5056	RTStrFree(*ppszFreeMe);
5057	*ppszFreeMe = NULL;
5058	}
5059	else if (rc != VERR_CFGM_VALUE_NOT_FOUND)
5060	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5061	N_("VMDK: Image path: '%s'. Getting 'BootSector' configuration failed (%Rrc)"), pImage->pszFilename, rc);
5062
5063	/*
5064	* Relative=0/1
5065	*/
5066	*pfRelative = false;
5067	rc = VDCFGQueryBool(pImgCfg, "Relative", pfRelative);
5068	if (RT_SUCCESS(rc))
5069	{
5070	if (!pfPartitions && pfRelative != false)
5071	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5072	N_("VMDK: Image path: '%s'. The 'Relative' option is not supported for whole-drive configurations, only when selecting partitions"),
5073	pImage->pszFilename);
5074	#if !defined(RT_OS_DARWIN) && !defined(RT_OS_LINUX) && !defined(RT_OS_FREEBSD) && !defined(RT_OS_WINDOWS) && !defined(RT_OS_SOLARIS) /* PORTME */
5075	if (*pfRelative == true)
5076	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5077	N_("VMDK: Image path: '%s'. The 'Relative' option is not supported on this host OS"),
5078	pImage->pszFilename);
5079	#endif
5080	}
5081	else if (rc != VERR_CFGM_VALUE_NOT_FOUND)
5082	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5083	N_("VMDK: Image path: '%s'. Getting 'Relative' configuration failed (%Rrc)"), pImage->pszFilename, rc);
5084	else
5085	#ifdef RT_OS_DARWIN /* different default on macOS, see ticketref:1461 (comment 20). */
5086	*pfRelative = true;
5087	#else
5088	*pfRelative = false;
5089	#endif
5090
5091	return VINF_SUCCESS;
5092	}
5093
5094	/**
5095	* Creates a raw drive (nee disk) descriptor.
5096	*
5097	* This was originally done in VBoxInternalManage.cpp, but was copied (not move)
5098	* here much later. That's one of the reasons why we produce a descriptor just
5099	* like it does, rather than mixing directly into the vmdkCreateRawImage code.
5100	*
5101	* @returns VBox status code.
5102	* @param pImage The image.
5103	* @param ppRaw Where to return the raw drive descriptor. Caller must
5104	* free it using vmdkRawDescFree regardless of the status
5105	* code.
5106	* @internal
5107	*/
5108	static int vmdkMakeRawDescriptor(PVMDKIMAGE pImage, PVDISKRAW *ppRaw)
5109	{
5110	/* Make sure it's NULL. */
5111	*ppRaw = NULL;
5112
5113	/*
5114	* Read the configuration.
5115	*/
5116	char *pszRawDrive = NULL;
5117	uint32_t fPartitions = 0; /* zero if whole-drive */
5118	uint32_t fPartitionsReadOnly = 0; /* (subset of fPartitions) */
5119	void *pvBootSector = NULL;
5120	size_t cbBootSector = 0;
5121	bool fRelative = false;
5122	char pszFreeMe = NULL; / lazy bird cleanup. */
5123	int rc = vmdkRawDescParseConfig(pImage, &pszRawDrive, &fPartitions, &fPartitionsReadOnly,
5124	&pvBootSector, &cbBootSector, &fRelative, &pszFreeMe);
5125	RTStrFree(pszFreeMe);
5126	if (RT_SUCCESS(rc))
5127	{
5128	/*
5129	* Open the device, getting the sector size and drive size.
5130	*/
5131	uint64_t cbSize = 0;
5132	uint32_t cbSector = 0;
5133	RTFILE hRawDrive = NIL_RTFILE;
5134	rc = vmkdRawDescOpenDevice(pImage, pszRawDrive, &hRawDrive, &cbSize, &cbSector);
5135	if (RT_SUCCESS(rc))
5136	{
5137	pImage->cbSize = cbSize;
5138	/*
5139	* Create the raw-drive descriptor
5140	*/
5141	PVDISKRAW pRawDesc = (PVDISKRAW)RTMemAllocZ(sizeof(*pRawDesc));
5142	if (pRawDesc)
5143	{
5144	pRawDesc->szSignature[0] = 'R';
5145	pRawDesc->szSignature[1] = 'A';
5146	pRawDesc->szSignature[2] = 'W';
5147	//pRawDesc->szSignature[3] = '\0';
5148	if (!fPartitions)
5149	{
5150	/*
5151	* It's simple for when doing the whole drive.
5152	*/
5153	pRawDesc->uFlags = VDISKRAW_DISK;
5154	rc = RTStrDupEx(&pRawDesc->pszRawDisk, pszRawDrive);
5155	}
5156	else
5157	{
5158	/*
5159	* In selected partitions mode we've got a lot more work ahead of us.
5160	*/
5161	pRawDesc->uFlags = VDISKRAW_NORMAL;
5162	//pRawDesc->pszRawDisk = NULL;
5163	//pRawDesc->cPartDescs = 0;
5164	//pRawDesc->pPartDescs = NULL;
5165
5166	/* We need to parse the partition map to complete the descriptor: */
5167	RTDVM hVolMgr = NIL_RTDVM;
5168	rc = vmdkRawDescOpenVolMgr(pImage, hRawDrive, pszRawDrive, cbSector, &hVolMgr);
5169	if (RT_SUCCESS(rc))
5170	{
5171	RTDVMFORMATTYPE enmFormatType = RTDvmMapGetFormatType(hVolMgr);
5172	if ( enmFormatType == RTDVMFORMATTYPE_MBR
5173	\|\| enmFormatType == RTDVMFORMATTYPE_GPT)
5174	{
5175	pRawDesc->enmPartitioningType = enmFormatType == RTDVMFORMATTYPE_MBR
5176	? VDISKPARTTYPE_MBR : VDISKPARTTYPE_GPT;
5177
5178	/* Add copies of the partition tables: */
5179	rc = vmdkRawDescDoCopyPartitionTables(pImage, hVolMgr, pRawDesc, pszRawDrive, hRawDrive,
5180	pvBootSector, cbBootSector);
5181	if (RT_SUCCESS(rc))
5182	{
5183	/* Add descriptors for the partitions/volumes, indicating which
5184	should be accessible and how to access them: */
5185	RTDVMVOLUME hVolRelease = NIL_RTDVMVOLUME;
5186	rc = vmdkRawDescDoPartitions(pImage, hVolMgr, pRawDesc, hRawDrive, pszRawDrive, cbSector,
5187	fPartitions, fPartitionsReadOnly, fRelative, &hVolRelease);
5188	RTDvmVolumeRelease(hVolRelease);
5189
5190	/* Finally, sort the partition and check consistency (overlaps, etc): */
5191	if (RT_SUCCESS(rc))
5192	rc = vmdkRawDescPostProcessPartitions(pImage, pRawDesc, cbSize);
5193	}
5194	}
5195	else
5196	rc = vdIfError(pImage->pIfError, VERR_NOT_SUPPORTED, RT_SRC_POS,
5197	N_("VMDK: Image path: '%s'. Unsupported partitioning for the disk '%s': %s"),
5198	pImage->pszFilename, pszRawDrive, RTDvmMapGetFormatType(hVolMgr));
5199	RTDvmRelease(hVolMgr);
5200	}
5201	}
5202	if (RT_SUCCESS(rc))
5203	{
5204	/*
5205	* We succeeded.
5206	*/
5207	*ppRaw = pRawDesc;
5208	Log(("vmdkMakeRawDescriptor: fFlags=%#x enmPartitioningType=%d cPartDescs=%u pszRawDisk=%s\n",
5209	pRawDesc->uFlags, pRawDesc->enmPartitioningType, pRawDesc->cPartDescs, pRawDesc->pszRawDisk));
5210	if (pRawDesc->cPartDescs)
5211	{
5212	Log(("# VMDK offset Length Device offset PartDataPtr Device\n"));
5213	for (uint32_t i = 0; i < pRawDesc->cPartDescs; i++)
5214	Log(("%2u %14RU64 %14RU64 %14RU64 %#18p %s\n", i, pRawDesc->pPartDescs[i].offStartInVDisk,
5215	pRawDesc->pPartDescs[i].cbData, pRawDesc->pPartDescs[i].offStartInDevice,
5216	pRawDesc->pPartDescs[i].pvPartitionData, pRawDesc->pPartDescs[i].pszRawDevice));
5217	}
5218	}
5219	else
5220	vmdkRawDescFree(pRawDesc);
5221	}
5222	else
5223	rc = vdIfError(pImage->pIfError, VERR_NOT_SUPPORTED, RT_SRC_POS,
5224	N_("VMDK: Image path: '%s'. Failed to allocate %u bytes for the raw drive descriptor"),
5225	pImage->pszFilename, sizeof(*pRawDesc));
5226	RTFileClose(hRawDrive);
5227	}
5228	}
5229	RTStrFree(pszRawDrive);
5230	RTMemFree(pvBootSector);
5231	return rc;
5232	}
5233
5234	/**
5235	* Internal: create VMDK images for raw disk/partition access.
5236	*/
5237	static int vmdkCreateRawImage(PVMDKIMAGE pImage, const PVDISKRAW pRaw,
5238	uint64_t cbSize)
5239	{
5240	int rc = VINF_SUCCESS;
5241	PVMDKEXTENT pExtent;
5242
5243	if (pRaw->uFlags & VDISKRAW_DISK)
5244	{
5245	/* Full raw disk access. This requires setting up a descriptor
5246	* file and open the (flat) raw disk. */
5247	rc = vmdkCreateExtents(pImage, 1);
5248	if (RT_FAILURE(rc))
5249	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
5250	pExtent = &pImage->pExtents[0];
5251	/* Create raw disk descriptor file. */
5252	rc = vmdkFileOpen(pImage, &pImage->pFile, NULL, pImage->pszFilename,
5253	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5254	true /* fCreate */));
5255	if (RT_FAILURE(rc))
5256	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pImage->pszFilename);
5257
5258	/* Set up basename for extent description. Cannot use StrDup. */
5259	size_t cbBasename = strlen(pRaw->pszRawDisk) + 1;
5260	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
5261	if (!pszBasename)
5262	return VERR_NO_MEMORY;
5263	memcpy(pszBasename, pRaw->pszRawDisk, cbBasename);
5264	pExtent->pszBasename = pszBasename;
5265	/* For raw disks the full name is identical to the base name. */
5266	pExtent->pszFullname = RTStrDup(pszBasename);
5267	if (!pExtent->pszFullname)
5268	return VERR_NO_MEMORY;
5269	pExtent->enmType = VMDKETYPE_FLAT;
5270	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
5271	pExtent->uSectorOffset = 0;
5272	pExtent->enmAccess = (pRaw->uFlags & VDISKRAW_READONLY) ? VMDKACCESS_READONLY : VMDKACCESS_READWRITE;
5273	pExtent->fMetaDirty = false;
5274
5275	/* Open flat image, the raw disk. */
5276	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5277	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
5278	false /* fCreate */));
5279	if (RT_FAILURE(rc))
5280	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not open raw disk file '%s'"), pExtent->pszFullname);
5281	}
5282	else
5283	{
5284	/* Raw partition access. This requires setting up a descriptor
5285	* file, write the partition information to a flat extent and
5286	* open all the (flat) raw disk partitions. */
5287
5288	/* First pass over the partition data areas to determine how many
5289	* extents we need. One data area can require up to 2 extents, as
5290	* it might be necessary to skip over unpartitioned space. */
5291	unsigned cExtents = 0;
5292	uint64_t uStart = 0;
5293	for (unsigned i = 0; i < pRaw->cPartDescs; i++)
5294	{
5295	PVDISKRAWPARTDESC pPart = &pRaw->pPartDescs[i];
5296	if (uStart > pPart->offStartInVDisk)
5297	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5298	N_("VMDK: incorrect partition data area ordering set up by the caller in '%s'"), pImage->pszFilename);
5299
5300	if (uStart < pPart->offStartInVDisk)
5301	cExtents++;
5302	uStart = pPart->offStartInVDisk + pPart->cbData;
5303	cExtents++;
5304	}
5305	/* Another extent for filling up the rest of the image. */
5306	if (uStart != cbSize)
5307	cExtents++;
5308
5309	rc = vmdkCreateExtents(pImage, cExtents);
5310	if (RT_FAILURE(rc))
5311	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
5312
5313	/* Create raw partition descriptor file. */
5314	rc = vmdkFileOpen(pImage, &pImage->pFile, NULL, pImage->pszFilename,
5315	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5316	true /* fCreate */));
5317	if (RT_FAILURE(rc))
5318	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pImage->pszFilename);
5319
5320	/* Create base filename for the partition table extent. */
5321	/** @todo remove fixed buffer without creating memory leaks. */
5322	char pszPartition[1024];
5323	const char *pszBase = RTPathFilename(pImage->pszFilename);
5324	const char *pszSuff = RTPathSuffix(pszBase);
5325	if (pszSuff == NULL)
5326	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: invalid filename '%s'"), pImage->pszFilename);
5327	char *pszBaseBase = RTStrDup(pszBase);
5328	if (!pszBaseBase)
5329	return VERR_NO_MEMORY;
5330	RTPathStripSuffix(pszBaseBase);
5331	RTStrPrintf(pszPartition, sizeof(pszPartition), "%s-pt%s",
5332	pszBaseBase, pszSuff);
5333	RTStrFree(pszBaseBase);
5334
5335	/* Second pass over the partitions, now define all extents. */
5336	uint64_t uPartOffset = 0;
5337	cExtents = 0;
5338	uStart = 0;
5339	for (unsigned i = 0; i < pRaw->cPartDescs; i++)
5340	{
5341	PVDISKRAWPARTDESC pPart = &pRaw->pPartDescs[i];
5342	pExtent = &pImage->pExtents[cExtents++];
5343
5344	if (uStart < pPart->offStartInVDisk)
5345	{
5346	pExtent->pszBasename = NULL;
5347	pExtent->pszFullname = NULL;
5348	pExtent->enmType = VMDKETYPE_ZERO;
5349	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->offStartInVDisk - uStart);
5350	pExtent->uSectorOffset = 0;
5351	pExtent->enmAccess = VMDKACCESS_READWRITE;
5352	pExtent->fMetaDirty = false;
5353	/* go to next extent */
5354	pExtent = &pImage->pExtents[cExtents++];
5355	}
5356	uStart = pPart->offStartInVDisk + pPart->cbData;
5357
5358	if (pPart->pvPartitionData)
5359	{
5360	/* Set up basename for extent description. Can't use StrDup. */
5361	size_t cbBasename = strlen(pszPartition) + 1;
5362	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
5363	if (!pszBasename)
5364	return VERR_NO_MEMORY;
5365	memcpy(pszBasename, pszPartition, cbBasename);
5366	pExtent->pszBasename = pszBasename;
5367
5368	/* Set up full name for partition extent. */
5369	char *pszDirname = RTStrDup(pImage->pszFilename);
5370	if (!pszDirname)
5371	return VERR_NO_STR_MEMORY;
5372	RTPathStripFilename(pszDirname);
5373	char *pszFullname = RTPathJoinA(pszDirname, pExtent->pszBasename);
5374	RTStrFree(pszDirname);
5375	if (!pszFullname)
5376	return VERR_NO_STR_MEMORY;
5377	pExtent->pszFullname = pszFullname;
5378	pExtent->enmType = VMDKETYPE_FLAT;
5379	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
5380	pExtent->uSectorOffset = uPartOffset;
5381	pExtent->enmAccess = VMDKACCESS_READWRITE;
5382	pExtent->fMetaDirty = false;
5383
5384	/* Create partition table flat image. */
5385	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5386	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
5387	true /* fCreate */));
5388	if (RT_FAILURE(rc))
5389	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new partition data file '%s'"), pExtent->pszFullname);
5390	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
5391	VMDK_SECTOR2BYTE(uPartOffset),
5392	pPart->pvPartitionData,
5393	pPart->cbData);
5394	if (RT_FAILURE(rc))
5395	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not write partition data to '%s'"), pExtent->pszFullname);
5396	uPartOffset += VMDK_BYTE2SECTOR(pPart->cbData);
5397	}
5398	else
5399	{
5400	if (pPart->pszRawDevice)
5401	{
5402	/* Set up basename for extent descr. Can't use StrDup. */
5403	size_t cbBasename = strlen(pPart->pszRawDevice) + 1;
5404	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
5405	if (!pszBasename)
5406	return VERR_NO_MEMORY;
5407	memcpy(pszBasename, pPart->pszRawDevice, cbBasename);
5408	pExtent->pszBasename = pszBasename;
5409	/* For raw disks full name is identical to base name. */
5410	pExtent->pszFullname = RTStrDup(pszBasename);
5411	if (!pExtent->pszFullname)
5412	return VERR_NO_MEMORY;
5413	pExtent->enmType = VMDKETYPE_FLAT;
5414	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
5415	pExtent->uSectorOffset = VMDK_BYTE2SECTOR(pPart->offStartInDevice);
5416	pExtent->enmAccess = (pPart->uFlags & VDISKRAW_READONLY) ? VMDKACCESS_READONLY : VMDKACCESS_READWRITE;
5417	pExtent->fMetaDirty = false;
5418
5419	/* Open flat image, the raw partition. */
5420	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5421	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
5422	false /* fCreate */));
5423	if (RT_FAILURE(rc))
5424	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not open raw partition file '%s'"), pExtent->pszFullname);
5425	}
5426	else
5427	{
5428	pExtent->pszBasename = NULL;
5429	pExtent->pszFullname = NULL;
5430	pExtent->enmType = VMDKETYPE_ZERO;
5431	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
5432	pExtent->uSectorOffset = 0;
5433	pExtent->enmAccess = VMDKACCESS_READWRITE;
5434	pExtent->fMetaDirty = false;
5435	}
5436	}
5437	}
5438	/* Another extent for filling up the rest of the image. */
5439	if (uStart != cbSize)
5440	{
5441	pExtent = &pImage->pExtents[cExtents++];
5442	pExtent->pszBasename = NULL;
5443	pExtent->pszFullname = NULL;
5444	pExtent->enmType = VMDKETYPE_ZERO;
5445	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize - uStart);
5446	pExtent->uSectorOffset = 0;
5447	pExtent->enmAccess = VMDKACCESS_READWRITE;
5448	pExtent->fMetaDirty = false;
5449	}
5450	}
5451
5452	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
5453	(pRaw->uFlags & VDISKRAW_DISK) ?
5454	"fullDevice" : "partitionedDevice");
5455	if (RT_FAILURE(rc))
5456	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
5457	return rc;
5458	}
5459
5460	/**
5461	* Internal: create a regular (i.e. file-backed) VMDK image.
5462	*/
5463	static int vmdkCreateRegularImage(PVMDKIMAGE pImage, uint64_t cbSize,
5464	unsigned uImageFlags, PVDINTERFACEPROGRESS pIfProgress,
5465	unsigned uPercentStart, unsigned uPercentSpan)
5466	{
5467	int rc = VINF_SUCCESS;
5468	unsigned cExtents = 1;
5469	uint64_t cbOffset = 0;
5470	uint64_t cbRemaining = cbSize;
5471
5472	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
5473	{
5474	cExtents = cbSize / VMDK_2G_SPLIT_SIZE;
5475	/* Do proper extent computation: need one smaller extent if the total
5476	* size isn't evenly divisible by the split size. */
5477	if (cbSize % VMDK_2G_SPLIT_SIZE)
5478	cExtents++;
5479	}
5480	rc = vmdkCreateExtents(pImage, cExtents);
5481	if (RT_FAILURE(rc))
5482	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
5483
5484	/* Basename strings needed for constructing the extent names. */
5485	char *pszBasenameSubstr = RTPathFilename(pImage->pszFilename);
5486	AssertPtr(pszBasenameSubstr);
5487	size_t cbBasenameSubstr = strlen(pszBasenameSubstr) + 1;
5488
5489	/* Create separate descriptor file if necessary. */
5490	if (cExtents != 1 \|\| (uImageFlags & VD_IMAGE_FLAGS_FIXED))
5491	{
5492	rc = vmdkFileOpen(pImage, &pImage->pFile, NULL, pImage->pszFilename,
5493	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5494	true /* fCreate */));
5495	if (RT_FAILURE(rc))
5496	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new sparse descriptor file '%s'"), pImage->pszFilename);
5497	}
5498	else
5499	pImage->pFile = NULL;
5500
5501	/* Set up all extents. */
5502	for (unsigned i = 0; i < cExtents; i++)
5503	{
5504	PVMDKEXTENT pExtent = &pImage->pExtents[i];
5505	uint64_t cbExtent = cbRemaining;
5506
5507	/* Set up fullname/basename for extent description. Cannot use StrDup
5508	* for basename, as it is not guaranteed that the memory can be freed
5509	* with RTMemTmpFree, which must be used as in other code paths
5510	* StrDup is not usable. */
5511	if (cExtents == 1 && !(uImageFlags & VD_IMAGE_FLAGS_FIXED))
5512	{
5513	char pszBasename = (char )RTMemTmpAlloc(cbBasenameSubstr);
5514	if (!pszBasename)
5515	return VERR_NO_MEMORY;
5516	memcpy(pszBasename, pszBasenameSubstr, cbBasenameSubstr);
5517	pExtent->pszBasename = pszBasename;
5518	}
5519	else
5520	{
5521	char *pszBasenameSuff = RTPathSuffix(pszBasenameSubstr);
5522	char *pszBasenameBase = RTStrDup(pszBasenameSubstr);
5523	RTPathStripSuffix(pszBasenameBase);
5524	char *pszTmp;
5525	size_t cbTmp;
5526	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
5527	{
5528	if (cExtents == 1)
5529	RTStrAPrintf(&pszTmp, "%s-flat%s", pszBasenameBase,
5530	pszBasenameSuff);
5531	else
5532	RTStrAPrintf(&pszTmp, "%s-f%03d%s", pszBasenameBase,
5533	i+1, pszBasenameSuff);
5534	}
5535	else
5536	RTStrAPrintf(&pszTmp, "%s-s%03d%s", pszBasenameBase, i+1,
5537	pszBasenameSuff);
5538	RTStrFree(pszBasenameBase);
5539	if (!pszTmp)
5540	return VERR_NO_STR_MEMORY;
5541	cbTmp = strlen(pszTmp) + 1;
5542	char pszBasename = (char )RTMemTmpAlloc(cbTmp);
5543	if (!pszBasename)
5544	{
5545	RTStrFree(pszTmp);
5546	return VERR_NO_MEMORY;
5547	}
5548	memcpy(pszBasename, pszTmp, cbTmp);
5549	RTStrFree(pszTmp);
5550	pExtent->pszBasename = pszBasename;
5551	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
5552	cbExtent = RT_MIN(cbRemaining, VMDK_2G_SPLIT_SIZE);
5553	}
5554	char *pszBasedirectory = RTStrDup(pImage->pszFilename);
5555	if (!pszBasedirectory)
5556	return VERR_NO_STR_MEMORY;
5557	RTPathStripFilename(pszBasedirectory);
5558	char *pszFullname = RTPathJoinA(pszBasedirectory, pExtent->pszBasename);
5559	RTStrFree(pszBasedirectory);
5560	if (!pszFullname)
5561	return VERR_NO_STR_MEMORY;
5562	pExtent->pszFullname = pszFullname;
5563
5564	/* Create file for extent. */
5565	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5566	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5567	true /* fCreate */));
5568	if (RT_FAILURE(rc))
5569	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pExtent->pszFullname);
5570	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
5571	{
5572	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage, cbExtent,
5573	0 /* fFlags */, pIfProgress,
5574	uPercentStart + cbOffset * uPercentSpan / cbSize,
5575	cbExtent * uPercentSpan / cbSize);
5576	if (RT_FAILURE(rc))
5577	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
5578	}
5579
5580	/* Place descriptor file information (where integrated). */
5581	if (cExtents == 1 && !(uImageFlags & VD_IMAGE_FLAGS_FIXED))
5582	{
5583	pExtent->uDescriptorSector = 1;
5584	pExtent->cDescriptorSectors = VMDK_BYTE2SECTOR(pImage->cbDescAlloc);
5585	/* The descriptor is part of the (only) extent. */
5586	pExtent->pDescData = pImage->pDescData;
5587	pImage->pDescData = NULL;
5588	}
5589
5590	if (!(uImageFlags & VD_IMAGE_FLAGS_FIXED))
5591	{
5592	uint64_t cSectorsPerGDE, cSectorsPerGD;
5593	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE;
5594	pExtent->cSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64(cbExtent, _64K));
5595	pExtent->cSectorsPerGrain = VMDK_BYTE2SECTOR(_64K);
5596	pExtent->cGTEntries = 512;
5597	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
5598	pExtent->cSectorsPerGDE = cSectorsPerGDE;
5599	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
5600	cSectorsPerGD = (pExtent->cGDEntries + (512 / sizeof(uint32_t) - 1)) / (512 / sizeof(uint32_t));
5601	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5602	{
5603	/* The spec says version is 1 for all VMDKs, but the vast
5604	* majority of streamOptimized VMDKs actually contain
5605	* version 3 - so go with the majority. Both are accepted. */
5606	pExtent->uVersion = 3;
5607	pExtent->uCompression = VMDK_COMPRESSION_DEFLATE;
5608	}
5609	}
5610	else
5611	{
5612	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
5613	pExtent->enmType = VMDKETYPE_VMFS;
5614	else
5615	pExtent->enmType = VMDKETYPE_FLAT;
5616	}
5617
5618	pExtent->enmAccess = VMDKACCESS_READWRITE;
5619	pExtent->fUncleanShutdown = true;
5620	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbExtent);
5621	pExtent->uSectorOffset = 0;
5622	pExtent->fMetaDirty = true;
5623
5624	if (!(uImageFlags & VD_IMAGE_FLAGS_FIXED))
5625	{
5626	/* fPreAlloc should never be false because VMware can't use such images. */
5627	rc = vmdkCreateGrainDirectory(pImage, pExtent,
5628	RT_MAX( pExtent->uDescriptorSector
5629	+ pExtent->cDescriptorSectors,
5630	1),
5631	true /* fPreAlloc */);
5632	if (RT_FAILURE(rc))
5633	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new grain directory in '%s'"), pExtent->pszFullname);
5634	}
5635
5636	cbOffset += cbExtent;
5637
5638	if (RT_SUCCESS(rc))
5639	vdIfProgress(pIfProgress, uPercentStart + cbOffset * uPercentSpan / cbSize);
5640
5641	cbRemaining -= cbExtent;
5642	}
5643
5644	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
5645	{
5646	/* VirtualBox doesn't care, but VMWare ESX freaks out if the wrong
5647	* controller type is set in an image. */
5648	rc = vmdkDescDDBSetStr(pImage, &pImage->Descriptor, "ddb.adapterType", "lsilogic");
5649	if (RT_FAILURE(rc))
5650	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set controller type to lsilogic in '%s'"), pImage->pszFilename);
5651	}
5652
5653	const char *pszDescType = NULL;
5654	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
5655	{
5656	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
5657	pszDescType = "vmfs";
5658	else
5659	pszDescType = (cExtents == 1)
5660	? "monolithicFlat" : "twoGbMaxExtentFlat";
5661	}
5662	else
5663	{
5664	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5665	pszDescType = "streamOptimized";
5666	else
5667	{
5668	pszDescType = (cExtents == 1)
5669	? "monolithicSparse" : "twoGbMaxExtentSparse";
5670	}
5671	}
5672	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
5673	pszDescType);
5674	if (RT_FAILURE(rc))
5675	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
5676	return rc;
5677	}
5678
5679	/**
5680	* Internal: Create a real stream optimized VMDK using only linear writes.
5681	*/
5682	static int vmdkCreateStreamImage(PVMDKIMAGE pImage, uint64_t cbSize)
5683	{
5684	int rc = vmdkCreateExtents(pImage, 1);
5685	if (RT_FAILURE(rc))
5686	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
5687
5688	/* Basename strings needed for constructing the extent names. */
5689	const char *pszBasenameSubstr = RTPathFilename(pImage->pszFilename);
5690	AssertPtr(pszBasenameSubstr);
5691	size_t cbBasenameSubstr = strlen(pszBasenameSubstr) + 1;
5692
5693	/* No separate descriptor file. */
5694	pImage->pFile = NULL;
5695
5696	/* Set up all extents. */
5697	PVMDKEXTENT pExtent = &pImage->pExtents[0];
5698
5699	/* Set up fullname/basename for extent description. Cannot use StrDup
5700	* for basename, as it is not guaranteed that the memory can be freed
5701	* with RTMemTmpFree, which must be used as in other code paths
5702	* StrDup is not usable. */
5703	char pszBasename = (char )RTMemTmpAlloc(cbBasenameSubstr);
5704	if (!pszBasename)
5705	return VERR_NO_MEMORY;
5706	memcpy(pszBasename, pszBasenameSubstr, cbBasenameSubstr);
5707	pExtent->pszBasename = pszBasename;
5708
5709	char *pszBasedirectory = RTStrDup(pImage->pszFilename);
5710	RTPathStripFilename(pszBasedirectory);
5711	char *pszFullname = RTPathJoinA(pszBasedirectory, pExtent->pszBasename);
5712	RTStrFree(pszBasedirectory);
5713	if (!pszFullname)
5714	return VERR_NO_STR_MEMORY;
5715	pExtent->pszFullname = pszFullname;
5716
5717	/* Create file for extent. Make it write only, no reading allowed. */
5718	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5719	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5720	true /* fCreate */)
5721	& ~RTFILE_O_READ);
5722	if (RT_FAILURE(rc))
5723	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pExtent->pszFullname);
5724
5725	/* Place descriptor file information. */
5726	pExtent->uDescriptorSector = 1;
5727	pExtent->cDescriptorSectors = VMDK_BYTE2SECTOR(pImage->cbDescAlloc);
5728	/* The descriptor is part of the (only) extent. */
5729	pExtent->pDescData = pImage->pDescData;
5730	pImage->pDescData = NULL;
5731
5732	uint64_t cSectorsPerGDE, cSectorsPerGD;
5733	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE;
5734	pExtent->cSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64(cbSize, _64K));
5735	pExtent->cSectorsPerGrain = VMDK_BYTE2SECTOR(_64K);
5736	pExtent->cGTEntries = 512;
5737	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
5738	pExtent->cSectorsPerGDE = cSectorsPerGDE;
5739	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
5740	cSectorsPerGD = (pExtent->cGDEntries + (512 / sizeof(uint32_t) - 1)) / (512 / sizeof(uint32_t));
5741
5742	/* The spec says version is 1 for all VMDKs, but the vast
5743	* majority of streamOptimized VMDKs actually contain
5744	* version 3 - so go with the majority. Both are accepted. */
5745	pExtent->uVersion = 3;
5746	pExtent->uCompression = VMDK_COMPRESSION_DEFLATE;
5747	pExtent->fFooter = true;
5748
5749	pExtent->enmAccess = VMDKACCESS_READONLY;
5750	pExtent->fUncleanShutdown = false;
5751	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
5752	pExtent->uSectorOffset = 0;
5753	pExtent->fMetaDirty = true;
5754
5755	/* Create grain directory, without preallocating it straight away. It will
5756	* be constructed on the fly when writing out the data and written when
5757	* closing the image. The end effect is that the full grain directory is
5758	* allocated, which is a requirement of the VMDK specs. */
5759	rc = vmdkCreateGrainDirectory(pImage, pExtent, VMDK_GD_AT_END,
5760	false /* fPreAlloc */);
5761	if (RT_FAILURE(rc))
5762	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new grain directory in '%s'"), pExtent->pszFullname);
5763
5764	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
5765	"streamOptimized");
5766	if (RT_FAILURE(rc))
5767	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
5768
5769	return rc;
5770	}
5771
5772	/**
5773	* Initializes the UUID fields in the DDB.
5774	*
5775	* @returns VBox status code.
5776	* @param pImage The VMDK image instance.
5777	*/
5778	static int vmdkCreateImageDdbUuidsInit(PVMDKIMAGE pImage)
5779	{
5780	int rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_IMAGE_UUID, &pImage->ImageUuid);
5781	if (RT_SUCCESS(rc))
5782	{
5783	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_PARENT_UUID, &pImage->ParentUuid);
5784	if (RT_SUCCESS(rc))
5785	{
5786	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_MODIFICATION_UUID,
5787	&pImage->ModificationUuid);
5788	if (RT_SUCCESS(rc))
5789	{
5790	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_PARENT_MODIFICATION_UUID,
5791	&pImage->ParentModificationUuid);
5792	if (RT_FAILURE(rc))
5793	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5794	N_("VMDK: error storing parent modification UUID in new descriptor in '%s'"), pImage->pszFilename);
5795	}
5796	else
5797	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5798	N_("VMDK: error storing modification UUID in new descriptor in '%s'"), pImage->pszFilename);
5799	}
5800	else
5801	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5802	N_("VMDK: error storing parent image UUID in new descriptor in '%s'"), pImage->pszFilename);
5803	}
5804	else
5805	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5806	N_("VMDK: error storing image UUID in new descriptor in '%s'"), pImage->pszFilename);
5807
5808	return rc;
5809	}
5810
5811	/**
5812	* Internal: The actual code for creating any VMDK variant currently in
5813	* existence on hosted environments.
5814	*/
5815	static int vmdkCreateImage(PVMDKIMAGE pImage, uint64_t cbSize,
5816	unsigned uImageFlags, const char *pszComment,
5817	PCVDGEOMETRY pPCHSGeometry,
5818	PCVDGEOMETRY pLCHSGeometry, PCRTUUID pUuid,
5819	PVDINTERFACEPROGRESS pIfProgress,
5820	unsigned uPercentStart, unsigned uPercentSpan)
5821	{
5822	pImage->uImageFlags = uImageFlags;
5823
5824	pImage->pIfError = VDIfErrorGet(pImage->pVDIfsDisk);
5825	pImage->pIfIo = VDIfIoIntGet(pImage->pVDIfsImage);
5826	AssertPtrReturn(pImage->pIfIo, VERR_INVALID_PARAMETER);
5827
5828	int rc = vmdkCreateDescriptor(pImage, pImage->pDescData, pImage->cbDescAlloc,
5829	&pImage->Descriptor);
5830	if (RT_SUCCESS(rc))
5831	{
5832	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
5833	{
5834	/* Raw disk image (includes raw partition). */
5835	PVDISKRAW pRaw = NULL;
5836	rc = vmdkMakeRawDescriptor(pImage, &pRaw);
5837	if (RT_FAILURE(rc))
5838	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create raw descriptor for '%s'"),
5839	pImage->pszFilename);
5840	if (!cbSize)
5841	cbSize = pImage->cbSize;
5842
5843	rc = vmdkCreateRawImage(pImage, pRaw, cbSize);
5844	vmdkRawDescFree(pRaw);
5845	}
5846	else if (uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5847	{
5848	/* Stream optimized sparse image (monolithic). */
5849	rc = vmdkCreateStreamImage(pImage, cbSize);
5850	}
5851	else
5852	{
5853	/* Regular fixed or sparse image (monolithic or split). */
5854	rc = vmdkCreateRegularImage(pImage, cbSize, uImageFlags,
5855	pIfProgress, uPercentStart,
5856	uPercentSpan * 95 / 100);
5857	}
5858
5859	if (RT_SUCCESS(rc))
5860	{
5861	vdIfProgress(pIfProgress, uPercentStart + uPercentSpan * 98 / 100);
5862
5863	pImage->cbSize = cbSize;
5864
5865	for (unsigned i = 0; i < pImage->cExtents; i++)
5866	{
5867	PVMDKEXTENT pExtent = &pImage->pExtents[i];
5868
5869	rc = vmdkDescExtInsert(pImage, &pImage->Descriptor, pExtent->enmAccess,
5870	pExtent->cNominalSectors, pExtent->enmType,
5871	pExtent->pszBasename, pExtent->uSectorOffset);
5872	if (RT_FAILURE(rc))
5873	{
5874	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not insert the extent list into descriptor in '%s'"), pImage->pszFilename);
5875	break;
5876	}
5877	}
5878
5879	if (RT_SUCCESS(rc))
5880	vmdkDescExtRemoveDummy(pImage, &pImage->Descriptor);
5881
5882	pImage->LCHSGeometry = *pLCHSGeometry;
5883	pImage->PCHSGeometry = *pPCHSGeometry;
5884
5885	if (RT_SUCCESS(rc))
5886	{
5887	if ( pPCHSGeometry->cCylinders != 0
5888	&& pPCHSGeometry->cHeads != 0
5889	&& pPCHSGeometry->cSectors != 0)
5890	rc = vmdkDescSetPCHSGeometry(pImage, pPCHSGeometry);
5891	else if (uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
5892	{
5893	VDGEOMETRY RawDiskPCHSGeometry;
5894	RawDiskPCHSGeometry.cCylinders = (uint32_t)RT_MIN(pImage->cbSize / 512 / 16 / 63, 16383);
5895	RawDiskPCHSGeometry.cHeads = 16;
5896	RawDiskPCHSGeometry.cSectors = 63;
5897	rc = vmdkDescSetPCHSGeometry(pImage, &RawDiskPCHSGeometry);
5898	}
5899	}
5900
5901	if ( RT_SUCCESS(rc)
5902	&& pLCHSGeometry->cCylinders != 0
5903	&& pLCHSGeometry->cHeads != 0
5904	&& pLCHSGeometry->cSectors != 0)
5905	rc = vmdkDescSetLCHSGeometry(pImage, pLCHSGeometry);
5906
5907	pImage->ImageUuid = *pUuid;
5908	RTUuidClear(&pImage->ParentUuid);
5909	RTUuidClear(&pImage->ModificationUuid);
5910	RTUuidClear(&pImage->ParentModificationUuid);
5911
5912	if (RT_SUCCESS(rc))
5913	rc = vmdkCreateImageDdbUuidsInit(pImage);
5914
5915	if (RT_SUCCESS(rc))
5916	rc = vmdkAllocateGrainTableCache(pImage);
5917
5918	if (RT_SUCCESS(rc))
5919	{
5920	rc = vmdkSetImageComment(pImage, pszComment);
5921	if (RT_FAILURE(rc))
5922	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot set image comment in '%s'"), pImage->pszFilename);
5923	}
5924
5925	if (RT_SUCCESS(rc))
5926	{
5927	vdIfProgress(pIfProgress, uPercentStart + uPercentSpan * 99 / 100);
5928
5929	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5930	{
5931	/* streamOptimized is a bit special, we cannot trigger the flush
5932	* until all data has been written. So we write the necessary
5933	* information explicitly. */
5934	pImage->pExtents[0].cDescriptorSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64( pImage->Descriptor.aLines[pImage->Descriptor.cLines]
5935	- pImage->Descriptor.aLines[0], 512));
5936	rc = vmdkWriteMetaSparseExtent(pImage, &pImage->pExtents[0], 0, NULL);
5937	if (RT_SUCCESS(rc))
5938	{
5939	rc = vmdkWriteDescriptor(pImage, NULL);
5940	if (RT_FAILURE(rc))
5941	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write VMDK descriptor in '%s'"), pImage->pszFilename);
5942	}
5943	else
5944	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write VMDK header in '%s'"), pImage->pszFilename);
5945	}
5946	else
5947	rc = vmdkFlushImage(pImage, NULL);
5948	}
5949	}
5950	}
5951	else
5952	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new descriptor in '%s'"), pImage->pszFilename);
5953
5954
5955	if (RT_SUCCESS(rc))
5956	{
5957	PVDREGIONDESC pRegion = &pImage->RegionList.aRegions[0];
5958	pImage->RegionList.fFlags = 0;
5959	pImage->RegionList.cRegions = 1;
5960
5961	pRegion->offRegion = 0; /* Disk start. */
5962	pRegion->cbBlock = 512;
5963	pRegion->enmDataForm = VDREGIONDATAFORM_RAW;
5964	pRegion->enmMetadataForm = VDREGIONMETADATAFORM_NONE;
5965	pRegion->cbData = 512;
5966	pRegion->cbMetadata = 0;
5967	pRegion->cRegionBlocksOrBytes = pImage->cbSize;
5968
5969	vdIfProgress(pIfProgress, uPercentStart + uPercentSpan);
5970	}
5971	else
5972	vmdkFreeImage(pImage, rc != VERR_ALREADY_EXISTS, false /fFlush/);
5973	return rc;
5974	}
5975
5976	/**
5977	* Internal: Update image comment.
5978	*/
5979	static int vmdkSetImageComment(PVMDKIMAGE pImage, const char *pszComment)
5980	{
5981	char *pszCommentEncoded = NULL;
5982	if (pszComment)
5983	{
5984	pszCommentEncoded = vmdkEncodeString(pszComment);
5985	if (!pszCommentEncoded)
5986	return VERR_NO_MEMORY;
5987	}
5988
5989	int rc = vmdkDescDDBSetStr(pImage, &pImage->Descriptor,
5990	"ddb.comment", pszCommentEncoded);
5991	if (pszCommentEncoded)
5992	RTStrFree(pszCommentEncoded);
5993	if (RT_FAILURE(rc))
5994	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image comment in descriptor in '%s'"), pImage->pszFilename);
5995	return VINF_SUCCESS;
5996	}
5997
5998	/**
5999	* Internal. Clear the grain table buffer for real stream optimized writing.
6000	*/
6001	static void vmdkStreamClearGT(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
6002	{
6003	uint32_t cCacheLines = RT_ALIGN(pExtent->cGTEntries, VMDK_GT_CACHELINE_SIZE) / VMDK_GT_CACHELINE_SIZE;
6004	for (uint32_t i = 0; i < cCacheLines; i++)
6005	memset(&pImage->pGTCache->aGTCache[i].aGTData[0], '\0',
6006	VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t));
6007	}
6008
6009	/**
6010	* Internal. Flush the grain table buffer for real stream optimized writing.
6011	*/
6012	static int vmdkStreamFlushGT(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
6013	uint32_t uGDEntry)
6014	{
6015	int rc = VINF_SUCCESS;
6016	uint32_t cCacheLines = RT_ALIGN(pExtent->cGTEntries, VMDK_GT_CACHELINE_SIZE) / VMDK_GT_CACHELINE_SIZE;
6017
6018	/* VMware does not write out completely empty grain tables in the case
6019	* of streamOptimized images, which according to my interpretation of
6020	* the VMDK 1.1 spec is bending the rules. Since they do it and we can
6021	* handle it without problems do it the same way and save some bytes. */
6022	bool fAllZero = true;
6023	for (uint32_t i = 0; i < cCacheLines; i++)
6024	{
6025	/* Convert the grain table to little endian in place, as it will not
6026	* be used at all after this function has been called. */
6027	uint32_t *pGTTmp = &pImage->pGTCache->aGTCache[i].aGTData[0];
6028	for (uint32_t j = 0; j < VMDK_GT_CACHELINE_SIZE; j++, pGTTmp++)
6029	if (*pGTTmp)
6030	{
6031	fAllZero = false;
6032	break;
6033	}
6034	if (!fAllZero)
6035	break;
6036	}
6037	if (fAllZero)
6038	return VINF_SUCCESS;
6039
6040	uint64_t uFileOffset = pExtent->uAppendPosition;
6041	if (!uFileOffset)
6042	return VERR_INTERNAL_ERROR;
6043	/* Align to sector, as the previous write could have been any size. */
6044	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
6045
6046	/* Grain table marker. */
6047	uint8_t aMarker[512];
6048	PVMDKMARKER pMarker = (PVMDKMARKER)&aMarker[0];
6049	memset(pMarker, '\0', sizeof(aMarker));
6050	pMarker->uSector = RT_H2LE_U64(VMDK_BYTE2SECTOR((uint64_t)pExtent->cGTEntries * sizeof(uint32_t)));
6051	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_GT);
6052	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
6053	aMarker, sizeof(aMarker));
6054	AssertRC(rc);
6055	uFileOffset += 512;
6056
6057	if (!pExtent->pGD \|\| pExtent->pGD[uGDEntry])
6058	return VERR_INTERNAL_ERROR;
6059
6060	pExtent->pGD[uGDEntry] = VMDK_BYTE2SECTOR(uFileOffset);
6061
6062	for (uint32_t i = 0; i < cCacheLines; i++)
6063	{
6064	/* Convert the grain table to little endian in place, as it will not
6065	* be used at all after this function has been called. */
6066	uint32_t *pGTTmp = &pImage->pGTCache->aGTCache[i].aGTData[0];
6067	for (uint32_t j = 0; j < VMDK_GT_CACHELINE_SIZE; j++, pGTTmp++)
6068	pGTTmp = RT_H2LE_U32(pGTTmp);
6069
6070	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
6071	&pImage->pGTCache->aGTCache[i].aGTData[0],
6072	VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t));
6073	uFileOffset += VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t);
6074	if (RT_FAILURE(rc))
6075	break;
6076	}
6077	Assert(!(uFileOffset % 512));
6078	pExtent->uAppendPosition = RT_ALIGN_64(uFileOffset, 512);
6079	return rc;
6080	}
6081
6082	/**
6083	* Internal. Free all allocated space for representing an image, and optionally
6084	* delete the image from disk.
6085	*/
6086	static int vmdkFreeImage(PVMDKIMAGE pImage, bool fDelete, bool fFlush)
6087	{
6088	int rc = VINF_SUCCESS;
6089
6090	/* Freeing a never allocated image (e.g. because the open failed) is
6091	* not signalled as an error. After all nothing bad happens. */
6092	if (pImage)
6093	{
6094	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
6095	{
6096	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6097	{
6098	/* Check if all extents are clean. */
6099	for (unsigned i = 0; i < pImage->cExtents; i++)
6100	{
6101	Assert(!pImage->pExtents[i].fUncleanShutdown);
6102	}
6103	}
6104	else
6105	{
6106	/* Mark all extents as clean. */
6107	for (unsigned i = 0; i < pImage->cExtents; i++)
6108	{
6109	if ( pImage->pExtents[i].enmType == VMDKETYPE_HOSTED_SPARSE
6110	&& pImage->pExtents[i].fUncleanShutdown)
6111	{
6112	pImage->pExtents[i].fUncleanShutdown = false;
6113	pImage->pExtents[i].fMetaDirty = true;
6114	}
6115
6116	/* From now on it's not safe to append any more data. */
6117	pImage->pExtents[i].uAppendPosition = 0;
6118	}
6119	}
6120	}
6121
6122	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6123	{
6124	/* No need to write any pending data if the file will be deleted
6125	* or if the new file wasn't successfully created. */
6126	if ( !fDelete && pImage->pExtents
6127	&& pImage->pExtents[0].cGTEntries
6128	&& pImage->pExtents[0].uAppendPosition)
6129	{
6130	PVMDKEXTENT pExtent = &pImage->pExtents[0];
6131	uint32_t uLastGDEntry = pExtent->uLastGrainAccess / pExtent->cGTEntries;
6132	rc = vmdkStreamFlushGT(pImage, pExtent, uLastGDEntry);
6133	AssertRC(rc);
6134	vmdkStreamClearGT(pImage, pExtent);
6135	for (uint32_t i = uLastGDEntry + 1; i < pExtent->cGDEntries; i++)
6136	{
6137	rc = vmdkStreamFlushGT(pImage, pExtent, i);
6138	AssertRC(rc);
6139	}
6140
6141	uint64_t uFileOffset = pExtent->uAppendPosition;
6142	if (!uFileOffset)
6143	return VERR_INTERNAL_ERROR;
6144	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
6145
6146	/* From now on it's not safe to append any more data. */
6147	pExtent->uAppendPosition = 0;
6148
6149	/* Grain directory marker. */
6150	uint8_t aMarker[512];
6151	PVMDKMARKER pMarker = (PVMDKMARKER)&aMarker[0];
6152	memset(pMarker, '\0', sizeof(aMarker));
6153	pMarker->uSector = VMDK_BYTE2SECTOR(RT_ALIGN_64(RT_H2LE_U64((uint64_t)pExtent->cGDEntries * sizeof(uint32_t)), 512));
6154	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_GD);
6155	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
6156	aMarker, sizeof(aMarker));
6157	AssertRC(rc);
6158	uFileOffset += 512;
6159
6160	/* Write grain directory in little endian style. The array will
6161	* not be used after this, so convert in place. */
6162	uint32_t *pGDTmp = pExtent->pGD;
6163	for (uint32_t i = 0; i < pExtent->cGDEntries; i++, pGDTmp++)
6164	pGDTmp = RT_H2LE_U32(pGDTmp);
6165	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
6166	uFileOffset, pExtent->pGD,
6167	pExtent->cGDEntries * sizeof(uint32_t));
6168	AssertRC(rc);
6169
6170	pExtent->uSectorGD = VMDK_BYTE2SECTOR(uFileOffset);
6171	pExtent->uSectorRGD = VMDK_BYTE2SECTOR(uFileOffset);
6172	uFileOffset = RT_ALIGN_64( uFileOffset
6173	+ pExtent->cGDEntries * sizeof(uint32_t),
6174	512);
6175
6176	/* Footer marker. */
6177	memset(pMarker, '\0', sizeof(aMarker));
6178	pMarker->uSector = VMDK_BYTE2SECTOR(512);
6179	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_FOOTER);
6180	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
6181	uFileOffset, aMarker, sizeof(aMarker));
6182	AssertRC(rc);
6183
6184	uFileOffset += 512;
6185	rc = vmdkWriteMetaSparseExtent(pImage, pExtent, uFileOffset, NULL);
6186	AssertRC(rc);
6187
6188	uFileOffset += 512;
6189	/* End-of-stream marker. */
6190	memset(pMarker, '\0', sizeof(aMarker));
6191	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
6192	uFileOffset, aMarker, sizeof(aMarker));
6193	AssertRC(rc);
6194	}
6195	}
6196	else if (!fDelete && fFlush)
6197	vmdkFlushImage(pImage, NULL);
6198
6199	if (pImage->pExtents != NULL)
6200	{
6201	for (unsigned i = 0 ; i < pImage->cExtents; i++)
6202	{
6203	int rc2 = vmdkFreeExtentData(pImage, &pImage->pExtents[i], fDelete);
6204	if (RT_SUCCESS(rc))
6205	rc = rc2; /* Propogate any error when closing the file. */
6206	}
6207	RTMemFree(pImage->pExtents);
6208	pImage->pExtents = NULL;
6209	}
6210	pImage->cExtents = 0;
6211	if (pImage->pFile != NULL)
6212	{
6213	int rc2 = vmdkFileClose(pImage, &pImage->pFile, fDelete);
6214	if (RT_SUCCESS(rc))
6215	rc = rc2; /* Propogate any error when closing the file. */
6216	}
6217	int rc2 = vmdkFileCheckAllClose(pImage);
6218	if (RT_SUCCESS(rc))
6219	rc = rc2; /* Propogate any error when closing the file. */
6220
6221	if (pImage->pGTCache)
6222	{
6223	RTMemFree(pImage->pGTCache);
6224	pImage->pGTCache = NULL;
6225	}
6226	if (pImage->pDescData)
6227	{
6228	RTMemFree(pImage->pDescData);
6229	pImage->pDescData = NULL;
6230	}
6231	}
6232
6233	LogFlowFunc(("returns %Rrc\n", rc));
6234	return rc;
6235	}
6236
6237	/**
6238	* Internal. Flush image data (and metadata) to disk.
6239	*/
6240	static int vmdkFlushImage(PVMDKIMAGE pImage, PVDIOCTX pIoCtx)
6241	{
6242	PVMDKEXTENT pExtent;
6243	int rc = VINF_SUCCESS;
6244
6245	/* Update descriptor if changed. */
6246	if (pImage->Descriptor.fDirty)
6247	rc = vmdkWriteDescriptor(pImage, pIoCtx);
6248
6249	if (RT_SUCCESS(rc))
6250	{
6251	for (unsigned i = 0; i < pImage->cExtents; i++)
6252	{
6253	pExtent = &pImage->pExtents[i];
6254	if (pExtent->pFile != NULL && pExtent->fMetaDirty)
6255	{
6256	switch (pExtent->enmType)
6257	{
6258	case VMDKETYPE_HOSTED_SPARSE:
6259	if (!pExtent->fFooter)
6260	rc = vmdkWriteMetaSparseExtent(pImage, pExtent, 0, pIoCtx);
6261	else
6262	{
6263	uint64_t uFileOffset = pExtent->uAppendPosition;
6264	/* Simply skip writing anything if the streamOptimized
6265	* image hasn't been just created. */
6266	if (!uFileOffset)
6267	break;
6268	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
6269	rc = vmdkWriteMetaSparseExtent(pImage, pExtent,
6270	uFileOffset, pIoCtx);
6271	}
6272	break;
6273	case VMDKETYPE_VMFS:
6274	case VMDKETYPE_FLAT:
6275	/* Nothing to do. */
6276	break;
6277	case VMDKETYPE_ZERO:
6278	default:
6279	AssertMsgFailed(("extent with type %d marked as dirty\n",
6280	pExtent->enmType));
6281	break;
6282	}
6283	}
6284
6285	if (RT_FAILURE(rc))
6286	break;
6287
6288	switch (pExtent->enmType)
6289	{
6290	case VMDKETYPE_HOSTED_SPARSE:
6291	case VMDKETYPE_VMFS:
6292	case VMDKETYPE_FLAT:
6293	/** @todo implement proper path absolute check. */
6294	if ( pExtent->pFile != NULL
6295	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
6296	&& !(pExtent->pszBasename[0] == RTPATH_SLASH))
6297	rc = vdIfIoIntFileFlush(pImage->pIfIo, pExtent->pFile->pStorage, pIoCtx,
6298	NULL, NULL);
6299	break;
6300	case VMDKETYPE_ZERO:
6301	/* No need to do anything for this extent. */
6302	break;
6303	default:
6304	AssertMsgFailed(("unknown extent type %d\n", pExtent->enmType));
6305	break;
6306	}
6307	}
6308	}
6309
6310	return rc;
6311	}
6312
6313	/**
6314	* Internal. Find extent corresponding to the sector number in the disk.
6315	*/
6316	static int vmdkFindExtent(PVMDKIMAGE pImage, uint64_t offSector,
6317	PVMDKEXTENT ppExtent, uint64_t puSectorInExtent)
6318	{
6319	PVMDKEXTENT pExtent = NULL;
6320	int rc = VINF_SUCCESS;
6321
6322	for (unsigned i = 0; i < pImage->cExtents; i++)
6323	{
6324	if (offSector < pImage->pExtents[i].cNominalSectors)
6325	{
6326	pExtent = &pImage->pExtents[i];
6327	*puSectorInExtent = offSector + pImage->pExtents[i].uSectorOffset;
6328	break;
6329	}
6330	offSector -= pImage->pExtents[i].cNominalSectors;
6331	}
6332
6333	if (pExtent)
6334	*ppExtent = pExtent;
6335	else
6336	rc = VERR_IO_SECTOR_NOT_FOUND;
6337
6338	return rc;
6339	}
6340
6341	/**
6342	* Internal. Hash function for placing the grain table hash entries.
6343	*/
6344	static uint32_t vmdkGTCacheHash(PVMDKGTCACHE pCache, uint64_t uSector,
6345	unsigned uExtent)
6346	{
6347	/** @todo this hash function is quite simple, maybe use a better one which
6348	* scrambles the bits better. */
6349	return (uSector + uExtent) % pCache->cEntries;
6350	}
6351
6352	/**
6353	* Internal. Get sector number in the extent file from the relative sector
6354	* number in the extent.
6355	*/
6356	static int vmdkGetSector(PVMDKIMAGE pImage, PVDIOCTX pIoCtx,
6357	PVMDKEXTENT pExtent, uint64_t uSector,
6358	uint64_t *puExtentSector)
6359	{
6360	PVMDKGTCACHE pCache = pImage->pGTCache;
6361	uint64_t uGDIndex, uGTSector, uGTBlock;
6362	uint32_t uGTHash, uGTBlockIndex;
6363	PVMDKGTCACHEENTRY pGTCacheEntry;
6364	uint32_t aGTDataTmp[VMDK_GT_CACHELINE_SIZE];
6365	int rc;
6366
6367	/* For newly created and readonly/sequentially opened streamOptimized
6368	* images this must be a no-op, as the grain directory is not there. */
6369	if ( ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
6370	&& pExtent->uAppendPosition)
6371	\|\| ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
6372	&& pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY
6373	&& pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
6374	{
6375	*puExtentSector = 0;
6376	return VINF_SUCCESS;
6377	}
6378
6379	uGDIndex = uSector / pExtent->cSectorsPerGDE;
6380	if (uGDIndex >= pExtent->cGDEntries)
6381	return VERR_OUT_OF_RANGE;
6382	uGTSector = pExtent->pGD[uGDIndex];
6383	if (!uGTSector)
6384	{
6385	/* There is no grain table referenced by this grain directory
6386	* entry. So there is absolutely no data in this area. */
6387	*puExtentSector = 0;
6388	return VINF_SUCCESS;
6389	}
6390
6391	uGTBlock = uSector / (pExtent->cSectorsPerGrain * VMDK_GT_CACHELINE_SIZE);
6392	uGTHash = vmdkGTCacheHash(pCache, uGTBlock, pExtent->uExtent);
6393	pGTCacheEntry = &pCache->aGTCache[uGTHash];
6394	if ( pGTCacheEntry->uExtent != pExtent->uExtent
6395	\|\| pGTCacheEntry->uGTBlock != uGTBlock)
6396	{
6397	/* Cache miss, fetch data from disk. */
6398	PVDMETAXFER pMetaXfer;
6399	rc = vdIfIoIntFileReadMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6400	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
6401	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx, &pMetaXfer, NULL, NULL);
6402	if (RT_FAILURE(rc))
6403	return rc;
6404	/* We can release the metadata transfer immediately. */
6405	vdIfIoIntMetaXferRelease(pImage->pIfIo, pMetaXfer);
6406	pGTCacheEntry->uExtent = pExtent->uExtent;
6407	pGTCacheEntry->uGTBlock = uGTBlock;
6408	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
6409	pGTCacheEntry->aGTData[i] = RT_LE2H_U32(aGTDataTmp[i]);
6410	}
6411	uGTBlockIndex = (uSector / pExtent->cSectorsPerGrain) % VMDK_GT_CACHELINE_SIZE;
6412	uint32_t uGrainSector = pGTCacheEntry->aGTData[uGTBlockIndex];
6413	if (uGrainSector)
6414	*puExtentSector = uGrainSector + uSector % pExtent->cSectorsPerGrain;
6415	else
6416	*puExtentSector = 0;
6417	return VINF_SUCCESS;
6418	}
6419
6420	/**
6421	* Internal. Writes the grain and also if necessary the grain tables.
6422	* Uses the grain table cache as a true grain table.
6423	*/
6424	static int vmdkStreamAllocGrain(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
6425	uint64_t uSector, PVDIOCTX pIoCtx,
6426	uint64_t cbWrite)
6427	{
6428	uint32_t uGrain;
6429	uint32_t uGDEntry, uLastGDEntry;
6430	uint32_t cbGrain = 0;
6431	uint32_t uCacheLine, uCacheEntry;
6432	const void *pData;
6433	int rc;
6434
6435	/* Very strict requirements: always write at least one full grain, with
6436	* proper alignment. Everything else would require reading of already
6437	* written data, which we don't support for obvious reasons. The only
6438	* exception is the last grain, and only if the image size specifies
6439	* that only some portion holds data. In any case the write must be
6440	* within the image limits, no "overshoot" allowed. */
6441	if ( cbWrite == 0
6442	\|\| ( cbWrite < VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain)
6443	&& pExtent->cNominalSectors - uSector >= pExtent->cSectorsPerGrain)
6444	\|\| uSector % pExtent->cSectorsPerGrain
6445	\|\| uSector + VMDK_BYTE2SECTOR(cbWrite) > pExtent->cNominalSectors)
6446	return VERR_INVALID_PARAMETER;
6447
6448	/* Clip write range to at most the rest of the grain. */
6449	cbWrite = RT_MIN(cbWrite, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain - uSector % pExtent->cSectorsPerGrain));
6450
6451	/* Do not allow to go back. */
6452	uGrain = uSector / pExtent->cSectorsPerGrain;
6453	uCacheLine = uGrain % pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE;
6454	uCacheEntry = uGrain % VMDK_GT_CACHELINE_SIZE;
6455	uGDEntry = uGrain / pExtent->cGTEntries;
6456	uLastGDEntry = pExtent->uLastGrainAccess / pExtent->cGTEntries;
6457	if (uGrain < pExtent->uLastGrainAccess)
6458	return VERR_VD_VMDK_INVALID_WRITE;
6459
6460	/* Zero byte write optimization. Since we don't tell VBoxHDD that we need
6461	* to allocate something, we also need to detect the situation ourself. */
6462	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_HONOR_ZEROES)
6463	&& vdIfIoIntIoCtxIsZero(pImage->pIfIo, pIoCtx, cbWrite, true /* fAdvance */))
6464	return VINF_SUCCESS;
6465
6466	if (uGDEntry != uLastGDEntry)
6467	{
6468	rc = vmdkStreamFlushGT(pImage, pExtent, uLastGDEntry);
6469	if (RT_FAILURE(rc))
6470	return rc;
6471	vmdkStreamClearGT(pImage, pExtent);
6472	for (uint32_t i = uLastGDEntry + 1; i < uGDEntry; i++)
6473	{
6474	rc = vmdkStreamFlushGT(pImage, pExtent, i);
6475	if (RT_FAILURE(rc))
6476	return rc;
6477	}
6478	}
6479
6480	uint64_t uFileOffset;
6481	uFileOffset = pExtent->uAppendPosition;
6482	if (!uFileOffset)
6483	return VERR_INTERNAL_ERROR;
6484	/* Align to sector, as the previous write could have been any size. */
6485	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
6486
6487	/* Paranoia check: extent type, grain table buffer presence and
6488	* grain table buffer space. Also grain table entry must be clear. */
6489	if ( pExtent->enmType != VMDKETYPE_HOSTED_SPARSE
6490	\|\| !pImage->pGTCache
6491	\|\| pExtent->cGTEntries > VMDK_GT_CACHE_SIZE * VMDK_GT_CACHELINE_SIZE
6492	\|\| pImage->pGTCache->aGTCache[uCacheLine].aGTData[uCacheEntry])
6493	return VERR_INTERNAL_ERROR;
6494
6495	/* Update grain table entry. */
6496	pImage->pGTCache->aGTCache[uCacheLine].aGTData[uCacheEntry] = VMDK_BYTE2SECTOR(uFileOffset);
6497
6498	if (cbWrite != VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
6499	{
6500	vdIfIoIntIoCtxCopyFrom(pImage->pIfIo, pIoCtx, pExtent->pvGrain, cbWrite);
6501	memset((char *)pExtent->pvGrain + cbWrite, '\0',
6502	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain) - cbWrite);
6503	pData = pExtent->pvGrain;
6504	}
6505	else
6506	{
6507	RTSGSEG Segment;
6508	unsigned cSegments = 1;
6509	size_t cbSeg = 0;
6510
6511	cbSeg = vdIfIoIntIoCtxSegArrayCreate(pImage->pIfIo, pIoCtx, &Segment,
6512	&cSegments, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
6513	Assert(cbSeg == VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
6514	pData = Segment.pvSeg;
6515	}
6516	rc = vmdkFileDeflateSync(pImage, pExtent, uFileOffset, pData,
6517	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
6518	uSector, &cbGrain);
6519	if (RT_FAILURE(rc))
6520	{
6521	pExtent->uGrainSectorAbs = 0;
6522	AssertRC(rc);
6523	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write compressed data block in '%s'"), pExtent->pszFullname);
6524	}
6525	pExtent->uLastGrainAccess = uGrain;
6526	pExtent->uAppendPosition += cbGrain;
6527
6528	return rc;
6529	}
6530
6531	/**
6532	* Internal: Updates the grain table during grain allocation.
6533	*/
6534	static int vmdkAllocGrainGTUpdate(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, PVDIOCTX pIoCtx,
6535	PVMDKGRAINALLOCASYNC pGrainAlloc)
6536	{
6537	int rc = VINF_SUCCESS;
6538	PVMDKGTCACHE pCache = pImage->pGTCache;
6539	uint32_t aGTDataTmp[VMDK_GT_CACHELINE_SIZE];
6540	uint32_t uGTHash, uGTBlockIndex;
6541	uint64_t uGTSector, uRGTSector, uGTBlock;
6542	uint64_t uSector = pGrainAlloc->uSector;
6543	PVMDKGTCACHEENTRY pGTCacheEntry;
6544
6545	LogFlowFunc(("pImage=%#p pExtent=%#p pCache=%#p pIoCtx=%#p pGrainAlloc=%#p\n",
6546	pImage, pExtent, pCache, pIoCtx, pGrainAlloc));
6547
6548	uGTSector = pGrainAlloc->uGTSector;
6549	uRGTSector = pGrainAlloc->uRGTSector;
6550	LogFlow(("uGTSector=%llu uRGTSector=%llu\n", uGTSector, uRGTSector));
6551
6552	/* Update the grain table (and the cache). */
6553	uGTBlock = uSector / (pExtent->cSectorsPerGrain * VMDK_GT_CACHELINE_SIZE);
6554	uGTHash = vmdkGTCacheHash(pCache, uGTBlock, pExtent->uExtent);
6555	pGTCacheEntry = &pCache->aGTCache[uGTHash];
6556	if ( pGTCacheEntry->uExtent != pExtent->uExtent
6557	\|\| pGTCacheEntry->uGTBlock != uGTBlock)
6558	{
6559	/* Cache miss, fetch data from disk. */
6560	LogFlow(("Cache miss, fetch data from disk\n"));
6561	PVDMETAXFER pMetaXfer = NULL;
6562	rc = vdIfIoIntFileReadMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6563	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
6564	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
6565	&pMetaXfer, vmdkAllocGrainComplete, pGrainAlloc);
6566	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6567	{
6568	pGrainAlloc->cIoXfersPending++;
6569	pGrainAlloc->fGTUpdateNeeded = true;
6570	/* Leave early, we will be called again after the read completed. */
6571	LogFlowFunc(("Metadata read in progress, leaving\n"));
6572	return rc;
6573	}
6574	else if (RT_FAILURE(rc))
6575	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot read allocated grain table entry in '%s'"), pExtent->pszFullname);
6576	vdIfIoIntMetaXferRelease(pImage->pIfIo, pMetaXfer);
6577	pGTCacheEntry->uExtent = pExtent->uExtent;
6578	pGTCacheEntry->uGTBlock = uGTBlock;
6579	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
6580	pGTCacheEntry->aGTData[i] = RT_LE2H_U32(aGTDataTmp[i]);
6581	}
6582	else
6583	{
6584	/* Cache hit. Convert grain table block back to disk format, otherwise
6585	* the code below will write garbage for all but the updated entry. */
6586	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
6587	aGTDataTmp[i] = RT_H2LE_U32(pGTCacheEntry->aGTData[i]);
6588	}
6589	pGrainAlloc->fGTUpdateNeeded = false;
6590	uGTBlockIndex = (uSector / pExtent->cSectorsPerGrain) % VMDK_GT_CACHELINE_SIZE;
6591	aGTDataTmp[uGTBlockIndex] = RT_H2LE_U32(VMDK_BYTE2SECTOR(pGrainAlloc->uGrainOffset));
6592	pGTCacheEntry->aGTData[uGTBlockIndex] = VMDK_BYTE2SECTOR(pGrainAlloc->uGrainOffset);
6593	/* Update grain table on disk. */
6594	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6595	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
6596	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
6597	vmdkAllocGrainComplete, pGrainAlloc);
6598	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6599	pGrainAlloc->cIoXfersPending++;
6600	else if (RT_FAILURE(rc))
6601	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write updated grain table in '%s'"), pExtent->pszFullname);
6602	if (pExtent->pRGD)
6603	{
6604	/* Update backup grain table on disk. */
6605	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6606	VMDK_SECTOR2BYTE(uRGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
6607	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
6608	vmdkAllocGrainComplete, pGrainAlloc);
6609	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6610	pGrainAlloc->cIoXfersPending++;
6611	else if (RT_FAILURE(rc))
6612	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write updated backup grain table in '%s'"), pExtent->pszFullname);
6613	}
6614
6615	LogFlowFunc(("leaving rc=%Rrc\n", rc));
6616	return rc;
6617	}
6618
6619	/**
6620	* Internal - complete the grain allocation by updating disk grain table if required.
6621	*/
6622	static DECLCALLBACK(int) vmdkAllocGrainComplete(void pBackendData, PVDIOCTX pIoCtx, void pvUser, int rcReq)
6623	{
6624	RT_NOREF1(rcReq);
6625	int rc = VINF_SUCCESS;
6626	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6627	PVMDKGRAINALLOCASYNC pGrainAlloc = (PVMDKGRAINALLOCASYNC)pvUser;
6628
6629	LogFlowFunc(("pBackendData=%#p pIoCtx=%#p pvUser=%#p rcReq=%Rrc\n",
6630	pBackendData, pIoCtx, pvUser, rcReq));
6631
6632	pGrainAlloc->cIoXfersPending--;
6633	if (!pGrainAlloc->cIoXfersPending && pGrainAlloc->fGTUpdateNeeded)
6634	rc = vmdkAllocGrainGTUpdate(pImage, pGrainAlloc->pExtent, pIoCtx, pGrainAlloc);
6635
6636	if (!pGrainAlloc->cIoXfersPending)
6637	{
6638	/* Grain allocation completed. */
6639	RTMemFree(pGrainAlloc);
6640	}
6641
6642	LogFlowFunc(("Leaving rc=%Rrc\n", rc));
6643	return rc;
6644	}
6645
6646	/**
6647	* Internal. Allocates a new grain table (if necessary).
6648	*/
6649	static int vmdkAllocGrain(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, PVDIOCTX pIoCtx,
6650	uint64_t uSector, uint64_t cbWrite)
6651	{
6652	PVMDKGTCACHE pCache = pImage->pGTCache; NOREF(pCache);
6653	uint64_t uGDIndex, uGTSector, uRGTSector;
6654	uint64_t uFileOffset;
6655	PVMDKGRAINALLOCASYNC pGrainAlloc = NULL;
6656	int rc;
6657
6658	LogFlowFunc(("pCache=%#p pExtent=%#p pIoCtx=%#p uSector=%llu cbWrite=%llu\n",
6659	pCache, pExtent, pIoCtx, uSector, cbWrite));
6660
6661	pGrainAlloc = (PVMDKGRAINALLOCASYNC)RTMemAllocZ(sizeof(VMDKGRAINALLOCASYNC));
6662	if (!pGrainAlloc)
6663	return VERR_NO_MEMORY;
6664
6665	pGrainAlloc->pExtent = pExtent;
6666	pGrainAlloc->uSector = uSector;
6667
6668	uGDIndex = uSector / pExtent->cSectorsPerGDE;
6669	if (uGDIndex >= pExtent->cGDEntries)
6670	{
6671	RTMemFree(pGrainAlloc);
6672	return VERR_OUT_OF_RANGE;
6673	}
6674	uGTSector = pExtent->pGD[uGDIndex];
6675	if (pExtent->pRGD)
6676	uRGTSector = pExtent->pRGD[uGDIndex];
6677	else
6678	uRGTSector = 0; /*< avoid compiler warning /
6679	if (!uGTSector)
6680	{
6681	LogFlow(("Allocating new grain table\n"));
6682
6683	/* There is no grain table referenced by this grain directory
6684	* entry. So there is absolutely no data in this area. Allocate
6685	* a new grain table and put the reference to it in the GDs. */
6686	uFileOffset = pExtent->uAppendPosition;
6687	if (!uFileOffset)
6688	{
6689	RTMemFree(pGrainAlloc);
6690	return VERR_INTERNAL_ERROR;
6691	}
6692	Assert(!(uFileOffset % 512));
6693
6694	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
6695	uGTSector = VMDK_BYTE2SECTOR(uFileOffset);
6696
6697	/* Normally the grain table is preallocated for hosted sparse extents
6698	* that support more than 32 bit sector numbers. So this shouldn't
6699	* ever happen on a valid extent. */
6700	if (uGTSector > UINT32_MAX)
6701	{
6702	RTMemFree(pGrainAlloc);
6703	return VERR_VD_VMDK_INVALID_HEADER;
6704	}
6705
6706	/* Write grain table by writing the required number of grain table
6707	* cache chunks. Allocate memory dynamically here or we flood the
6708	* metadata cache with very small entries. */
6709	size_t cbGTDataTmp = pExtent->cGTEntries * sizeof(uint32_t);
6710	uint32_t paGTDataTmp = (uint32_t )RTMemTmpAllocZ(cbGTDataTmp);
6711
6712	if (!paGTDataTmp)
6713	{
6714	RTMemFree(pGrainAlloc);
6715	return VERR_NO_MEMORY;
6716	}
6717
6718	memset(paGTDataTmp, '\0', cbGTDataTmp);
6719	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6720	VMDK_SECTOR2BYTE(uGTSector),
6721	paGTDataTmp, cbGTDataTmp, pIoCtx,
6722	vmdkAllocGrainComplete, pGrainAlloc);
6723	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6724	pGrainAlloc->cIoXfersPending++;
6725	else if (RT_FAILURE(rc))
6726	{
6727	RTMemTmpFree(paGTDataTmp);
6728	RTMemFree(pGrainAlloc);
6729	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write grain table allocation in '%s'"), pExtent->pszFullname);
6730	}
6731	pExtent->uAppendPosition = RT_ALIGN_64( pExtent->uAppendPosition
6732	+ cbGTDataTmp, 512);
6733
6734	if (pExtent->pRGD)
6735	{
6736	AssertReturn(!uRGTSector, VERR_VD_VMDK_INVALID_HEADER);
6737	uFileOffset = pExtent->uAppendPosition;
6738	if (!uFileOffset)
6739	return VERR_INTERNAL_ERROR;
6740	Assert(!(uFileOffset % 512));
6741	uRGTSector = VMDK_BYTE2SECTOR(uFileOffset);
6742
6743	/* Normally the redundant grain table is preallocated for hosted
6744	* sparse extents that support more than 32 bit sector numbers. So
6745	* this shouldn't ever happen on a valid extent. */
6746	if (uRGTSector > UINT32_MAX)
6747	{
6748	RTMemTmpFree(paGTDataTmp);
6749	return VERR_VD_VMDK_INVALID_HEADER;
6750	}
6751
6752	/* Write grain table by writing the required number of grain table
6753	* cache chunks. Allocate memory dynamically here or we flood the
6754	* metadata cache with very small entries. */
6755	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6756	VMDK_SECTOR2BYTE(uRGTSector),
6757	paGTDataTmp, cbGTDataTmp, pIoCtx,
6758	vmdkAllocGrainComplete, pGrainAlloc);
6759	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6760	pGrainAlloc->cIoXfersPending++;
6761	else if (RT_FAILURE(rc))
6762	{
6763	RTMemTmpFree(paGTDataTmp);
6764	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write backup grain table allocation in '%s'"), pExtent->pszFullname);
6765	}
6766
6767	pExtent->uAppendPosition = pExtent->uAppendPosition + cbGTDataTmp;
6768	}
6769
6770	RTMemTmpFree(paGTDataTmp);
6771
6772	/* Update the grain directory on disk (doing it before writing the
6773	* grain table will result in a garbled extent if the operation is
6774	* aborted for some reason. Otherwise the worst that can happen is
6775	* some unused sectors in the extent. */
6776	uint32_t uGTSectorLE = RT_H2LE_U64(uGTSector);
6777	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6778	VMDK_SECTOR2BYTE(pExtent->uSectorGD) + uGDIndex * sizeof(uGTSectorLE),
6779	&uGTSectorLE, sizeof(uGTSectorLE), pIoCtx,
6780	vmdkAllocGrainComplete, pGrainAlloc);
6781	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6782	pGrainAlloc->cIoXfersPending++;
6783	else if (RT_FAILURE(rc))
6784	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write grain directory entry in '%s'"), pExtent->pszFullname);
6785	if (pExtent->pRGD)
6786	{
6787	uint32_t uRGTSectorLE = RT_H2LE_U64(uRGTSector);
6788	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6789	VMDK_SECTOR2BYTE(pExtent->uSectorRGD) + uGDIndex * sizeof(uGTSectorLE),
6790	&uRGTSectorLE, sizeof(uRGTSectorLE), pIoCtx,
6791	vmdkAllocGrainComplete, pGrainAlloc);
6792	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6793	pGrainAlloc->cIoXfersPending++;
6794	else if (RT_FAILURE(rc))
6795	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write backup grain directory entry in '%s'"), pExtent->pszFullname);
6796	}
6797
6798	/* As the final step update the in-memory copy of the GDs. */
6799	pExtent->pGD[uGDIndex] = uGTSector;
6800	if (pExtent->pRGD)
6801	pExtent->pRGD[uGDIndex] = uRGTSector;
6802	}
6803
6804	LogFlow(("uGTSector=%llu uRGTSector=%llu\n", uGTSector, uRGTSector));
6805	pGrainAlloc->uGTSector = uGTSector;
6806	pGrainAlloc->uRGTSector = uRGTSector;
6807
6808	uFileOffset = pExtent->uAppendPosition;
6809	if (!uFileOffset)
6810	return VERR_INTERNAL_ERROR;
6811	Assert(!(uFileOffset % 512));
6812
6813	pGrainAlloc->uGrainOffset = uFileOffset;
6814
6815	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6816	{
6817	AssertMsgReturn(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
6818	("Accesses to stream optimized images must be synchronous\n"),
6819	VERR_INVALID_STATE);
6820
6821	if (cbWrite != VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
6822	return vdIfError(pImage->pIfError, VERR_INTERNAL_ERROR, RT_SRC_POS, N_("VMDK: not enough data for a compressed data block in '%s'"), pExtent->pszFullname);
6823
6824	/* Invalidate cache, just in case some code incorrectly allows mixing
6825	* of reads and writes. Normally shouldn't be needed. */
6826	pExtent->uGrainSectorAbs = 0;
6827
6828	/* Write compressed data block and the markers. */
6829	uint32_t cbGrain = 0;
6830	size_t cbSeg = 0;
6831	RTSGSEG Segment;
6832	unsigned cSegments = 1;
6833
6834	cbSeg = vdIfIoIntIoCtxSegArrayCreate(pImage->pIfIo, pIoCtx, &Segment,
6835	&cSegments, cbWrite);
6836	Assert(cbSeg == cbWrite);
6837
6838	rc = vmdkFileDeflateSync(pImage, pExtent, uFileOffset,
6839	Segment.pvSeg, cbWrite, uSector, &cbGrain);
6840	if (RT_FAILURE(rc))
6841	{
6842	AssertRC(rc);
6843	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write allocated compressed data block in '%s'"), pExtent->pszFullname);
6844	}
6845	pExtent->uLastGrainAccess = uSector / pExtent->cSectorsPerGrain;
6846	pExtent->uAppendPosition += cbGrain;
6847	}
6848	else
6849	{
6850	/* Write the data. Always a full grain, or we're in big trouble. */
6851	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
6852	uFileOffset, pIoCtx, cbWrite,
6853	vmdkAllocGrainComplete, pGrainAlloc);
6854	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6855	pGrainAlloc->cIoXfersPending++;
6856	else if (RT_FAILURE(rc))
6857	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write allocated data block in '%s'"), pExtent->pszFullname);
6858
6859	pExtent->uAppendPosition += cbWrite;
6860	}
6861
6862	rc = vmdkAllocGrainGTUpdate(pImage, pExtent, pIoCtx, pGrainAlloc);
6863
6864	if (!pGrainAlloc->cIoXfersPending)
6865	{
6866	/* Grain allocation completed. */
6867	RTMemFree(pGrainAlloc);
6868	}
6869
6870	LogFlowFunc(("leaving rc=%Rrc\n", rc));
6871
6872	return rc;
6873	}
6874
6875	/**
6876	* Internal. Reads the contents by sequentially going over the compressed
6877	* grains (hoping that they are in sequence).
6878	*/
6879	static int vmdkStreamReadSequential(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
6880	uint64_t uSector, PVDIOCTX pIoCtx,
6881	uint64_t cbRead)
6882	{
6883	int rc;
6884
6885	LogFlowFunc(("pImage=%#p pExtent=%#p uSector=%llu pIoCtx=%#p cbRead=%llu\n",
6886	pImage, pExtent, uSector, pIoCtx, cbRead));
6887
6888	AssertMsgReturn(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
6889	("Async I/O not supported for sequential stream optimized images\n"),
6890	VERR_INVALID_STATE);
6891
6892	/* Do not allow to go back. */
6893	uint32_t uGrain = uSector / pExtent->cSectorsPerGrain;
6894	if (uGrain < pExtent->uLastGrainAccess)
6895	return VERR_VD_VMDK_INVALID_STATE;
6896	pExtent->uLastGrainAccess = uGrain;
6897
6898	/* After a previous error do not attempt to recover, as it would need
6899	* seeking (in the general case backwards which is forbidden). */
6900	if (!pExtent->uGrainSectorAbs)
6901	return VERR_VD_VMDK_INVALID_STATE;
6902
6903	/* Check if we need to read something from the image or if what we have
6904	* in the buffer is good to fulfill the request. */
6905	if (!pExtent->cbGrainStreamRead \|\| uGrain > pExtent->uGrain)
6906	{
6907	uint32_t uGrainSectorAbs = pExtent->uGrainSectorAbs
6908	+ VMDK_BYTE2SECTOR(pExtent->cbGrainStreamRead);
6909
6910	/* Get the marker from the next data block - and skip everything which
6911	* is not a compressed grain. If it's a compressed grain which is for
6912	* the requested sector (or after), read it. */
6913	VMDKMARKER Marker;
6914	do
6915	{
6916	RT_ZERO(Marker);
6917	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
6918	VMDK_SECTOR2BYTE(uGrainSectorAbs),
6919	&Marker, RT_UOFFSETOF(VMDKMARKER, uType));
6920	if (RT_FAILURE(rc))
6921	return rc;
6922	Marker.uSector = RT_LE2H_U64(Marker.uSector);
6923	Marker.cbSize = RT_LE2H_U32(Marker.cbSize);
6924
6925	if (Marker.cbSize == 0)
6926	{
6927	/* A marker for something else than a compressed grain. */
6928	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
6929	VMDK_SECTOR2BYTE(uGrainSectorAbs)
6930	+ RT_UOFFSETOF(VMDKMARKER, uType),
6931	&Marker.uType, sizeof(Marker.uType));
6932	if (RT_FAILURE(rc))
6933	return rc;
6934	Marker.uType = RT_LE2H_U32(Marker.uType);
6935	switch (Marker.uType)
6936	{
6937	case VMDK_MARKER_EOS:
6938	uGrainSectorAbs++;
6939	/* Read (or mostly skip) to the end of file. Uses the
6940	* Marker (LBA sector) as it is unused anyway. This
6941	* makes sure that really everything is read in the
6942	* success case. If this read fails it means the image
6943	* is truncated, but this is harmless so ignore. */
6944	vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
6945	VMDK_SECTOR2BYTE(uGrainSectorAbs)
6946	+ 511,
6947	&Marker.uSector, 1);
6948	break;
6949	case VMDK_MARKER_GT:
6950	uGrainSectorAbs += 1 + VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
6951	break;
6952	case VMDK_MARKER_GD:
6953	uGrainSectorAbs += 1 + VMDK_BYTE2SECTOR(RT_ALIGN(pExtent->cGDEntries * sizeof(uint32_t), 512));
6954	break;
6955	case VMDK_MARKER_FOOTER:
6956	uGrainSectorAbs += 2;
6957	break;
6958	case VMDK_MARKER_UNSPECIFIED:
6959	/* Skip over the contents of the unspecified marker
6960	* type 4 which exists in some vSphere created files. */
6961	/** @todo figure out what the payload means. */
6962	uGrainSectorAbs += 1;
6963	break;
6964	default:
6965	AssertMsgFailed(("VMDK: corrupted marker, type=%#x\n", Marker.uType));
6966	pExtent->uGrainSectorAbs = 0;
6967	return VERR_VD_VMDK_INVALID_STATE;
6968	}
6969	pExtent->cbGrainStreamRead = 0;
6970	}
6971	else
6972	{
6973	/* A compressed grain marker. If it is at/after what we're
6974	* interested in read and decompress data. */
6975	if (uSector > Marker.uSector + pExtent->cSectorsPerGrain)
6976	{
6977	uGrainSectorAbs += VMDK_BYTE2SECTOR(RT_ALIGN(Marker.cbSize + RT_UOFFSETOF(VMDKMARKER, uType), 512));
6978	continue;
6979	}
6980	uint64_t uLBA = 0;
6981	uint32_t cbGrainStreamRead = 0;
6982	rc = vmdkFileInflateSync(pImage, pExtent,
6983	VMDK_SECTOR2BYTE(uGrainSectorAbs),
6984	pExtent->pvGrain,
6985	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
6986	&Marker, &uLBA, &cbGrainStreamRead);
6987	if (RT_FAILURE(rc))
6988	{
6989	pExtent->uGrainSectorAbs = 0;
6990	return rc;
6991	}
6992	if ( pExtent->uGrain
6993	&& uLBA / pExtent->cSectorsPerGrain <= pExtent->uGrain)
6994	{
6995	pExtent->uGrainSectorAbs = 0;
6996	return VERR_VD_VMDK_INVALID_STATE;
6997	}
6998	pExtent->uGrain = uLBA / pExtent->cSectorsPerGrain;
6999	pExtent->cbGrainStreamRead = cbGrainStreamRead;
7000	break;
7001	}
7002	} while (Marker.uType != VMDK_MARKER_EOS);
7003
7004	pExtent->uGrainSectorAbs = uGrainSectorAbs;
7005
7006	if (!pExtent->cbGrainStreamRead && Marker.uType == VMDK_MARKER_EOS)
7007	{
7008	pExtent->uGrain = UINT32_MAX;
7009	/* Must set a non-zero value for pExtent->cbGrainStreamRead or
7010	* the next read would try to get more data, and we're at EOF. */
7011	pExtent->cbGrainStreamRead = 1;
7012	}
7013	}
7014
7015	if (pExtent->uGrain > uSector / pExtent->cSectorsPerGrain)
7016	{
7017	/* The next data block we have is not for this area, so just return
7018	* that there is no data. */
7019	LogFlowFunc(("returns VERR_VD_BLOCK_FREE\n"));
7020	return VERR_VD_BLOCK_FREE;
7021	}
7022
7023	uint32_t uSectorInGrain = uSector % pExtent->cSectorsPerGrain;
7024	vdIfIoIntIoCtxCopyTo(pImage->pIfIo, pIoCtx,
7025	(uint8_t *)pExtent->pvGrain + VMDK_SECTOR2BYTE(uSectorInGrain),
7026	cbRead);
7027	LogFlowFunc(("returns VINF_SUCCESS\n"));
7028	return VINF_SUCCESS;
7029	}
7030
7031	/**
7032	* Replaces a fragment of a string with the specified string.
7033	*
7034	* @returns Pointer to the allocated UTF-8 string.
7035	* @param pszWhere UTF-8 string to search in.
7036	* @param pszWhat UTF-8 string to search for.
7037	* @param pszByWhat UTF-8 string to replace the found string with.
7038	*
7039	* @note r=bird: This is only used by vmdkRenameWorker(). The first use is
7040	* for updating the base name in the descriptor, the second is for
7041	* generating new filenames for extents. This code borked when
7042	* RTPathAbs started correcting the driver letter case on windows,
7043	* when strstr failed because the pExtent->pszFullname was not
7044	* subjected to RTPathAbs but while pExtent->pszFullname was. I fixed
7045	* this by apply RTPathAbs to the places it wasn't applied.
7046	*
7047	* However, this highlights some undocumented ASSUMPTIONS as well as
7048	* terrible short commings of the approach.
7049	*
7050	* Given the right filename, it may also screw up the descriptor. Take
7051	* the descriptor text 'RW 2048 SPARSE "Test0.vmdk"' for instance,
7052	* we'll be asked to replace "Test0" with something, no problem. No,
7053	* imagine 'RW 2048 SPARSE "SPARSE.vmdk"', 'RW 2048 SPARSE "RW.vmdk"'
7054	* or 'RW 2048 SPARSE "2048.vmdk"', and the strstr approach falls on
7055	* its bum. The descriptor string must be parsed and reconstructed,
7056	* the lazy strstr approach doesn't cut it.
7057	*
7058	* I'm also curious as to what would be the correct escaping of '"' in
7059	* the file name and how that is supposed to be handled, because it
7060	* needs to be or such names must be rejected in several places (maybe
7061	* they are, I didn't check).
7062	*
7063	* When this function is used to replace the start of a path, I think
7064	* the assumption from the prep/setup code is that we kind of knows
7065	* what we're working on (I could be wrong). However, using strstr
7066	* instead of strncmp/RTStrNICmp makes no sense and isn't future proof.
7067	* Especially on unix systems, weird stuff could happen if someone
7068	* unwittingly tinkers with the prep/setup code. What should really be
7069	* done here is using a new RTPathStartEx function that (via flags)
7070	* allows matching partial final component and returns the length of
7071	* what it matched up (in case it skipped slashes and '.' components).
7072	*
7073	*/
7074	static char vmdkStrReplace(const char pszWhere, const char *pszWhat,
7075	const char *pszByWhat)
7076	{
7077	AssertPtr(pszWhere);
7078	AssertPtr(pszWhat);
7079	AssertPtr(pszByWhat);
7080	const char *pszFoundStr = strstr(pszWhere, pszWhat);
7081	if (!pszFoundStr)
7082	{
7083	LogFlowFunc(("Failed to find '%s' in '%s'!\n", pszWhat, pszWhere));
7084	return NULL;
7085	}
7086	size_t cbFinal = strlen(pszWhere) + 1 + strlen(pszByWhat) - strlen(pszWhat);
7087	char *pszNewStr = RTStrAlloc(cbFinal);
7088	if (pszNewStr)
7089	{
7090	char *pszTmp = pszNewStr;
7091	memcpy(pszTmp, pszWhere, pszFoundStr - pszWhere);
7092	pszTmp += pszFoundStr - pszWhere;
7093	memcpy(pszTmp, pszByWhat, strlen(pszByWhat));
7094	pszTmp += strlen(pszByWhat);
7095	strcpy(pszTmp, pszFoundStr + strlen(pszWhat));
7096	}
7097	return pszNewStr;
7098	}
7099
7100
7101	/** @copydoc VDIMAGEBACKEND::pfnProbe */
7102	static DECLCALLBACK(int) vmdkProbe(const char *pszFilename, PVDINTERFACE pVDIfsDisk,
7103	PVDINTERFACE pVDIfsImage, VDTYPE enmDesiredType, VDTYPE *penmType)
7104	{
7105	RT_NOREF(enmDesiredType);
7106	LogFlowFunc(("pszFilename=\"%s\" pVDIfsDisk=%#p pVDIfsImage=%#p penmType=%#p\n",
7107	pszFilename, pVDIfsDisk, pVDIfsImage, penmType));
7108	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
7109	AssertReturn(*pszFilename != '\0', VERR_INVALID_PARAMETER);
7110
7111	int rc = VINF_SUCCESS;
7112	PVMDKIMAGE pImage = (PVMDKIMAGE)RTMemAllocZ(RT_UOFFSETOF(VMDKIMAGE, RegionList.aRegions[1]));
7113	if (RT_LIKELY(pImage))
7114	{
7115	pImage->pszFilename = pszFilename;
7116	pImage->pFile = NULL;
7117	pImage->pExtents = NULL;
7118	pImage->pFiles = NULL;
7119	pImage->pGTCache = NULL;
7120	pImage->pDescData = NULL;
7121	pImage->pVDIfsDisk = pVDIfsDisk;
7122	pImage->pVDIfsImage = pVDIfsImage;
7123	/** @todo speed up this test open (VD_OPEN_FLAGS_INFO) by skipping as
7124	* much as possible in vmdkOpenImage. */
7125	rc = vmdkOpenImage(pImage, VD_OPEN_FLAGS_INFO \| VD_OPEN_FLAGS_READONLY);
7126	vmdkFreeImage(pImage, false, false /fFlush/);
7127	RTMemFree(pImage);
7128
7129	if (RT_SUCCESS(rc))
7130	*penmType = VDTYPE_HDD;
7131	}
7132	else
7133	rc = VERR_NO_MEMORY;
7134
7135	LogFlowFunc(("returns %Rrc\n", rc));
7136	return rc;
7137	}
7138
7139	/** @copydoc VDIMAGEBACKEND::pfnOpen */
7140	static DECLCALLBACK(int) vmdkOpen(const char *pszFilename, unsigned uOpenFlags,
7141	PVDINTERFACE pVDIfsDisk, PVDINTERFACE pVDIfsImage,
7142	VDTYPE enmType, void **ppBackendData)
7143	{
7144	RT_NOREF1(enmType); /*< @todo r=klaus make use of the type info. /
7145
7146	LogFlowFunc(("pszFilename=\"%s\" uOpenFlags=%#x pVDIfsDisk=%#p pVDIfsImage=%#p enmType=%u ppBackendData=%#p\n",
7147	pszFilename, uOpenFlags, pVDIfsDisk, pVDIfsImage, enmType, ppBackendData));
7148	int rc;
7149
7150	/* Check open flags. All valid flags are supported. */
7151	AssertReturn(!(uOpenFlags & ~VD_OPEN_FLAGS_MASK), VERR_INVALID_PARAMETER);
7152	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
7153	AssertReturn(*pszFilename != '\0', VERR_INVALID_PARAMETER);
7154
7155
7156	PVMDKIMAGE pImage = (PVMDKIMAGE)RTMemAllocZ(RT_UOFFSETOF(VMDKIMAGE, RegionList.aRegions[1]));
7157	if (RT_LIKELY(pImage))
7158	{
7159	pImage->pszFilename = pszFilename;
7160	pImage->pFile = NULL;
7161	pImage->pExtents = NULL;
7162	pImage->pFiles = NULL;
7163	pImage->pGTCache = NULL;
7164	pImage->pDescData = NULL;
7165	pImage->pVDIfsDisk = pVDIfsDisk;
7166	pImage->pVDIfsImage = pVDIfsImage;
7167
7168	rc = vmdkOpenImage(pImage, uOpenFlags);
7169	if (RT_SUCCESS(rc))
7170	*ppBackendData = pImage;
7171	else
7172	RTMemFree(pImage);
7173	}
7174	else
7175	rc = VERR_NO_MEMORY;
7176
7177	LogFlowFunc(("returns %Rrc (pBackendData=%#p)\n", rc, *ppBackendData));
7178	return rc;
7179	}
7180
7181	/** @copydoc VDIMAGEBACKEND::pfnCreate */
7182	static DECLCALLBACK(int) vmdkCreate(const char *pszFilename, uint64_t cbSize,
7183	unsigned uImageFlags, const char *pszComment,
7184	PCVDGEOMETRY pPCHSGeometry, PCVDGEOMETRY pLCHSGeometry,
7185	PCRTUUID pUuid, unsigned uOpenFlags,
7186	unsigned uPercentStart, unsigned uPercentSpan,
7187	PVDINTERFACE pVDIfsDisk, PVDINTERFACE pVDIfsImage,
7188	PVDINTERFACE pVDIfsOperation, VDTYPE enmType,
7189	void **ppBackendData)
7190	{
7191	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",
7192	pszFilename, cbSize, uImageFlags, pszComment, pPCHSGeometry, pLCHSGeometry, pUuid, uOpenFlags, uPercentStart, uPercentSpan, pVDIfsDisk, pVDIfsImage, pVDIfsOperation, enmType, ppBackendData));
7193	int rc;
7194
7195	/* Check the VD container type and image flags. */
7196	if ( enmType != VDTYPE_HDD
7197	\|\| (uImageFlags & ~VD_VMDK_IMAGE_FLAGS_MASK) != 0)
7198	return VERR_VD_INVALID_TYPE;
7199
7200	/* Check size. Maximum 256TB-64K for sparse images, otherwise unlimited. */
7201	if ( !(uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
7202	&& ( !cbSize
7203	\|\| (!(uImageFlags & VD_IMAGE_FLAGS_FIXED) && cbSize >= _1T * 256 - _64K)))
7204	return VERR_VD_INVALID_SIZE;
7205
7206	/* Check image flags for invalid combinations. */
7207	if ( (uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7208	&& (uImageFlags & ~(VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED \| VD_IMAGE_FLAGS_DIFF)))
7209	return VERR_INVALID_PARAMETER;
7210
7211	/* Check open flags. All valid flags are supported. */
7212	AssertReturn(!(uOpenFlags & ~VD_OPEN_FLAGS_MASK), VERR_INVALID_PARAMETER);
7213	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
7214	AssertReturn(*pszFilename != '\0', VERR_INVALID_PARAMETER);
7215	AssertPtrReturn(pPCHSGeometry, VERR_INVALID_POINTER);
7216	AssertPtrReturn(pLCHSGeometry, VERR_INVALID_POINTER);
7217	AssertReturn(!( uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX
7218	&& !(uImageFlags & VD_IMAGE_FLAGS_FIXED)),
7219	VERR_INVALID_PARAMETER);
7220
7221	PVMDKIMAGE pImage = (PVMDKIMAGE)RTMemAllocZ(RT_UOFFSETOF(VMDKIMAGE, RegionList.aRegions[1]));
7222	if (RT_LIKELY(pImage))
7223	{
7224	PVDINTERFACEPROGRESS pIfProgress = VDIfProgressGet(pVDIfsOperation);
7225
7226	pImage->pszFilename = pszFilename;
7227	pImage->pFile = NULL;
7228	pImage->pExtents = NULL;
7229	pImage->pFiles = NULL;
7230	pImage->pGTCache = NULL;
7231	pImage->pDescData = NULL;
7232	pImage->pVDIfsDisk = pVDIfsDisk;
7233	pImage->pVDIfsImage = pVDIfsImage;
7234	/* Descriptors for split images can be pretty large, especially if the
7235	* filename is long. So prepare for the worst, and allocate quite some
7236	* memory for the descriptor in this case. */
7237	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
7238	pImage->cbDescAlloc = VMDK_SECTOR2BYTE(200);
7239	else
7240	pImage->cbDescAlloc = VMDK_SECTOR2BYTE(20);
7241	pImage->pDescData = (char *)RTMemAllocZ(pImage->cbDescAlloc);
7242	if (RT_LIKELY(pImage->pDescData))
7243	{
7244	rc = vmdkCreateImage(pImage, cbSize, uImageFlags, pszComment,
7245	pPCHSGeometry, pLCHSGeometry, pUuid,
7246	pIfProgress, uPercentStart, uPercentSpan);
7247	if (RT_SUCCESS(rc))
7248	{
7249	/* So far the image is opened in read/write mode. Make sure the
7250	* image is opened in read-only mode if the caller requested that. */
7251	if (uOpenFlags & VD_OPEN_FLAGS_READONLY)
7252	{
7253	vmdkFreeImage(pImage, false, true /fFlush/);
7254	rc = vmdkOpenImage(pImage, uOpenFlags);
7255	}
7256
7257	if (RT_SUCCESS(rc))
7258	*ppBackendData = pImage;
7259	}
7260
7261	if (RT_FAILURE(rc))
7262	RTMemFree(pImage->pDescData);
7263	}
7264	else
7265	rc = VERR_NO_MEMORY;
7266
7267	if (RT_FAILURE(rc))
7268	RTMemFree(pImage);
7269	}
7270	else
7271	rc = VERR_NO_MEMORY;
7272
7273	LogFlowFunc(("returns %Rrc (pBackendData=%#p)\n", rc, *ppBackendData));
7274	return rc;
7275	}
7276
7277	/**
7278	* Prepares the state for renaming a VMDK image, setting up the state and allocating
7279	* memory.
7280	*
7281	* @returns VBox status code.
7282	* @param pImage VMDK image instance.
7283	* @param pRenameState The state to initialize.
7284	* @param pszFilename The new filename.
7285	*/
7286	static int vmdkRenameStatePrepare(PVMDKIMAGE pImage, PVMDKRENAMESTATE pRenameState, const char *pszFilename)
7287	{
7288	AssertReturn(RTPathFilename(pszFilename) != NULL, VERR_INVALID_PARAMETER);
7289
7290	int rc = VINF_SUCCESS;
7291
7292	memset(&pRenameState->DescriptorCopy, 0, sizeof(pRenameState->DescriptorCopy));
7293
7294	/*
7295	* Allocate an array to store both old and new names of renamed files
7296	* in case we have to roll back the changes. Arrays are initialized
7297	* with zeros. We actually save stuff when and if we change it.
7298	*/
7299	pRenameState->cExtents = pImage->cExtents;
7300	pRenameState->apszOldName = (char *)RTMemTmpAllocZ((pRenameState->cExtents + 1) sizeof(char *));
7301	pRenameState->apszNewName = (char *)RTMemTmpAllocZ((pRenameState->cExtents + 1) sizeof(char *));
7302	pRenameState->apszNewLines = (char *)RTMemTmpAllocZ(pRenameState->cExtents sizeof(char *));
7303	if ( pRenameState->apszOldName
7304	&& pRenameState->apszNewName
7305	&& pRenameState->apszNewLines)
7306	{
7307	/* Save the descriptor size and position. */
7308	if (pImage->pDescData)
7309	{
7310	/* Separate descriptor file. */
7311	pRenameState->fEmbeddedDesc = false;
7312	}
7313	else
7314	{
7315	/* Embedded descriptor file. */
7316	pRenameState->ExtentCopy = pImage->pExtents[0];
7317	pRenameState->fEmbeddedDesc = true;
7318	}
7319
7320	/* Save the descriptor content. */
7321	pRenameState->DescriptorCopy.cLines = pImage->Descriptor.cLines;
7322	for (unsigned i = 0; i < pRenameState->DescriptorCopy.cLines; i++)
7323	{
7324	pRenameState->DescriptorCopy.aLines[i] = RTStrDup(pImage->Descriptor.aLines[i]);
7325	if (!pRenameState->DescriptorCopy.aLines[i])
7326	{
7327	rc = VERR_NO_MEMORY;
7328	break;
7329	}
7330	}
7331
7332	if (RT_SUCCESS(rc))
7333	{
7334	/* Prepare both old and new base names used for string replacement. */
7335	pRenameState->pszNewBaseName = RTStrDup(RTPathFilename(pszFilename));
7336	AssertReturn(pRenameState->pszNewBaseName, VERR_NO_STR_MEMORY);
7337	RTPathStripSuffix(pRenameState->pszNewBaseName);
7338
7339	pRenameState->pszOldBaseName = RTStrDup(RTPathFilename(pImage->pszFilename));
7340	AssertReturn(pRenameState->pszOldBaseName, VERR_NO_STR_MEMORY);
7341	RTPathStripSuffix(pRenameState->pszOldBaseName);
7342
7343	/* Prepare both old and new full names used for string replacement.
7344	Note! Must abspath the stuff here, so the strstr weirdness later in
7345	the renaming process get a match against abspath'ed extent paths.
7346	See RTPathAbsDup call in vmdkDescriptorReadSparse(). */
7347	pRenameState->pszNewFullName = RTPathAbsDup(pszFilename);
7348	AssertReturn(pRenameState->pszNewFullName, VERR_NO_STR_MEMORY);
7349	RTPathStripSuffix(pRenameState->pszNewFullName);
7350
7351	pRenameState->pszOldFullName = RTPathAbsDup(pImage->pszFilename);
7352	AssertReturn(pRenameState->pszOldFullName, VERR_NO_STR_MEMORY);
7353	RTPathStripSuffix(pRenameState->pszOldFullName);
7354
7355	/* Save the old name for easy access to the old descriptor file. */
7356	pRenameState->pszOldDescName = RTStrDup(pImage->pszFilename);
7357	AssertReturn(pRenameState->pszOldDescName, VERR_NO_STR_MEMORY);
7358
7359	/* Save old image name. */
7360	pRenameState->pszOldImageName = pImage->pszFilename;
7361	}
7362	}
7363	else
7364	rc = VERR_NO_TMP_MEMORY;
7365
7366	return rc;
7367	}
7368
7369	/**
7370	* Destroys the given rename state, freeing all allocated memory.
7371	*
7372	* @param pRenameState The rename state to destroy.
7373	*/
7374	static void vmdkRenameStateDestroy(PVMDKRENAMESTATE pRenameState)
7375	{
7376	for (unsigned i = 0; i < pRenameState->DescriptorCopy.cLines; i++)
7377	if (pRenameState->DescriptorCopy.aLines[i])
7378	RTStrFree(pRenameState->DescriptorCopy.aLines[i]);
7379	if (pRenameState->apszOldName)
7380	{
7381	for (unsigned i = 0; i <= pRenameState->cExtents; i++)
7382	if (pRenameState->apszOldName[i])
7383	RTStrFree(pRenameState->apszOldName[i]);
7384	RTMemTmpFree(pRenameState->apszOldName);
7385	}
7386	if (pRenameState->apszNewName)
7387	{
7388	for (unsigned i = 0; i <= pRenameState->cExtents; i++)
7389	if (pRenameState->apszNewName[i])
7390	RTStrFree(pRenameState->apszNewName[i]);
7391	RTMemTmpFree(pRenameState->apszNewName);
7392	}
7393	if (pRenameState->apszNewLines)
7394	{
7395	for (unsigned i = 0; i < pRenameState->cExtents; i++)
7396	if (pRenameState->apszNewLines[i])
7397	RTStrFree(pRenameState->apszNewLines[i]);
7398	RTMemTmpFree(pRenameState->apszNewLines);
7399	}
7400	if (pRenameState->pszOldDescName)
7401	RTStrFree(pRenameState->pszOldDescName);
7402	if (pRenameState->pszOldBaseName)
7403	RTStrFree(pRenameState->pszOldBaseName);
7404	if (pRenameState->pszNewBaseName)
7405	RTStrFree(pRenameState->pszNewBaseName);
7406	if (pRenameState->pszOldFullName)
7407	RTStrFree(pRenameState->pszOldFullName);
7408	if (pRenameState->pszNewFullName)
7409	RTStrFree(pRenameState->pszNewFullName);
7410	}
7411
7412	/**
7413	* Rolls back the rename operation to the original state.
7414	*
7415	* @returns VBox status code.
7416	* @param pImage VMDK image instance.
7417	* @param pRenameState The rename state.
7418	*/
7419	static int vmdkRenameRollback(PVMDKIMAGE pImage, PVMDKRENAMESTATE pRenameState)
7420	{
7421	int rc = VINF_SUCCESS;
7422
7423	if (!pRenameState->fImageFreed)
7424	{
7425	/*
7426	* Some extents may have been closed, close the rest. We will
7427	* re-open the whole thing later.
7428	*/
7429	vmdkFreeImage(pImage, false, true /fFlush/);
7430	}
7431
7432	/* Rename files back. */
7433	for (unsigned i = 0; i <= pRenameState->cExtents; i++)
7434	{
7435	if (pRenameState->apszOldName[i])
7436	{
7437	rc = vdIfIoIntFileMove(pImage->pIfIo, pRenameState->apszNewName[i], pRenameState->apszOldName[i], 0);
7438	AssertRC(rc);
7439	}
7440	}
7441	/* Restore the old descriptor. */
7442	PVMDKFILE pFile;
7443	rc = vmdkFileOpen(pImage, &pFile, NULL, pRenameState->pszOldDescName,
7444	VDOpenFlagsToFileOpenFlags(VD_OPEN_FLAGS_NORMAL,
7445	false /* fCreate */));
7446	AssertRC(rc);
7447	if (pRenameState->fEmbeddedDesc)
7448	{
7449	pRenameState->ExtentCopy.pFile = pFile;
7450	pImage->pExtents = &pRenameState->ExtentCopy;
7451	}
7452	else
7453	{
7454	/* Shouldn't be null for separate descriptor.
7455	* There will be no access to the actual content.
7456	*/
7457	pImage->pDescData = pRenameState->pszOldDescName;
7458	pImage->pFile = pFile;
7459	}
7460	pImage->Descriptor = pRenameState->DescriptorCopy;
7461	vmdkWriteDescriptor(pImage, NULL);
7462	vmdkFileClose(pImage, &pFile, false);
7463	/* Get rid of the stuff we implanted. */
7464	pImage->pExtents = NULL;
7465	pImage->pFile = NULL;
7466	pImage->pDescData = NULL;
7467	/* Re-open the image back. */
7468	pImage->pszFilename = pRenameState->pszOldImageName;
7469	rc = vmdkOpenImage(pImage, pImage->uOpenFlags);
7470
7471	return rc;
7472	}
7473
7474	/**
7475	* Rename worker doing the real work.
7476	*
7477	* @returns VBox status code.
7478	* @param pImage VMDK image instance.
7479	* @param pRenameState The rename state.
7480	* @param pszFilename The new filename.
7481	*/
7482	static int vmdkRenameWorker(PVMDKIMAGE pImage, PVMDKRENAMESTATE pRenameState, const char *pszFilename)
7483	{
7484	int rc = VINF_SUCCESS;
7485	unsigned i, line;
7486
7487	/* Update the descriptor with modified extent names. */
7488	for (i = 0, line = pImage->Descriptor.uFirstExtent;
7489	i < pRenameState->cExtents;
7490	i++, line = pImage->Descriptor.aNextLines[line])
7491	{
7492	/* Update the descriptor. */
7493	pRenameState->apszNewLines[i] = vmdkStrReplace(pImage->Descriptor.aLines[line],
7494	pRenameState->pszOldBaseName,
7495	pRenameState->pszNewBaseName);
7496	if (!pRenameState->apszNewLines[i])
7497	{
7498	rc = VERR_NO_MEMORY;
7499	break;
7500	}
7501	pImage->Descriptor.aLines[line] = pRenameState->apszNewLines[i];
7502	}
7503
7504	if (RT_SUCCESS(rc))
7505	{
7506	/* Make sure the descriptor gets written back. */
7507	pImage->Descriptor.fDirty = true;
7508	/* Flush the descriptor now, in case it is embedded. */
7509	vmdkFlushImage(pImage, NULL);
7510
7511	/* Close and rename/move extents. */
7512	for (i = 0; i < pRenameState->cExtents; i++)
7513	{
7514	PVMDKEXTENT pExtent = &pImage->pExtents[i];
7515	/* Compose new name for the extent. */
7516	pRenameState->apszNewName[i] = vmdkStrReplace(pExtent->pszFullname,
7517	pRenameState->pszOldFullName,
7518	pRenameState->pszNewFullName);
7519	if (!pRenameState->apszNewName[i])
7520	{
7521	rc = VERR_NO_MEMORY;
7522	break;
7523	}
7524	/* Close the extent file. */
7525	rc = vmdkFileClose(pImage, &pExtent->pFile, false);
7526	if (RT_FAILURE(rc))
7527	break;;
7528
7529	/* Rename the extent file. */
7530	rc = vdIfIoIntFileMove(pImage->pIfIo, pExtent->pszFullname, pRenameState->apszNewName[i], 0);
7531	if (RT_FAILURE(rc))
7532	break;
7533	/* Remember the old name. */
7534	pRenameState->apszOldName[i] = RTStrDup(pExtent->pszFullname);
7535	}
7536
7537	if (RT_SUCCESS(rc))
7538	{
7539	/* Release all old stuff. */
7540	rc = vmdkFreeImage(pImage, false, true /fFlush/);
7541	if (RT_SUCCESS(rc))
7542	{
7543	pRenameState->fImageFreed = true;
7544
7545	/* Last elements of new/old name arrays are intended for
7546	* storing descriptor's names.
7547	*/
7548	pRenameState->apszNewName[pRenameState->cExtents] = RTStrDup(pszFilename);
7549	/* Rename the descriptor file if it's separate. */
7550	if (!pRenameState->fEmbeddedDesc)
7551	{
7552	rc = vdIfIoIntFileMove(pImage->pIfIo, pImage->pszFilename, pRenameState->apszNewName[pRenameState->cExtents], 0);
7553	if (RT_SUCCESS(rc))
7554	{
7555	/* Save old name only if we may need to change it back. */
7556	pRenameState->apszOldName[pRenameState->cExtents] = RTStrDup(pszFilename);
7557	}
7558	}
7559
7560	/* Update pImage with the new information. */
7561	pImage->pszFilename = pszFilename;
7562
7563	/* Open the new image. */
7564	rc = vmdkOpenImage(pImage, pImage->uOpenFlags);
7565	}
7566	}
7567	}
7568
7569	return rc;
7570	}
7571
7572	/** @copydoc VDIMAGEBACKEND::pfnRename */
7573	static DECLCALLBACK(int) vmdkRename(void pBackendData, const char pszFilename)
7574	{
7575	LogFlowFunc(("pBackendData=%#p pszFilename=%#p\n", pBackendData, pszFilename));
7576
7577	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7578	VMDKRENAMESTATE RenameState;
7579
7580	memset(&RenameState, 0, sizeof(RenameState));
7581
7582	/* Check arguments. */
7583	AssertPtrReturn(pImage, VERR_INVALID_POINTER);
7584	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
7585	AssertReturn(*pszFilename != '\0', VERR_INVALID_PARAMETER);
7586	AssertReturn(!(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK), VERR_INVALID_PARAMETER);
7587
7588	int rc = vmdkRenameStatePrepare(pImage, &RenameState, pszFilename);
7589	if (RT_SUCCESS(rc))
7590	{
7591	/* --- Up to this point we have not done any damage yet. --- */
7592
7593	rc = vmdkRenameWorker(pImage, &RenameState, pszFilename);
7594	/* Roll back all changes in case of failure. */
7595	if (RT_FAILURE(rc))
7596	{
7597	int rrc = vmdkRenameRollback(pImage, &RenameState);
7598	AssertRC(rrc);
7599	}
7600	}
7601
7602	vmdkRenameStateDestroy(&RenameState);
7603	LogFlowFunc(("returns %Rrc\n", rc));
7604	return rc;
7605	}
7606
7607	/** @copydoc VDIMAGEBACKEND::pfnClose */
7608	static DECLCALLBACK(int) vmdkClose(void *pBackendData, bool fDelete)
7609	{
7610	LogFlowFunc(("pBackendData=%#p fDelete=%d\n", pBackendData, fDelete));
7611	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7612
7613	int rc = vmdkFreeImage(pImage, fDelete, true /fFlush/);
7614	RTMemFree(pImage);
7615
7616	LogFlowFunc(("returns %Rrc\n", rc));
7617	return rc;
7618	}
7619
7620	/** @copydoc VDIMAGEBACKEND::pfnRead */
7621	static DECLCALLBACK(int) vmdkRead(void *pBackendData, uint64_t uOffset, size_t cbToRead,
7622	PVDIOCTX pIoCtx, size_t *pcbActuallyRead)
7623	{
7624	LogFlowFunc(("pBackendData=%#p uOffset=%llu pIoCtx=%#p cbToRead=%zu pcbActuallyRead=%#p\n",
7625	pBackendData, uOffset, pIoCtx, cbToRead, pcbActuallyRead));
7626	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7627
7628	AssertPtr(pImage);
7629	Assert(uOffset % 512 == 0);
7630	Assert(cbToRead % 512 == 0);
7631	AssertPtrReturn(pIoCtx, VERR_INVALID_POINTER);
7632	AssertReturn(cbToRead, VERR_INVALID_PARAMETER);
7633	AssertReturn(uOffset + cbToRead <= pImage->cbSize, VERR_INVALID_PARAMETER);
7634
7635	/* Find the extent and check access permissions as defined in the extent descriptor. */
7636	PVMDKEXTENT pExtent;
7637	uint64_t uSectorExtentRel;
7638	int rc = vmdkFindExtent(pImage, VMDK_BYTE2SECTOR(uOffset),
7639	&pExtent, &uSectorExtentRel);
7640	if ( RT_SUCCESS(rc)
7641	&& pExtent->enmAccess != VMDKACCESS_NOACCESS)
7642	{
7643	/* Clip read range to remain in this extent. */
7644	cbToRead = RT_MIN(cbToRead, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
7645
7646	/* Handle the read according to the current extent type. */
7647	switch (pExtent->enmType)
7648	{
7649	case VMDKETYPE_HOSTED_SPARSE:
7650	{
7651	uint64_t uSectorExtentAbs;
7652
7653	rc = vmdkGetSector(pImage, pIoCtx, pExtent, uSectorExtentRel, &uSectorExtentAbs);
7654	if (RT_FAILURE(rc))
7655	break;
7656	/* Clip read range to at most the rest of the grain. */
7657	cbToRead = RT_MIN(cbToRead, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain - uSectorExtentRel % pExtent->cSectorsPerGrain));
7658	Assert(!(cbToRead % 512));
7659	if (uSectorExtentAbs == 0)
7660	{
7661	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7662	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
7663	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
7664	rc = VERR_VD_BLOCK_FREE;
7665	else
7666	rc = vmdkStreamReadSequential(pImage, pExtent,
7667	uSectorExtentRel,
7668	pIoCtx, cbToRead);
7669	}
7670	else
7671	{
7672	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7673	{
7674	AssertMsg(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
7675	("Async I/O is not supported for stream optimized VMDK's\n"));
7676
7677	uint32_t uSectorInGrain = uSectorExtentRel % pExtent->cSectorsPerGrain;
7678	uSectorExtentAbs -= uSectorInGrain;
7679	if (pExtent->uGrainSectorAbs != uSectorExtentAbs)
7680	{
7681	uint64_t uLBA = 0; /* gcc maybe uninitialized */
7682	rc = vmdkFileInflateSync(pImage, pExtent,
7683	VMDK_SECTOR2BYTE(uSectorExtentAbs),
7684	pExtent->pvGrain,
7685	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
7686	NULL, &uLBA, NULL);
7687	if (RT_FAILURE(rc))
7688	{
7689	pExtent->uGrainSectorAbs = 0;
7690	break;
7691	}
7692	pExtent->uGrainSectorAbs = uSectorExtentAbs;
7693	pExtent->uGrain = uSectorExtentRel / pExtent->cSectorsPerGrain;
7694	Assert(uLBA == uSectorExtentRel);
7695	}
7696	vdIfIoIntIoCtxCopyTo(pImage->pIfIo, pIoCtx,
7697	(uint8_t *)pExtent->pvGrain
7698	+ VMDK_SECTOR2BYTE(uSectorInGrain),
7699	cbToRead);
7700	}
7701	else
7702	rc = vdIfIoIntFileReadUser(pImage->pIfIo, pExtent->pFile->pStorage,
7703	VMDK_SECTOR2BYTE(uSectorExtentAbs),
7704	pIoCtx, cbToRead);
7705	}
7706	break;
7707	}
7708	case VMDKETYPE_VMFS:
7709	case VMDKETYPE_FLAT:
7710	rc = vdIfIoIntFileReadUser(pImage->pIfIo, pExtent->pFile->pStorage,
7711	VMDK_SECTOR2BYTE(uSectorExtentRel),
7712	pIoCtx, cbToRead);
7713	break;
7714	case VMDKETYPE_ZERO:
7715	{
7716	size_t cbSet;
7717
7718	cbSet = vdIfIoIntIoCtxSet(pImage->pIfIo, pIoCtx, 0, cbToRead);
7719	Assert(cbSet == cbToRead);
7720	break;
7721	}
7722	}
7723	if (pcbActuallyRead)
7724	*pcbActuallyRead = cbToRead;
7725	}
7726	else if (RT_SUCCESS(rc))
7727	rc = VERR_VD_VMDK_INVALID_STATE;
7728
7729	LogFlowFunc(("returns %Rrc\n", rc));
7730	return rc;
7731	}
7732
7733	/** @copydoc VDIMAGEBACKEND::pfnWrite */
7734	static DECLCALLBACK(int) vmdkWrite(void *pBackendData, uint64_t uOffset, size_t cbToWrite,
7735	PVDIOCTX pIoCtx, size_t pcbWriteProcess, size_t pcbPreRead,
7736	size_t *pcbPostRead, unsigned fWrite)
7737	{
7738	LogFlowFunc(("pBackendData=%#p uOffset=%llu pIoCtx=%#p cbToWrite=%zu pcbWriteProcess=%#p pcbPreRead=%#p pcbPostRead=%#p\n",
7739	pBackendData, uOffset, pIoCtx, cbToWrite, pcbWriteProcess, pcbPreRead, pcbPostRead));
7740	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7741	int rc;
7742
7743	AssertPtr(pImage);
7744	Assert(uOffset % 512 == 0);
7745	Assert(cbToWrite % 512 == 0);
7746	AssertPtrReturn(pIoCtx, VERR_INVALID_POINTER);
7747	AssertReturn(cbToWrite, VERR_INVALID_PARAMETER);
7748
7749	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7750	{
7751	PVMDKEXTENT pExtent;
7752	uint64_t uSectorExtentRel;
7753	uint64_t uSectorExtentAbs;
7754
7755	/* No size check here, will do that later when the extent is located.
7756	* There are sparse images out there which according to the spec are
7757	* invalid, because the total size is not a multiple of the grain size.
7758	* Also for sparse images which are stitched together in odd ways (not at
7759	* grain boundaries, and with the nominal size not being a multiple of the
7760	* grain size), this would prevent writing to the last grain. */
7761
7762	rc = vmdkFindExtent(pImage, VMDK_BYTE2SECTOR(uOffset),
7763	&pExtent, &uSectorExtentRel);
7764	if (RT_SUCCESS(rc))
7765	{
7766	if ( pExtent->enmAccess != VMDKACCESS_READWRITE
7767	&& ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7768	&& !pImage->pExtents[0].uAppendPosition
7769	&& pExtent->enmAccess != VMDKACCESS_READONLY))
7770	rc = VERR_VD_VMDK_INVALID_STATE;
7771	else
7772	{
7773	/* Handle the write according to the current extent type. */
7774	switch (pExtent->enmType)
7775	{
7776	case VMDKETYPE_HOSTED_SPARSE:
7777	rc = vmdkGetSector(pImage, pIoCtx, pExtent, uSectorExtentRel, &uSectorExtentAbs);
7778	if (RT_SUCCESS(rc))
7779	{
7780	if ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
7781	&& uSectorExtentRel < (uint64_t)pExtent->uLastGrainAccess * pExtent->cSectorsPerGrain)
7782	rc = VERR_VD_VMDK_INVALID_WRITE;
7783	else
7784	{
7785	/* Clip write range to at most the rest of the grain. */
7786	cbToWrite = RT_MIN(cbToWrite,
7787	VMDK_SECTOR2BYTE( pExtent->cSectorsPerGrain
7788	- uSectorExtentRel % pExtent->cSectorsPerGrain));
7789	if (uSectorExtentAbs == 0)
7790	{
7791	if (!(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7792	{
7793	if (cbToWrite == VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
7794	{
7795	/* Full block write to a previously unallocated block.
7796	* Check if the caller wants to avoid the automatic alloc. */
7797	if (!(fWrite & VD_WRITE_NO_ALLOC))
7798	{
7799	/* Allocate GT and find out where to store the grain. */
7800	rc = vmdkAllocGrain(pImage, pExtent, pIoCtx,
7801	uSectorExtentRel, cbToWrite);
7802	}
7803	else
7804	rc = VERR_VD_BLOCK_FREE;
7805	*pcbPreRead = 0;
7806	*pcbPostRead = 0;
7807	}
7808	else
7809	{
7810	/* Clip write range to remain in this extent. */
7811	cbToWrite = RT_MIN(cbToWrite,
7812	VMDK_SECTOR2BYTE( pExtent->uSectorOffset
7813	+ pExtent->cNominalSectors - uSectorExtentRel));
7814	*pcbPreRead = VMDK_SECTOR2BYTE(uSectorExtentRel % pExtent->cSectorsPerGrain);
7815	pcbPostRead = VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain) - cbToWrite - pcbPreRead;
7816	rc = VERR_VD_BLOCK_FREE;
7817	}
7818	}
7819	else
7820	rc = vmdkStreamAllocGrain(pImage, pExtent, uSectorExtentRel,
7821	pIoCtx, cbToWrite);
7822	}
7823	else
7824	{
7825	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7826	{
7827	/* A partial write to a streamOptimized image is simply
7828	* invalid. It requires rewriting already compressed data
7829	* which is somewhere between expensive and impossible. */
7830	rc = VERR_VD_VMDK_INVALID_STATE;
7831	pExtent->uGrainSectorAbs = 0;
7832	AssertRC(rc);
7833	}
7834	else
7835	{
7836	Assert(!(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED));
7837	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
7838	VMDK_SECTOR2BYTE(uSectorExtentAbs),
7839	pIoCtx, cbToWrite, NULL, NULL);
7840	}
7841	}
7842	}
7843	}
7844	break;
7845	case VMDKETYPE_VMFS:
7846	case VMDKETYPE_FLAT:
7847	/* Clip write range to remain in this extent. */
7848	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
7849	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
7850	VMDK_SECTOR2BYTE(uSectorExtentRel),
7851	pIoCtx, cbToWrite, NULL, NULL);
7852	break;
7853	case VMDKETYPE_ZERO:
7854	/* Clip write range to remain in this extent. */
7855	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
7856	break;
7857	}
7858	}
7859
7860	if (pcbWriteProcess)
7861	*pcbWriteProcess = cbToWrite;
7862	}
7863	}
7864	else
7865	rc = VERR_VD_IMAGE_READ_ONLY;
7866
7867	LogFlowFunc(("returns %Rrc\n", rc));
7868	return rc;
7869	}
7870
7871	/** @copydoc VDIMAGEBACKEND::pfnFlush */
7872	static DECLCALLBACK(int) vmdkFlush(void *pBackendData, PVDIOCTX pIoCtx)
7873	{
7874	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7875
7876	return vmdkFlushImage(pImage, pIoCtx);
7877	}
7878
7879	/** @copydoc VDIMAGEBACKEND::pfnGetVersion */
7880	static DECLCALLBACK(unsigned) vmdkGetVersion(void *pBackendData)
7881	{
7882	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
7883	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7884
7885	AssertPtrReturn(pImage, 0);
7886
7887	return VMDK_IMAGE_VERSION;
7888	}
7889
7890	/** @copydoc VDIMAGEBACKEND::pfnGetFileSize */
7891	static DECLCALLBACK(uint64_t) vmdkGetFileSize(void *pBackendData)
7892	{
7893	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
7894	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7895	uint64_t cb = 0;
7896
7897	AssertPtrReturn(pImage, 0);
7898
7899	if (pImage->pFile != NULL)
7900	{
7901	uint64_t cbFile;
7902	int rc = vdIfIoIntFileGetSize(pImage->pIfIo, pImage->pFile->pStorage, &cbFile);
7903	if (RT_SUCCESS(rc))
7904	cb += cbFile;
7905	}
7906	for (unsigned i = 0; i < pImage->cExtents; i++)
7907	{
7908	if (pImage->pExtents[i].pFile != NULL)
7909	{
7910	uint64_t cbFile;
7911	int rc = vdIfIoIntFileGetSize(pImage->pIfIo, pImage->pExtents[i].pFile->pStorage, &cbFile);
7912	if (RT_SUCCESS(rc))
7913	cb += cbFile;
7914	}
7915	}
7916
7917	LogFlowFunc(("returns %lld\n", cb));
7918	return cb;
7919	}
7920
7921	/** @copydoc VDIMAGEBACKEND::pfnGetPCHSGeometry */
7922	static DECLCALLBACK(int) vmdkGetPCHSGeometry(void *pBackendData, PVDGEOMETRY pPCHSGeometry)
7923	{
7924	LogFlowFunc(("pBackendData=%#p pPCHSGeometry=%#p\n", pBackendData, pPCHSGeometry));
7925	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7926	int rc = VINF_SUCCESS;
7927
7928	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7929
7930	if (pImage->PCHSGeometry.cCylinders)
7931	*pPCHSGeometry = pImage->PCHSGeometry;
7932	else
7933	rc = VERR_VD_GEOMETRY_NOT_SET;
7934
7935	LogFlowFunc(("returns %Rrc (PCHS=%u/%u/%u)\n", rc, pPCHSGeometry->cCylinders, pPCHSGeometry->cHeads, pPCHSGeometry->cSectors));
7936	return rc;
7937	}
7938
7939	/** @copydoc VDIMAGEBACKEND::pfnSetPCHSGeometry */
7940	static DECLCALLBACK(int) vmdkSetPCHSGeometry(void *pBackendData, PCVDGEOMETRY pPCHSGeometry)
7941	{
7942	LogFlowFunc(("pBackendData=%#p pPCHSGeometry=%#p PCHS=%u/%u/%u\n",
7943	pBackendData, pPCHSGeometry, pPCHSGeometry->cCylinders, pPCHSGeometry->cHeads, pPCHSGeometry->cSectors));
7944	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7945	int rc = VINF_SUCCESS;
7946
7947	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7948
7949	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7950	{
7951	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7952	{
7953	rc = vmdkDescSetPCHSGeometry(pImage, pPCHSGeometry);
7954	if (RT_SUCCESS(rc))
7955	pImage->PCHSGeometry = *pPCHSGeometry;
7956	}
7957	else
7958	rc = VERR_NOT_SUPPORTED;
7959	}
7960	else
7961	rc = VERR_VD_IMAGE_READ_ONLY;
7962
7963	LogFlowFunc(("returns %Rrc\n", rc));
7964	return rc;
7965	}
7966
7967	/** @copydoc VDIMAGEBACKEND::pfnGetLCHSGeometry */
7968	static DECLCALLBACK(int) vmdkGetLCHSGeometry(void *pBackendData, PVDGEOMETRY pLCHSGeometry)
7969	{
7970	LogFlowFunc(("pBackendData=%#p pLCHSGeometry=%#p\n", pBackendData, pLCHSGeometry));
7971	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7972	int rc = VINF_SUCCESS;
7973
7974	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7975
7976	if (pImage->LCHSGeometry.cCylinders)
7977	*pLCHSGeometry = pImage->LCHSGeometry;
7978	else
7979	rc = VERR_VD_GEOMETRY_NOT_SET;
7980
7981	LogFlowFunc(("returns %Rrc (LCHS=%u/%u/%u)\n", rc, pLCHSGeometry->cCylinders, pLCHSGeometry->cHeads, pLCHSGeometry->cSectors));
7982	return rc;
7983	}
7984
7985	/** @copydoc VDIMAGEBACKEND::pfnSetLCHSGeometry */
7986	static DECLCALLBACK(int) vmdkSetLCHSGeometry(void *pBackendData, PCVDGEOMETRY pLCHSGeometry)
7987	{
7988	LogFlowFunc(("pBackendData=%#p pLCHSGeometry=%#p LCHS=%u/%u/%u\n",
7989	pBackendData, pLCHSGeometry, pLCHSGeometry->cCylinders, pLCHSGeometry->cHeads, pLCHSGeometry->cSectors));
7990	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7991	int rc = VINF_SUCCESS;
7992
7993	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7994
7995	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7996	{
7997	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7998	{
7999	rc = vmdkDescSetLCHSGeometry(pImage, pLCHSGeometry);
8000	if (RT_SUCCESS(rc))
8001	pImage->LCHSGeometry = *pLCHSGeometry;
8002	}
8003	else
8004	rc = VERR_NOT_SUPPORTED;
8005	}
8006	else
8007	rc = VERR_VD_IMAGE_READ_ONLY;
8008
8009	LogFlowFunc(("returns %Rrc\n", rc));
8010	return rc;
8011	}
8012
8013	/** @copydoc VDIMAGEBACKEND::pfnQueryRegions */
8014	static DECLCALLBACK(int) vmdkQueryRegions(void pBackendData, PCVDREGIONLIST ppRegionList)
8015	{
8016	LogFlowFunc(("pBackendData=%#p ppRegionList=%#p\n", pBackendData, ppRegionList));
8017	PVMDKIMAGE pThis = (PVMDKIMAGE)pBackendData;
8018
8019	AssertPtrReturn(pThis, VERR_VD_NOT_OPENED);
8020
8021	*ppRegionList = &pThis->RegionList;
8022	LogFlowFunc(("returns %Rrc\n", VINF_SUCCESS));
8023	return VINF_SUCCESS;
8024	}
8025
8026	/** @copydoc VDIMAGEBACKEND::pfnRegionListRelease */
8027	static DECLCALLBACK(void) vmdkRegionListRelease(void *pBackendData, PCVDREGIONLIST pRegionList)
8028	{
8029	RT_NOREF1(pRegionList);
8030	LogFlowFunc(("pBackendData=%#p pRegionList=%#p\n", pBackendData, pRegionList));
8031	PVMDKIMAGE pThis = (PVMDKIMAGE)pBackendData;
8032	AssertPtr(pThis); RT_NOREF(pThis);
8033
8034	/* Nothing to do here. */
8035	}
8036
8037	/** @copydoc VDIMAGEBACKEND::pfnGetImageFlags */
8038	static DECLCALLBACK(unsigned) vmdkGetImageFlags(void *pBackendData)
8039	{
8040	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
8041	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8042
8043	AssertPtrReturn(pImage, 0);
8044
8045	LogFlowFunc(("returns %#x\n", pImage->uImageFlags));
8046	return pImage->uImageFlags;
8047	}
8048
8049	/** @copydoc VDIMAGEBACKEND::pfnGetOpenFlags */
8050	static DECLCALLBACK(unsigned) vmdkGetOpenFlags(void *pBackendData)
8051	{
8052	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
8053	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8054
8055	AssertPtrReturn(pImage, 0);
8056
8057	LogFlowFunc(("returns %#x\n", pImage->uOpenFlags));
8058	return pImage->uOpenFlags;
8059	}
8060
8061	/** @copydoc VDIMAGEBACKEND::pfnSetOpenFlags */
8062	static DECLCALLBACK(int) vmdkSetOpenFlags(void *pBackendData, unsigned uOpenFlags)
8063	{
8064	LogFlowFunc(("pBackendData=%#p uOpenFlags=%#x\n", pBackendData, uOpenFlags));
8065	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8066	int rc;
8067
8068	/* Image must be opened and the new flags must be valid. */
8069	if (!pImage \|\| (uOpenFlags & ~( VD_OPEN_FLAGS_READONLY \| VD_OPEN_FLAGS_INFO
8070	\| VD_OPEN_FLAGS_ASYNC_IO \| VD_OPEN_FLAGS_SHAREABLE
8071	\| VD_OPEN_FLAGS_SEQUENTIAL \| VD_OPEN_FLAGS_SKIP_CONSISTENCY_CHECKS)))
8072	rc = VERR_INVALID_PARAMETER;
8073	else
8074	{
8075	/* StreamOptimized images need special treatment: reopen is prohibited. */
8076	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
8077	{
8078	if (pImage->uOpenFlags == uOpenFlags)
8079	rc = VINF_SUCCESS;
8080	else
8081	rc = VERR_INVALID_PARAMETER;
8082	}
8083	else
8084	{
8085	/* Implement this operation via reopening the image. */
8086	vmdkFreeImage(pImage, false, true /fFlush/);
8087	rc = vmdkOpenImage(pImage, uOpenFlags);
8088	}
8089	}
8090
8091	LogFlowFunc(("returns %Rrc\n", rc));
8092	return rc;
8093	}
8094
8095	/** @copydoc VDIMAGEBACKEND::pfnGetComment */
8096	static DECLCALLBACK(int) vmdkGetComment(void pBackendData, char pszComment, size_t cbComment)
8097	{
8098	LogFlowFunc(("pBackendData=%#p pszComment=%#p cbComment=%zu\n", pBackendData, pszComment, cbComment));
8099	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8100
8101	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8102
8103	char *pszCommentEncoded = NULL;
8104	int rc = vmdkDescDDBGetStr(pImage, &pImage->Descriptor,
8105	"ddb.comment", &pszCommentEncoded);
8106	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
8107	{
8108	pszCommentEncoded = NULL;
8109	rc = VINF_SUCCESS;
8110	}
8111
8112	if (RT_SUCCESS(rc))
8113	{
8114	if (pszComment && pszCommentEncoded)
8115	rc = vmdkDecodeString(pszCommentEncoded, pszComment, cbComment);
8116	else if (pszComment)
8117	*pszComment = '\0';
8118
8119	if (pszCommentEncoded)
8120	RTMemTmpFree(pszCommentEncoded);
8121	}
8122
8123	LogFlowFunc(("returns %Rrc comment='%s'\n", rc, pszComment));
8124	return rc;
8125	}
8126
8127	/** @copydoc VDIMAGEBACKEND::pfnSetComment */
8128	static DECLCALLBACK(int) vmdkSetComment(void pBackendData, const char pszComment)
8129	{
8130	LogFlowFunc(("pBackendData=%#p pszComment=\"%s\"\n", pBackendData, pszComment));
8131	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8132	int rc;
8133
8134	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8135
8136	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8137	{
8138	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8139	rc = vmdkSetImageComment(pImage, pszComment);
8140	else
8141	rc = VERR_NOT_SUPPORTED;
8142	}
8143	else
8144	rc = VERR_VD_IMAGE_READ_ONLY;
8145
8146	LogFlowFunc(("returns %Rrc\n", rc));
8147	return rc;
8148	}
8149
8150	/** @copydoc VDIMAGEBACKEND::pfnGetUuid */
8151	static DECLCALLBACK(int) vmdkGetUuid(void *pBackendData, PRTUUID pUuid)
8152	{
8153	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
8154	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8155
8156	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8157
8158	*pUuid = pImage->ImageUuid;
8159
8160	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
8161	return VINF_SUCCESS;
8162	}
8163
8164	/** @copydoc VDIMAGEBACKEND::pfnSetUuid */
8165	static DECLCALLBACK(int) vmdkSetUuid(void *pBackendData, PCRTUUID pUuid)
8166	{
8167	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
8168	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8169	int rc = VINF_SUCCESS;
8170
8171	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8172
8173	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8174	{
8175	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8176	{
8177	pImage->ImageUuid = *pUuid;
8178	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
8179	VMDK_DDB_IMAGE_UUID, pUuid);
8180	if (RT_FAILURE(rc))
8181	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
8182	N_("VMDK: error storing image UUID in descriptor in '%s'"), pImage->pszFilename);
8183	}
8184	else
8185	rc = VERR_NOT_SUPPORTED;
8186	}
8187	else
8188	rc = VERR_VD_IMAGE_READ_ONLY;
8189
8190	LogFlowFunc(("returns %Rrc\n", rc));
8191	return rc;
8192	}
8193
8194	/** @copydoc VDIMAGEBACKEND::pfnGetModificationUuid */
8195	static DECLCALLBACK(int) vmdkGetModificationUuid(void *pBackendData, PRTUUID pUuid)
8196	{
8197	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
8198	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8199
8200	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8201
8202	*pUuid = pImage->ModificationUuid;
8203
8204	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
8205	return VINF_SUCCESS;
8206	}
8207
8208	/** @copydoc VDIMAGEBACKEND::pfnSetModificationUuid */
8209	static DECLCALLBACK(int) vmdkSetModificationUuid(void *pBackendData, PCRTUUID pUuid)
8210	{
8211	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
8212	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8213	int rc = VINF_SUCCESS;
8214
8215	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8216
8217	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8218	{
8219	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8220	{
8221	/* Only touch the modification uuid if it changed. */
8222	if (RTUuidCompare(&pImage->ModificationUuid, pUuid))
8223	{
8224	pImage->ModificationUuid = *pUuid;
8225	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
8226	VMDK_DDB_MODIFICATION_UUID, pUuid);
8227	if (RT_FAILURE(rc))
8228	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing modification UUID in descriptor in '%s'"), pImage->pszFilename);
8229	}
8230	}
8231	else
8232	rc = VERR_NOT_SUPPORTED;
8233	}
8234	else
8235	rc = VERR_VD_IMAGE_READ_ONLY;
8236
8237	LogFlowFunc(("returns %Rrc\n", rc));
8238	return rc;
8239	}
8240
8241	/** @copydoc VDIMAGEBACKEND::pfnGetParentUuid */
8242	static DECLCALLBACK(int) vmdkGetParentUuid(void *pBackendData, PRTUUID pUuid)
8243	{
8244	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
8245	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8246
8247	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8248
8249	*pUuid = pImage->ParentUuid;
8250
8251	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
8252	return VINF_SUCCESS;
8253	}
8254
8255	/** @copydoc VDIMAGEBACKEND::pfnSetParentUuid */
8256	static DECLCALLBACK(int) vmdkSetParentUuid(void *pBackendData, PCRTUUID pUuid)
8257	{
8258	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
8259	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8260	int rc = VINF_SUCCESS;
8261
8262	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8263
8264	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8265	{
8266	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8267	{
8268	pImage->ParentUuid = *pUuid;
8269	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
8270	VMDK_DDB_PARENT_UUID, pUuid);
8271	if (RT_FAILURE(rc))
8272	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
8273	N_("VMDK: error storing parent image UUID in descriptor in '%s'"), pImage->pszFilename);
8274	}
8275	else
8276	rc = VERR_NOT_SUPPORTED;
8277	}
8278	else
8279	rc = VERR_VD_IMAGE_READ_ONLY;
8280
8281	LogFlowFunc(("returns %Rrc\n", rc));
8282	return rc;
8283	}
8284
8285	/** @copydoc VDIMAGEBACKEND::pfnGetParentModificationUuid */
8286	static DECLCALLBACK(int) vmdkGetParentModificationUuid(void *pBackendData, PRTUUID pUuid)
8287	{
8288	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
8289	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8290
8291	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8292
8293	*pUuid = pImage->ParentModificationUuid;
8294
8295	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
8296	return VINF_SUCCESS;
8297	}
8298
8299	/** @copydoc VDIMAGEBACKEND::pfnSetParentModificationUuid */
8300	static DECLCALLBACK(int) vmdkSetParentModificationUuid(void *pBackendData, PCRTUUID pUuid)
8301	{
8302	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
8303	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8304	int rc = VINF_SUCCESS;
8305
8306	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8307
8308	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8309	{
8310	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8311	{
8312	pImage->ParentModificationUuid = *pUuid;
8313	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
8314	VMDK_DDB_PARENT_MODIFICATION_UUID, pUuid);
8315	if (RT_FAILURE(rc))
8316	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent image UUID in descriptor in '%s'"), pImage->pszFilename);
8317	}
8318	else
8319	rc = VERR_NOT_SUPPORTED;
8320	}
8321	else
8322	rc = VERR_VD_IMAGE_READ_ONLY;
8323
8324	LogFlowFunc(("returns %Rrc\n", rc));
8325	return rc;
8326	}
8327
8328	/** @copydoc VDIMAGEBACKEND::pfnDump */
8329	static DECLCALLBACK(void) vmdkDump(void *pBackendData)
8330	{
8331	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8332
8333	AssertPtrReturnVoid(pImage);
8334	vdIfErrorMessage(pImage->pIfError, "Header: Geometry PCHS=%u/%u/%u LCHS=%u/%u/%u cbSector=%llu\n",
8335	pImage->PCHSGeometry.cCylinders, pImage->PCHSGeometry.cHeads, pImage->PCHSGeometry.cSectors,
8336	pImage->LCHSGeometry.cCylinders, pImage->LCHSGeometry.cHeads, pImage->LCHSGeometry.cSectors,
8337	VMDK_BYTE2SECTOR(pImage->cbSize));
8338	vdIfErrorMessage(pImage->pIfError, "Header: uuidCreation={%RTuuid}\n", &pImage->ImageUuid);
8339	vdIfErrorMessage(pImage->pIfError, "Header: uuidModification={%RTuuid}\n", &pImage->ModificationUuid);
8340	vdIfErrorMessage(pImage->pIfError, "Header: uuidParent={%RTuuid}\n", &pImage->ParentUuid);
8341	vdIfErrorMessage(pImage->pIfError, "Header: uuidParentModification={%RTuuid}\n", &pImage->ParentModificationUuid);
8342	}
8343
8344
8345	/**
8346	* Returns the size, in bytes, of the sparse extent overhead for
8347	* the number of desired total sectors and based on the current
8348	* sectors of the extent.
8349	*
8350	* @returns uint64_t size of new overhead in bytes.
8351	* @param pExtent VMDK extent instance.
8352	* @param cSectorsNew Number of desired total sectors.
8353	*/
8354	static uint64_t vmdkGetNewOverhead(PVMDKEXTENT pExtent, uint64_t cSectorsNew)
8355	{
8356	uint64_t cNewDirEntries = cSectorsNew / pExtent->cSectorsPerGDE;
8357	if (cSectorsNew % pExtent->cSectorsPerGDE)
8358	cNewDirEntries++;
8359
8360	size_t cbNewGD = cNewDirEntries * sizeof(uint32_t);
8361	uint64_t cbNewDirSize = RT_ALIGN_64(cbNewGD, 512);
8362	uint64_t cbNewAllTablesSize = RT_ALIGN_64(cNewDirEntries * pExtent->cGTEntries * sizeof(uint32_t), 512);
8363	uint64_t cbNewOverhead = RT_ALIGN_Z(RT_MAX(pExtent->uDescriptorSector
8364	+ pExtent->cDescriptorSectors, 1)
8365	+ cbNewDirSize + cbNewAllTablesSize, 512);
8366	cbNewOverhead += cbNewDirSize + cbNewAllTablesSize;
8367	cbNewOverhead = RT_ALIGN_64(cbNewOverhead,
8368	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
8369
8370	return cbNewOverhead;
8371	}
8372
8373	/**
8374	* Internal: Replaces the size (in sectors) of an extent in the descriptor file.
8375	*
8376	* @returns VBox status code.
8377	* @param pImage VMDK image instance.
8378	* @param pExtent VMDK extent instance.
8379	* @param uLine Line number of descriptor to change.
8380	* @param cSectorsOld Existing number of sectors.
8381	* @param cSectorsNew New number of sectors.
8382	*/
8383	static int vmdkReplaceExtentSize(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, unsigned uLine, uint64_t cSectorsOld,
8384	uint64_t cSectorsNew)
8385	{
8386	char szOldExtentSectors[UINT64_MAX_BUFF_SIZE];
8387	char szNewExtentSectors[UINT64_MAX_BUFF_SIZE];
8388
8389	ssize_t cbWritten = RTStrPrintf2(szOldExtentSectors, sizeof(szOldExtentSectors), "%llu", cSectorsOld);
8390	if (cbWritten <= 0 \|\| cbWritten > (ssize_t)sizeof(szOldExtentSectors))
8391	return VERR_BUFFER_OVERFLOW;
8392
8393	cbWritten = RTStrPrintf2(szNewExtentSectors, sizeof(szNewExtentSectors), "%llu", cSectorsNew);
8394	if (cbWritten <= 0 \|\| cbWritten > (ssize_t)sizeof(szNewExtentSectors))
8395	return VERR_BUFFER_OVERFLOW;
8396
8397	char *pszNewExtentLine = vmdkStrReplace(pImage->Descriptor.aLines[uLine],
8398	szOldExtentSectors,
8399	szNewExtentSectors);
8400
8401	if (RT_UNLIKELY(!pszNewExtentLine))
8402	return VERR_INVALID_PARAMETER;
8403
8404	vmdkDescExtRemoveByLine(pImage, &pImage->Descriptor, uLine);
8405	vmdkDescExtInsert(pImage, &pImage->Descriptor,
8406	pExtent->enmAccess, cSectorsNew,
8407	pExtent->enmType, pExtent->pszBasename, pExtent->uSectorOffset);
8408
8409	RTStrFree(pszNewExtentLine);
8410	pszNewExtentLine = NULL;
8411
8412	pImage->Descriptor.fDirty = true;
8413
8414	return VINF_SUCCESS;
8415	}
8416
8417	/**
8418	* Moves sectors down to make room for new overhead.
8419	* Used for sparse extent resize.
8420	*
8421	* @returns VBox status code.
8422	* @param pImage VMDK image instance.
8423	* @param pExtent VMDK extent instance.
8424	* @param cSectorsNew Number of sectors after resize.
8425	*/
8426	static int vmdkRelocateSectorsForSparseResize(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
8427	uint64_t cSectorsNew)
8428	{
8429	int rc = VINF_SUCCESS;
8430
8431	uint64_t cbNewOverhead = vmdkGetNewOverhead(pExtent, cSectorsNew);
8432
8433	uint64_t cNewOverheadSectors = VMDK_BYTE2SECTOR(cbNewOverhead);
8434	uint64_t cOverheadSectorDiff = cNewOverheadSectors - pExtent->cOverheadSectors;
8435
8436	uint64_t cbFile = 0;
8437	rc = vdIfIoIntFileGetSize(pImage->pIfIo, pExtent->pFile->pStorage, &cbFile);
8438
8439	uint64_t uNewAppendPosition;
8440
8441	/* Calculate how many sectors need to be relocated. */
8442	unsigned cSectorsReloc = cOverheadSectorDiff;
8443	if (cbNewOverhead % VMDK_SECTOR_SIZE)
8444	cSectorsReloc++;
8445
8446	if (cSectorsReloc < pExtent->cSectors)
8447	uNewAppendPosition = RT_ALIGN_Z(cbFile + VMDK_SECTOR2BYTE(cOverheadSectorDiff), 512);
8448	else
8449	uNewAppendPosition = cbFile;
8450
8451	/*
8452	* Get the blocks we need to relocate first, they are appended to the end
8453	* of the image.
8454	*/
8455	void pvBuf = NULL, pvZero = NULL;
8456	do
8457	{
8458	/* Allocate data buffer. */
8459	pvBuf = RTMemAllocZ(VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
8460	if (!pvBuf)
8461	{
8462	rc = VERR_NO_MEMORY;
8463	break;
8464	}
8465
8466	/* Allocate buffer for overwriting with zeroes. */
8467	pvZero = RTMemAllocZ(VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
8468	if (!pvZero)
8469	{
8470	RTMemFree(pvBuf);
8471	pvBuf = NULL;
8472
8473	rc = VERR_NO_MEMORY;
8474	break;
8475	}
8476
8477	uint32_t aGTDataTmp = (uint32_t )RTMemAllocZ(sizeof(uint32_t) * pExtent->cGTEntries);
8478	if(!aGTDataTmp)
8479	{
8480	RTMemFree(pvBuf);
8481	pvBuf = NULL;
8482
8483	RTMemFree(pvZero);
8484	pvZero = NULL;
8485
8486	rc = VERR_NO_MEMORY;
8487	break;
8488	}
8489
8490	uint32_t aRGTDataTmp = (uint32_t )RTMemAllocZ(sizeof(uint32_t) * pExtent->cGTEntries);
8491	if(!aRGTDataTmp)
8492	{
8493	RTMemFree(pvBuf);
8494	pvBuf = NULL;
8495
8496	RTMemFree(pvZero);
8497	pvZero = NULL;
8498
8499	RTMemFree(aGTDataTmp);
8500	aGTDataTmp = NULL;
8501
8502	rc = VERR_NO_MEMORY;
8503	break;
8504	}
8505
8506	/* Search for overlap sector in the grain table. */
8507	for (uint32_t idxGD = 0; idxGD < pExtent->cGDEntries; idxGD++)
8508	{
8509	uint64_t uGTSector = pExtent->pGD[idxGD];
8510	uint64_t uRGTSector = pExtent->pRGD[idxGD];
8511
8512	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
8513	VMDK_SECTOR2BYTE(uGTSector),
8514	aGTDataTmp, sizeof(uint32_t) * pExtent->cGTEntries);
8515
8516	if (RT_FAILURE(rc))
8517	break;
8518
8519	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
8520	VMDK_SECTOR2BYTE(uRGTSector),
8521	aRGTDataTmp, sizeof(uint32_t) * pExtent->cGTEntries);
8522
8523	if (RT_FAILURE(rc))
8524	break;
8525
8526	for (uint32_t idxGT = 0; idxGT < pExtent->cGTEntries; idxGT++)
8527	{
8528	uint64_t aGTEntryLE = RT_LE2H_U64(aGTDataTmp[idxGT]);
8529	uint64_t aRGTEntryLE = RT_LE2H_U64(aRGTDataTmp[idxGT]);
8530
8531	/**
8532	* Check if grain table is valid. If not dump out with an error.
8533	* Shoudln't ever get here (given other checks) but good sanity check.
8534	*/
8535	if (aGTEntryLE != aRGTEntryLE)
8536	{
8537	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
8538	N_("VMDK: inconsistent references within grain table in '%s'"), pExtent->pszFullname);
8539	break;
8540	}
8541
8542	if (aGTEntryLE < cNewOverheadSectors
8543	&& aGTEntryLE != 0)
8544	{
8545	/* Read data and append grain to the end of the image. */
8546	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
8547	VMDK_SECTOR2BYTE(aGTEntryLE), pvBuf,
8548	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
8549	if (RT_FAILURE(rc))
8550	break;
8551
8552	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8553	uNewAppendPosition, pvBuf,
8554	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
8555	if (RT_FAILURE(rc))
8556	break;
8557
8558	/* Zero out the old block area. */
8559	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8560	VMDK_SECTOR2BYTE(aGTEntryLE), pvZero,
8561	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
8562	if (RT_FAILURE(rc))
8563	break;
8564
8565	/* Write updated grain tables to file */
8566	aGTDataTmp[idxGT] = VMDK_BYTE2SECTOR(uNewAppendPosition);
8567	aRGTDataTmp[idxGT] = VMDK_BYTE2SECTOR(uNewAppendPosition);
8568
8569	if (memcmp(aGTDataTmp, aRGTDataTmp, sizeof(uint32_t) * pExtent->cGTEntries))
8570	{
8571	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
8572	N_("VMDK: inconsistency between grain table and backup grain table in '%s'"), pExtent->pszFullname);
8573	break;
8574	}
8575
8576	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8577	VMDK_SECTOR2BYTE(uGTSector),
8578	aGTDataTmp, sizeof(uint32_t) * pExtent->cGTEntries);
8579
8580	if (RT_FAILURE(rc))
8581	break;
8582
8583	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8584	VMDK_SECTOR2BYTE(uRGTSector),
8585	aRGTDataTmp, sizeof(uint32_t) * pExtent->cGTEntries);
8586
8587	break;
8588	}
8589	}
8590	}
8591
8592	RTMemFree(aGTDataTmp);
8593	aGTDataTmp = NULL;
8594
8595	RTMemFree(aRGTDataTmp);
8596	aRGTDataTmp = NULL;
8597
8598	if (RT_FAILURE(rc))
8599	break;
8600
8601	uNewAppendPosition += VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain);
8602	} while (0);
8603
8604	if (pvBuf)
8605	{
8606	RTMemFree(pvBuf);
8607	pvBuf = NULL;
8608	}
8609
8610	if (pvZero)
8611	{
8612	RTMemFree(pvZero);
8613	pvZero = NULL;
8614	}
8615
8616	// Update append position for extent
8617	pExtent->uAppendPosition = uNewAppendPosition;
8618
8619	return rc;
8620	}
8621
8622	/**
8623	* Resizes meta/overhead for sparse extent resize.
8624	*
8625	* @returns VBox status code.
8626	* @param pImage VMDK image instance.
8627	* @param pExtent VMDK extent instance.
8628	* @param cSectorsNew Number of sectors after resize.
8629	*/
8630	static int vmdkResizeSparseMeta(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
8631	uint64_t cSectorsNew)
8632	{
8633	int rc = VINF_SUCCESS;
8634	uint32_t cOldGDEntries = pExtent->cGDEntries;
8635
8636	uint64_t cNewDirEntries = cSectorsNew / pExtent->cSectorsPerGDE;
8637	if (cSectorsNew % pExtent->cSectorsPerGDE)
8638	cNewDirEntries++;
8639
8640	size_t cbNewGD = cNewDirEntries * sizeof(uint32_t);
8641
8642	uint64_t cbNewDirSize = RT_ALIGN_64(cbNewGD, 512);
8643	uint64_t cbCurrDirSize = RT_ALIGN_64(pExtent->cGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE, 512);
8644	uint64_t cDirSectorDiff = VMDK_BYTE2SECTOR(cbNewDirSize - cbCurrDirSize);
8645
8646	uint64_t cbNewAllTablesSize = RT_ALIGN_64(cNewDirEntries * pExtent->cGTEntries * sizeof(uint32_t), 512);
8647	uint64_t cbCurrAllTablesSize = RT_ALIGN_64(pExtent->cGDEntries * VMDK_GRAIN_TABLE_SIZE, 512);
8648	uint64_t cTableSectorDiff = VMDK_BYTE2SECTOR(cbNewAllTablesSize - cbCurrAllTablesSize);
8649
8650	uint64_t cbNewOverhead = vmdkGetNewOverhead(pExtent, cSectorsNew);
8651	uint64_t cNewOverheadSectors = VMDK_BYTE2SECTOR(cbNewOverhead);
8652	uint64_t cOverheadSectorDiff = cNewOverheadSectors - pExtent->cOverheadSectors;
8653
8654	/*
8655	* Get the blocks we need to relocate first, they are appended to the end
8656	* of the image.
8657	*/
8658	void pvBuf = NULL, pvZero = NULL;
8659
8660	do
8661	{
8662	/* Allocate data buffer. */
8663	pvBuf = RTMemAllocZ(VMDK_GRAIN_TABLE_SIZE);
8664	if (!pvBuf)
8665	{
8666	rc = VERR_NO_MEMORY;
8667	break;
8668	}
8669
8670	/* Allocate buffer for overwriting with zeroes. */
8671	pvZero = RTMemAllocZ(VMDK_GRAIN_TABLE_SIZE);
8672	if (!pvZero)
8673	{
8674	RTMemFree(pvBuf);
8675	pvBuf = NULL;
8676
8677	rc = VERR_NO_MEMORY;
8678	break;
8679	}
8680
8681	uint32_t uGTStart = VMDK_SECTOR2BYTE(pExtent->uSectorGD) + (cOldGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8682
8683	// points to last element in the grain table
8684	uint32_t uGTTail = uGTStart + (pExtent->cGDEntries * VMDK_GRAIN_TABLE_SIZE) - VMDK_GRAIN_TABLE_SIZE;
8685	uint32_t cbGTOff = RT_ALIGN_Z(VMDK_SECTOR2BYTE(cDirSectorDiff + cTableSectorDiff + cDirSectorDiff), 512);
8686
8687	for (int i = pExtent->cGDEntries - 1; i >= 0; i--)
8688	{
8689	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
8690	uGTTail, pvBuf,
8691	VMDK_GRAIN_TABLE_SIZE);
8692	if (RT_FAILURE(rc))
8693	break;
8694
8695	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8696	RT_ALIGN_Z(uGTTail + cbGTOff, 512), pvBuf,
8697	VMDK_GRAIN_TABLE_SIZE);
8698	if (RT_FAILURE(rc))
8699	break;
8700
8701	// This overshoots when i == 0, but we don't need it anymore.
8702	uGTTail -= VMDK_GRAIN_TABLE_SIZE;
8703	}
8704
8705
8706	/* Find the end of the grain directory and start bumping everything down. Update locations of GT entries. */
8707	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
8708	VMDK_SECTOR2BYTE(pExtent->uSectorGD), pvBuf,
8709	pExtent->cGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8710	if (RT_FAILURE(rc))
8711	break;
8712
8713	int * tmpBuf = (int *)pvBuf;
8714
8715	for (uint32_t i = 0; i < pExtent->cGDEntries; i++)
8716	{
8717	tmpBuf[i] = tmpBuf[i] + VMDK_BYTE2SECTOR(cbGTOff);
8718	pExtent->pGD[i] = pExtent->pGD[i] + VMDK_BYTE2SECTOR(cbGTOff);
8719	}
8720
8721	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8722	RT_ALIGN_Z(VMDK_SECTOR2BYTE(pExtent->uSectorGD + cTableSectorDiff + cDirSectorDiff), 512), pvBuf,
8723	pExtent->cGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8724	if (RT_FAILURE(rc))
8725	break;
8726
8727	pExtent->uSectorGD = pExtent->uSectorGD + cDirSectorDiff + cTableSectorDiff;
8728
8729	/* Repeat both steps with the redundant grain table/directory. */
8730
8731	uint32_t uRGTStart = VMDK_SECTOR2BYTE(pExtent->uSectorRGD) + (cOldGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8732
8733	// points to last element in the grain table
8734	uint32_t uRGTTail = uRGTStart + (pExtent->cGDEntries * VMDK_GRAIN_TABLE_SIZE) - VMDK_GRAIN_TABLE_SIZE;
8735	uint32_t cbRGTOff = RT_ALIGN_Z(VMDK_SECTOR2BYTE(cDirSectorDiff), 512);
8736
8737	for (int i = pExtent->cGDEntries - 1; i >= 0; i--)
8738	{
8739	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
8740	uRGTTail, pvBuf,
8741	VMDK_GRAIN_TABLE_SIZE);
8742	if (RT_FAILURE(rc))
8743	break;
8744
8745	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8746	RT_ALIGN_Z(uRGTTail + cbRGTOff, 512), pvBuf,
8747	VMDK_GRAIN_TABLE_SIZE);
8748	if (RT_FAILURE(rc))
8749	break;
8750
8751	// This overshoots when i == 0, but we don't need it anymore.
8752	uRGTTail -= VMDK_GRAIN_TABLE_SIZE;
8753	}
8754
8755	/* Update locations of GT entries. */
8756	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
8757	VMDK_SECTOR2BYTE(pExtent->uSectorRGD), pvBuf,
8758	pExtent->cGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8759	if (RT_FAILURE(rc))
8760	break;
8761
8762	tmpBuf = (int *)pvBuf;
8763
8764	for (uint32_t i = 0; i < pExtent->cGDEntries; i++)
8765	{
8766	tmpBuf[i] = tmpBuf[i] + cDirSectorDiff;
8767	pExtent->pRGD[i] = pExtent->pRGD[i] + cDirSectorDiff;
8768	}
8769
8770	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8771	VMDK_SECTOR2BYTE(pExtent->uSectorRGD), pvBuf,
8772	pExtent->cGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8773	if (RT_FAILURE(rc))
8774	break;
8775
8776	pExtent->uSectorRGD = pExtent->uSectorRGD;
8777	pExtent->cOverheadSectors += cOverheadSectorDiff;
8778
8779	} while (0);
8780
8781	if (pvBuf)
8782	{
8783	RTMemFree(pvBuf);
8784	pvBuf = NULL;
8785	}
8786
8787	if (pvZero)
8788	{
8789	RTMemFree(pvZero);
8790	pvZero = NULL;
8791	}
8792
8793	pExtent->cGDEntries = cNewDirEntries;
8794
8795	/* Allocate buffer for overwriting with zeroes. */
8796	pvZero = RTMemAllocZ(VMDK_GRAIN_TABLE_SIZE);
8797	if (!pvZero)
8798	return VERR_NO_MEMORY;
8799
8800	// Allocate additional grain dir
8801	pExtent->pGD = (uint32_t ) RTMemReallocZ(pExtent->pGD, pExtent->cGDEntries sizeof(uint32_t), cbNewGD);
8802	if (RT_LIKELY(pExtent->pGD))
8803	{
8804	if (pExtent->uSectorRGD)
8805	{
8806	pExtent->pRGD = (uint32_t )RTMemReallocZ(pExtent->pRGD, pExtent->cGDEntries sizeof(uint32_t), cbNewGD);
8807	if (RT_UNLIKELY(!pExtent->pRGD))
8808	rc = VERR_NO_MEMORY;
8809	}
8810	}
8811	else
8812	return VERR_NO_MEMORY;
8813
8814
8815	uint32_t uTmpDirVal = pExtent->pGD[cOldGDEntries - 1] + VMDK_GRAIN_DIR_ENTRY_SIZE;
8816	for (uint32_t i = cOldGDEntries; i < pExtent->cGDEntries; i++)
8817	{
8818	pExtent->pGD[i] = uTmpDirVal;
8819
8820	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8821	VMDK_SECTOR2BYTE(uTmpDirVal), pvZero,
8822	VMDK_GRAIN_TABLE_SIZE);
8823
8824	if (RT_FAILURE(rc))
8825	return rc;
8826
8827	uTmpDirVal += VMDK_GRAIN_DIR_ENTRY_SIZE;
8828	}
8829
8830	uint32_t uRTmpDirVal = pExtent->pRGD[cOldGDEntries - 1] + VMDK_GRAIN_DIR_ENTRY_SIZE;
8831	for (uint32_t i = cOldGDEntries; i < pExtent->cGDEntries; i++)
8832	{
8833	pExtent->pRGD[i] = uRTmpDirVal;
8834
8835	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8836	VMDK_SECTOR2BYTE(uRTmpDirVal), pvZero,
8837	VMDK_GRAIN_TABLE_SIZE);
8838
8839	if (RT_FAILURE(rc))
8840	return rc;
8841
8842	uRTmpDirVal += VMDK_GRAIN_DIR_ENTRY_SIZE;
8843	}
8844
8845	RTMemFree(pvZero);
8846	pvZero = NULL;
8847
8848	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8849	VMDK_SECTOR2BYTE(pExtent->uSectorGD), pExtent->pGD,
8850	pExtent->cGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8851	if (RT_FAILURE(rc))
8852	return rc;
8853
8854	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
8855	VMDK_SECTOR2BYTE(pExtent->uSectorRGD), pExtent->pRGD,
8856	pExtent->cGDEntries * VMDK_GRAIN_DIR_ENTRY_SIZE);
8857	if (RT_FAILURE(rc))
8858	return rc;
8859
8860	rc = vmdkReplaceExtentSize(pImage, pExtent, pImage->Descriptor.uFirstExtent + pExtent->uExtent,
8861	pExtent->cNominalSectors, cSectorsNew);
8862	if (RT_FAILURE(rc))
8863	return rc;
8864
8865	return rc;
8866	}
8867
8868	/** @copydoc VDIMAGEBACKEND::pfnResize */
8869	static DECLCALLBACK(int) vmdkResize(void *pBackendData, uint64_t cbSize,
8870	PCVDGEOMETRY pPCHSGeometry, PCVDGEOMETRY pLCHSGeometry,
8871	unsigned uPercentStart, unsigned uPercentSpan,
8872	PVDINTERFACE pVDIfsDisk, PVDINTERFACE pVDIfsImage,
8873	PVDINTERFACE pVDIfsOperation)
8874	{
8875	RT_NOREF5(uPercentStart, uPercentSpan, pVDIfsDisk, pVDIfsImage, pVDIfsOperation);
8876
8877	// Establish variables and objects needed
8878	int rc = VINF_SUCCESS;
8879	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8880	unsigned uImageFlags = pImage->uImageFlags;
8881	PVMDKEXTENT pExtent = &pImage->pExtents[0];
8882	pExtent->fMetaDirty = true;
8883
8884	uint64_t cSectorsNew = cbSize / VMDK_SECTOR_SIZE; /** < New number of sectors in the image after the resize */
8885	if (cbSize % VMDK_SECTOR_SIZE)
8886	cSectorsNew++;
8887
8888	uint64_t cSectorsOld = pImage->cbSize / VMDK_SECTOR_SIZE; /** < Number of sectors before the resize. Only for FLAT images. */
8889	if (pImage->cbSize % VMDK_SECTOR_SIZE)
8890	cSectorsOld++;
8891	unsigned cExtents = pImage->cExtents;
8892
8893	/* Check size is within min/max bounds. */
8894	if ( !(uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
8895	&& ( !cbSize
8896	\|\| (!(uImageFlags & VD_IMAGE_FLAGS_FIXED) && cbSize >= _1T * 256 - _64K)) )
8897	return VERR_VD_INVALID_SIZE;
8898
8899	/*
8900	* Making the image smaller is not supported at the moment.
8901	*/
8902	/** @todo implement making the image smaller, it is the responsibility of
8903	* the user to know what they're doing. */
8904	if (cbSize < pImage->cbSize)
8905	rc = VERR_VD_SHRINK_NOT_SUPPORTED;
8906	else if (cbSize > pImage->cbSize)
8907	{
8908	/**
8909	* monolithicFlat. FIXED flag and not split up into 2 GB parts.
8910	*/
8911	if ((uImageFlags & VD_IMAGE_FLAGS_FIXED) && !(uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G))
8912	{
8913	/** Required space in bytes for the extent after the resize. */
8914	uint64_t cbSectorSpaceNew = cSectorsNew * VMDK_SECTOR_SIZE;
8915	pExtent = &pImage->pExtents[0];
8916
8917	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage, cbSectorSpaceNew,
8918	0 /* fFlags */, NULL,
8919	uPercentStart, uPercentSpan);
8920	if (RT_FAILURE(rc))
8921	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
8922
8923	rc = vmdkReplaceExtentSize(pImage, pExtent, pImage->Descriptor.uFirstExtent, cSectorsOld, cSectorsNew);
8924	if (RT_FAILURE(rc))
8925	return rc;
8926	}
8927
8928	/**
8929	* twoGbMaxExtentFlat. FIXED flag and SPLIT into 2 GB parts.
8930	*/
8931	if ((uImageFlags & VD_IMAGE_FLAGS_FIXED) && (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G))
8932	{
8933	/* Check to see how much space remains in last extent */
8934	bool fSpaceAvailible = false;
8935	uint64_t cLastExtentRemSectors = cSectorsOld % VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
8936	if (cLastExtentRemSectors)
8937	fSpaceAvailible = true;
8938
8939	uint64_t cSectorsNeeded = cSectorsNew - cSectorsOld;
8940
8941	/** Space remaining in current last extent file that we don't need to create another one. */
8942	if (fSpaceAvailible && cSectorsNeeded + cLastExtentRemSectors <= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE))
8943	{
8944	pExtent = &pImage->pExtents[cExtents - 1];
8945	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage,
8946	VMDK_SECTOR2BYTE(cSectorsNeeded + cLastExtentRemSectors),
8947	0 /* fFlags */, NULL, uPercentStart, uPercentSpan);
8948	if (RT_FAILURE(rc))
8949	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
8950
8951	rc = vmdkReplaceExtentSize(pImage, pExtent, pImage->Descriptor.uFirstExtent + cExtents - 1,
8952	pExtent->cNominalSectors, cSectorsNeeded + cLastExtentRemSectors);
8953	if (RT_FAILURE(rc))
8954	return rc;
8955	}
8956	//** Need more extent files to handle all the requested space. */
8957	else
8958	{
8959	if (fSpaceAvailible)
8960	{
8961	pExtent = &pImage->pExtents[cExtents - 1];
8962	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage, VMDK_2G_SPLIT_SIZE,
8963	0 /* fFlags */, NULL,
8964	uPercentStart, uPercentSpan);
8965	if (RT_FAILURE(rc))
8966	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
8967
8968	cSectorsNeeded = cSectorsNeeded - VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE) + cLastExtentRemSectors;
8969
8970	rc = vmdkReplaceExtentSize(pImage, pExtent, pImage->Descriptor.uFirstExtent + cExtents - 1,
8971	pExtent->cNominalSectors, VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE));
8972	if (RT_FAILURE(rc))
8973	return rc;
8974	}
8975
8976	unsigned cNewExtents = VMDK_SECTOR2BYTE(cSectorsNeeded) / VMDK_2G_SPLIT_SIZE;
8977	if (cNewExtents % VMDK_2G_SPLIT_SIZE \|\| cNewExtents < VMDK_2G_SPLIT_SIZE)
8978	cNewExtents++;
8979
8980	for (unsigned i = cExtents;
8981	i < cExtents + cNewExtents && cSectorsNeeded >= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
8982	i++)
8983	{
8984	rc = vmdkAddFileBackedExtent(pImage, VMDK_2G_SPLIT_SIZE);
8985	if (RT_FAILURE(rc))
8986	return rc;
8987
8988	pExtent = &pImage->pExtents[i];
8989
8990	pExtent->cSectors = VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
8991	cSectorsNeeded -= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
8992	}
8993
8994	if (cSectorsNeeded)
8995	{
8996	rc = vmdkAddFileBackedExtent(pImage, VMDK_SECTOR2BYTE(cSectorsNeeded));
8997	if (RT_FAILURE(rc))
8998	return rc;
8999	}
9000	}
9001	}
9002
9003	/**
9004	* monolithicSparse.
9005	*/
9006	if (pExtent->enmType == VMDKETYPE_HOSTED_SPARSE && !(uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G))
9007	{
9008	// 1. Calculate sectors needed for new overhead.
9009
9010	uint64_t cbNewOverhead = vmdkGetNewOverhead(pExtent, cSectorsNew);
9011	uint64_t cNewOverheadSectors = VMDK_BYTE2SECTOR(cbNewOverhead);
9012	uint64_t cOverheadSectorDiff = cNewOverheadSectors - pExtent->cOverheadSectors;
9013
9014	// 2. Relocate sectors to make room for new GD/GT, update entries in GD/GT
9015	if (cOverheadSectorDiff > 0)
9016	{
9017	if (pExtent->cSectors > 0)
9018	{
9019	/* Do the relocation. */
9020	LogFlow(("Relocating VMDK sectors\n"));
9021	rc = vmdkRelocateSectorsForSparseResize(pImage, pExtent, cSectorsNew);
9022	if (RT_FAILURE(rc))
9023	return rc;
9024
9025	rc = vmdkFlushImage(pImage, NULL);
9026	if (RT_FAILURE(rc))
9027	return rc;
9028	}
9029
9030	rc = vmdkResizeSparseMeta(pImage, pExtent, cSectorsNew);
9031	if (RT_FAILURE(rc))
9032	return rc;
9033	}
9034	}
9035
9036	/**
9037	* twoGbSparseExtent
9038	*/
9039	if (pExtent->enmType == VMDKETYPE_HOSTED_SPARSE && (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G))
9040	{
9041	/* Check to see how much space remains in last extent */
9042	bool fSpaceAvailible = false;
9043	uint64_t cLastExtentRemSectors = cSectorsOld % VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
9044	if (cLastExtentRemSectors)
9045	fSpaceAvailible = true;
9046
9047	uint64_t cSectorsNeeded = cSectorsNew - cSectorsOld;
9048
9049	if (fSpaceAvailible && cSectorsNeeded + cLastExtentRemSectors <= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE))
9050	{
9051	pExtent = &pImage->pExtents[cExtents - 1];
9052	rc = vmdkRelocateSectorsForSparseResize(pImage, pExtent, cSectorsNeeded + cLastExtentRemSectors);
9053	if (RT_FAILURE(rc))
9054	return rc;
9055
9056	rc = vmdkFlushImage(pImage, NULL);
9057	if (RT_FAILURE(rc))
9058	return rc;
9059
9060	rc = vmdkResizeSparseMeta(pImage, pExtent, cSectorsNeeded + cLastExtentRemSectors);
9061	if (RT_FAILURE(rc))
9062	return rc;
9063	}
9064	else
9065	{
9066	if (fSpaceAvailible)
9067	{
9068	pExtent = &pImage->pExtents[cExtents - 1];
9069	rc = vmdkRelocateSectorsForSparseResize(pImage, pExtent, VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE));
9070	if (RT_FAILURE(rc))
9071	return rc;
9072
9073	rc = vmdkFlushImage(pImage, NULL);
9074	if (RT_FAILURE(rc))
9075	return rc;
9076
9077	rc = vmdkResizeSparseMeta(pImage, pExtent, VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE));
9078	if (RT_FAILURE(rc))
9079	return rc;
9080
9081	cSectorsNeeded = cSectorsNeeded - VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE) + cLastExtentRemSectors;
9082	}
9083
9084	unsigned cNewExtents = VMDK_SECTOR2BYTE(cSectorsNeeded) / VMDK_2G_SPLIT_SIZE;
9085	if (cNewExtents % VMDK_2G_SPLIT_SIZE \|\| cNewExtents < VMDK_2G_SPLIT_SIZE)
9086	cNewExtents++;
9087
9088	for (unsigned i = cExtents;
9089	i < cExtents + cNewExtents && cSectorsNeeded >= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
9090	i++)
9091	{
9092	rc = vmdkAddFileBackedExtent(pImage, VMDK_2G_SPLIT_SIZE);
9093	if (RT_FAILURE(rc))
9094	return rc;
9095
9096	pExtent = &pImage->pExtents[i];
9097
9098	rc = vmdkFlushImage(pImage, NULL);
9099	if (RT_FAILURE(rc))
9100	return rc;
9101
9102	pExtent->cSectors = VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
9103	cSectorsNeeded -= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
9104	}
9105
9106	if (cSectorsNeeded)
9107	{
9108	rc = vmdkAddFileBackedExtent(pImage, VMDK_SECTOR2BYTE(cSectorsNeeded));
9109	if (RT_FAILURE(rc))
9110	return rc;
9111
9112	pExtent = &pImage->pExtents[pImage->cExtents];
9113
9114	rc = vmdkFlushImage(pImage, NULL);
9115	if (RT_FAILURE(rc))
9116	return rc;
9117	}
9118	}
9119	}
9120
9121	/* Successful resize. Update metadata */
9122	if (RT_SUCCESS(rc))
9123	{
9124	/* Update size and new block count. */
9125	pImage->cbSize = cbSize;
9126	pExtent->cNominalSectors = cSectorsNew;
9127	pExtent->cSectors = cSectorsNew;
9128
9129	/* Update geometry. */
9130	pImage->PCHSGeometry = *pPCHSGeometry;
9131	pImage->LCHSGeometry = *pLCHSGeometry;
9132	}
9133
9134	/* Update header information in base image file. */
9135	pImage->Descriptor.fDirty = true;
9136	rc = vmdkWriteDescriptor(pImage, NULL);
9137
9138	if (RT_SUCCESS(rc))
9139	rc = vmdkFlushImage(pImage, NULL);
9140	}
9141	/* Same size doesn't change the image at all. */
9142
9143	LogFlowFunc(("returns %Rrc\n", rc));
9144	return rc;
9145	}
9146
9147	const VDIMAGEBACKEND g_VmdkBackend =
9148	{
9149	/* u32Version */
9150	VD_IMGBACKEND_VERSION,
9151	/* pszBackendName */
9152	"VMDK",
9153	/* uBackendCaps */
9154	VD_CAP_UUID \| VD_CAP_CREATE_FIXED \| VD_CAP_CREATE_DYNAMIC
9155	\| VD_CAP_CREATE_SPLIT_2G \| VD_CAP_DIFF \| VD_CAP_FILE \| VD_CAP_ASYNC
9156	\| VD_CAP_VFS \| VD_CAP_PREFERRED,
9157	/* paFileExtensions */
9158	s_aVmdkFileExtensions,
9159	/* paConfigInfo */
9160	s_aVmdkConfigInfo,
9161	/* pfnProbe */
9162	vmdkProbe,
9163	/* pfnOpen */
9164	vmdkOpen,
9165	/* pfnCreate */
9166	vmdkCreate,
9167	/* pfnRename */
9168	vmdkRename,
9169	/* pfnClose */
9170	vmdkClose,
9171	/* pfnRead */
9172	vmdkRead,
9173	/* pfnWrite */
9174	vmdkWrite,
9175	/* pfnFlush */
9176	vmdkFlush,
9177	/* pfnDiscard */
9178	NULL,
9179	/* pfnGetVersion */
9180	vmdkGetVersion,
9181	/* pfnGetFileSize */
9182	vmdkGetFileSize,
9183	/* pfnGetPCHSGeometry */
9184	vmdkGetPCHSGeometry,
9185	/* pfnSetPCHSGeometry */
9186	vmdkSetPCHSGeometry,
9187	/* pfnGetLCHSGeometry */
9188	vmdkGetLCHSGeometry,
9189	/* pfnSetLCHSGeometry */
9190	vmdkSetLCHSGeometry,
9191	/* pfnQueryRegions */
9192	vmdkQueryRegions,
9193	/* pfnRegionListRelease */
9194	vmdkRegionListRelease,
9195	/* pfnGetImageFlags */
9196	vmdkGetImageFlags,
9197	/* pfnGetOpenFlags */
9198	vmdkGetOpenFlags,
9199	/* pfnSetOpenFlags */
9200	vmdkSetOpenFlags,
9201	/* pfnGetComment */
9202	vmdkGetComment,
9203	/* pfnSetComment */
9204	vmdkSetComment,
9205	/* pfnGetUuid */
9206	vmdkGetUuid,
9207	/* pfnSetUuid */
9208	vmdkSetUuid,
9209	/* pfnGetModificationUuid */
9210	vmdkGetModificationUuid,
9211	/* pfnSetModificationUuid */
9212	vmdkSetModificationUuid,
9213	/* pfnGetParentUuid */
9214	vmdkGetParentUuid,
9215	/* pfnSetParentUuid */
9216	vmdkSetParentUuid,
9217	/* pfnGetParentModificationUuid */
9218	vmdkGetParentModificationUuid,
9219	/* pfnSetParentModificationUuid */
9220	vmdkSetParentModificationUuid,
9221	/* pfnDump */
9222	vmdkDump,
9223	/* pfnGetTimestamp */
9224	NULL,
9225	/* pfnGetParentTimestamp */
9226	NULL,
9227	/* pfnSetParentTimestamp */
9228	NULL,
9229	/* pfnGetParentFilename */
9230	NULL,
9231	/* pfnSetParentFilename */
9232	NULL,
9233	/* pfnComposeLocation */
9234	genericFileComposeLocation,
9235	/* pfnComposeName */
9236	genericFileComposeName,
9237	/* pfnCompact */
9238	NULL,
9239	/* pfnResize */
9240	vmdkResize,
9241	/* pfnRepair */
9242	NULL,
9243	/* pfnTraverseMetadata */
9244	NULL,
9245	/* u32VersionEnd */
9246	VD_IMGBACKEND_VERSION
9247	};

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

source: vbox/trunk/src/VBox/Storage/VMDK.cpp@ 99960

Download in other formats: