VMDK.cpp@ 97839

Last change on this file since 97839 was 97839, checked in by vboxsync, 2 years ago
undo. bugref:8707
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 353.9 KB

Line
1	/* $Id: VMDK.cpp 97839 2022-12-20 09:48:25Z vboxsync $ */
2	/** @file
3	* VMDK disk image, core code.
4	*/
5
6	/*
7	* Copyright (C) 2006-2022 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	* @returns nothing.
1169	* @param pGD The grain directory.
1170	* @param cGDEntries Number of entries in the grain directory to convert.
1171	*/
1172	DECLINLINE(void) vmdkGrainDirectoryConvToHost(uint32_t *pGD, uint32_t cGDEntries)
1173	{
1174	uint32_t *pGDTmp = pGD;
1175
1176	for (uint32_t i = 0; i < cGDEntries; i++, pGDTmp++)
1177	pGDTmp = RT_LE2H_U32(pGDTmp);
1178	}
1179
1180	/**
1181	* Read the grain directory and allocated grain tables verifying them against
1182	* their back up copies if available.
1183	*
1184	* @returns VBox status code.
1185	* @param pImage Image instance data.
1186	* @param pExtent The VMDK extent.
1187	*/
1188	static int vmdkReadGrainDirectory(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
1189	{
1190	int rc = VINF_SUCCESS;
1191	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
1192
1193	AssertReturn(( pExtent->enmType == VMDKETYPE_HOSTED_SPARSE
1194	&& pExtent->uSectorGD != VMDK_GD_AT_END
1195	&& pExtent->uSectorRGD != VMDK_GD_AT_END), VERR_INTERNAL_ERROR);
1196
1197	rc = vmdkAllocGrainDirectory(pImage, pExtent);
1198	if (RT_SUCCESS(rc))
1199	{
1200	/* The VMDK 1.1 spec seems to talk about compressed grain directories,
1201	* but in reality they are not compressed. */
1202	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1203	VMDK_SECTOR2BYTE(pExtent->uSectorGD),
1204	pExtent->pGD, cbGD);
1205	if (RT_SUCCESS(rc))
1206	{
1207	vmdkGrainDirectoryConvToHost(pExtent->pGD, pExtent->cGDEntries);
1208
1209	if ( pExtent->uSectorRGD
1210	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_SKIP_CONSISTENCY_CHECKS))
1211	{
1212	/* The VMDK 1.1 spec seems to talk about compressed grain directories,
1213	* but in reality they are not compressed. */
1214	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1215	VMDK_SECTOR2BYTE(pExtent->uSectorRGD),
1216	pExtent->pRGD, cbGD);
1217	if (RT_SUCCESS(rc))
1218	{
1219	vmdkGrainDirectoryConvToHost(pExtent->pRGD, pExtent->cGDEntries);
1220
1221	/* Check grain table and redundant grain table for consistency. */
1222	size_t cbGT = pExtent->cGTEntries * sizeof(uint32_t);
1223	size_t cbGTBuffers = cbGT; /* Start with space for one GT. */
1224	size_t cbGTBuffersMax = _1M;
1225
1226	uint32_t pTmpGT1 = (uint32_t )RTMemAlloc(cbGTBuffers);
1227	uint32_t pTmpGT2 = (uint32_t )RTMemAlloc(cbGTBuffers);
1228
1229	if ( !pTmpGT1
1230	\|\| !pTmpGT2)
1231	rc = VERR_NO_MEMORY;
1232
1233	size_t i = 0;
1234	uint32_t *pGDTmp = pExtent->pGD;
1235	uint32_t *pRGDTmp = pExtent->pRGD;
1236
1237	/* Loop through all entries. */
1238	while (i < pExtent->cGDEntries)
1239	{
1240	uint32_t uGTStart = *pGDTmp;
1241	uint32_t uRGTStart = *pRGDTmp;
1242	size_t cbGTRead = cbGT;
1243
1244	/* If no grain table is allocated skip the entry. */
1245	if (pGDTmp == 0 && pRGDTmp == 0)
1246	{
1247	i++;
1248	continue;
1249	}
1250
1251	if (pGDTmp == 0 \|\| pRGDTmp == 0 \|\| pGDTmp == pRGDTmp)
1252	{
1253	/* Just one grain directory entry refers to a not yet allocated
1254	* grain table or both grain directory copies refer to the same
1255	* grain table. Not allowed. */
1256	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1257	N_("VMDK: inconsistent references to grain directory in '%s'"), pExtent->pszFullname);
1258	break;
1259	}
1260
1261	i++;
1262	pGDTmp++;
1263	pRGDTmp++;
1264
1265	/*
1266	* Read a few tables at once if adjacent to decrease the number
1267	* of I/O requests. Read at maximum 1MB at once.
1268	*/
1269	while ( i < pExtent->cGDEntries
1270	&& cbGTRead < cbGTBuffersMax)
1271	{
1272	/* If no grain table is allocated skip the entry. */
1273	if (pGDTmp == 0 && pRGDTmp == 0)
1274	{
1275	i++;
1276	continue;
1277	}
1278
1279	if (pGDTmp == 0 \|\| pRGDTmp == 0 \|\| pGDTmp == pRGDTmp)
1280	{
1281	/* Just one grain directory entry refers to a not yet allocated
1282	* grain table or both grain directory copies refer to the same
1283	* grain table. Not allowed. */
1284	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1285	N_("VMDK: inconsistent references to grain directory in '%s'"), pExtent->pszFullname);
1286	break;
1287	}
1288
1289	/* Check that the start offsets are adjacent.*/
1290	if ( VMDK_SECTOR2BYTE(uGTStart) + cbGTRead != VMDK_SECTOR2BYTE(*pGDTmp)
1291	\|\| VMDK_SECTOR2BYTE(uRGTStart) + cbGTRead != VMDK_SECTOR2BYTE(*pRGDTmp))
1292	break;
1293
1294	i++;
1295	pGDTmp++;
1296	pRGDTmp++;
1297	cbGTRead += cbGT;
1298	}
1299
1300	/* Increase buffers if required. */
1301	if ( RT_SUCCESS(rc)
1302	&& cbGTBuffers < cbGTRead)
1303	{
1304	uint32_t *pTmp;
1305	pTmp = (uint32_t *)RTMemRealloc(pTmpGT1, cbGTRead);
1306	if (pTmp)
1307	{
1308	pTmpGT1 = pTmp;
1309	pTmp = (uint32_t *)RTMemRealloc(pTmpGT2, cbGTRead);
1310	if (pTmp)
1311	pTmpGT2 = pTmp;
1312	else
1313	rc = VERR_NO_MEMORY;
1314	}
1315	else
1316	rc = VERR_NO_MEMORY;
1317
1318	if (rc == VERR_NO_MEMORY)
1319	{
1320	/* Reset to the old values. */
1321	rc = VINF_SUCCESS;
1322	i -= cbGTRead / cbGT;
1323	cbGTRead = cbGT;
1324
1325	/* Don't try to increase the buffer again in the next run. */
1326	cbGTBuffersMax = cbGTBuffers;
1327	}
1328	}
1329
1330	if (RT_SUCCESS(rc))
1331	{
1332	/* The VMDK 1.1 spec seems to talk about compressed grain tables,
1333	* but in reality they are not compressed. */
1334	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1335	VMDK_SECTOR2BYTE(uGTStart),
1336	pTmpGT1, cbGTRead);
1337	if (RT_FAILURE(rc))
1338	{
1339	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1340	N_("VMDK: error reading grain table in '%s'"), pExtent->pszFullname);
1341	break;
1342	}
1343	/* The VMDK 1.1 spec seems to talk about compressed grain tables,
1344	* but in reality they are not compressed. */
1345	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
1346	VMDK_SECTOR2BYTE(uRGTStart),
1347	pTmpGT2, cbGTRead);
1348	if (RT_FAILURE(rc))
1349	{
1350	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1351	N_("VMDK: error reading backup grain table in '%s'"), pExtent->pszFullname);
1352	break;
1353	}
1354	if (memcmp(pTmpGT1, pTmpGT2, cbGTRead))
1355	{
1356	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
1357	N_("VMDK: inconsistency between grain table and backup grain table in '%s'"), pExtent->pszFullname);
1358	break;
1359	}
1360	}
1361	} /* while (i < pExtent->cGDEntries) */
1362
1363	/** @todo figure out what to do for unclean VMDKs. */
1364	if (pTmpGT1)
1365	RTMemFree(pTmpGT1);
1366	if (pTmpGT2)
1367	RTMemFree(pTmpGT2);
1368	}
1369	else
1370	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1371	N_("VMDK: could not read redundant grain directory in '%s'"), pExtent->pszFullname);
1372	}
1373	}
1374	else
1375	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
1376	N_("VMDK: could not read grain directory in '%s': %Rrc"), pExtent->pszFullname, rc);
1377	}
1378
1379	if (RT_FAILURE(rc))
1380	vmdkFreeGrainDirectory(pExtent);
1381	return rc;
1382	}
1383
1384	/**
1385	* Creates a new grain directory for the given extent at the given start sector.
1386	*
1387	* @returns VBox status code.
1388	* @param pImage Image instance data.
1389	* @param pExtent The VMDK extent.
1390	* @param uStartSector Where the grain directory should be stored in the image.
1391	* @param fPreAlloc Flag whether to pre allocate the grain tables at this point.
1392	*/
1393	static int vmdkCreateGrainDirectory(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
1394	uint64_t uStartSector, bool fPreAlloc)
1395	{
1396	int rc = VINF_SUCCESS;
1397	unsigned i;
1398	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
1399	size_t cbGDRounded = RT_ALIGN_64(cbGD, 512);
1400	size_t cbGTRounded;
1401	uint64_t cbOverhead;
1402
1403	if (fPreAlloc)
1404	{
1405	cbGTRounded = RT_ALIGN_64(pExtent->cGDEntries * pExtent->cGTEntries * sizeof(uint32_t), 512);
1406	cbOverhead = VMDK_SECTOR2BYTE(uStartSector) + cbGDRounded + cbGTRounded;
1407	}
1408	else
1409	{
1410	/* Use a dummy start sector for layout computation. */
1411	if (uStartSector == VMDK_GD_AT_END)
1412	uStartSector = 1;
1413	cbGTRounded = 0;
1414	cbOverhead = VMDK_SECTOR2BYTE(uStartSector) + cbGDRounded;
1415	}
1416
1417	/* For streamOptimized extents there is only one grain directory,
1418	* and for all others take redundant grain directory into account. */
1419	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
1420	{
1421	cbOverhead = RT_ALIGN_64(cbOverhead,
1422	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
1423	}
1424	else
1425	{
1426	cbOverhead += cbGDRounded + cbGTRounded;
1427	cbOverhead = RT_ALIGN_64(cbOverhead,
1428	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
1429	rc = vdIfIoIntFileSetSize(pImage->pIfIo, pExtent->pFile->pStorage, cbOverhead);
1430	}
1431
1432	if (RT_SUCCESS(rc))
1433	{
1434	pExtent->uAppendPosition = cbOverhead;
1435	pExtent->cOverheadSectors = VMDK_BYTE2SECTOR(cbOverhead);
1436
1437	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
1438	{
1439	pExtent->uSectorRGD = 0;
1440	pExtent->uSectorGD = uStartSector;
1441	}
1442	else
1443	{
1444	pExtent->uSectorRGD = uStartSector;
1445	pExtent->uSectorGD = uStartSector + VMDK_BYTE2SECTOR(cbGDRounded + cbGTRounded);
1446	}
1447
1448	rc = vmdkAllocStreamBuffers(pImage, pExtent);
1449	if (RT_SUCCESS(rc))
1450	{
1451	rc = vmdkAllocGrainDirectory(pImage, pExtent);
1452	if ( RT_SUCCESS(rc)
1453	&& fPreAlloc)
1454	{
1455	uint32_t uGTSectorLE;
1456	uint64_t uOffsetSectors;
1457
1458	if (pExtent->pRGD)
1459	{
1460	uOffsetSectors = pExtent->uSectorRGD + VMDK_BYTE2SECTOR(cbGDRounded);
1461	for (i = 0; i < pExtent->cGDEntries; i++)
1462	{
1463	pExtent->pRGD[i] = uOffsetSectors;
1464	uGTSectorLE = RT_H2LE_U64(uOffsetSectors);
1465	/* Write the redundant grain directory entry to disk. */
1466	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
1467	VMDK_SECTOR2BYTE(pExtent->uSectorRGD) + i * sizeof(uGTSectorLE),
1468	&uGTSectorLE, sizeof(uGTSectorLE));
1469	if (RT_FAILURE(rc))
1470	{
1471	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write new redundant grain directory entry in '%s'"), pExtent->pszFullname);
1472	break;
1473	}
1474	uOffsetSectors += VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
1475	}
1476	}
1477
1478	if (RT_SUCCESS(rc))
1479	{
1480	uOffsetSectors = pExtent->uSectorGD + VMDK_BYTE2SECTOR(cbGDRounded);
1481	for (i = 0; i < pExtent->cGDEntries; i++)
1482	{
1483	pExtent->pGD[i] = uOffsetSectors;
1484	uGTSectorLE = RT_H2LE_U64(uOffsetSectors);
1485	/* Write the grain directory entry to disk. */
1486	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
1487	VMDK_SECTOR2BYTE(pExtent->uSectorGD) + i * sizeof(uGTSectorLE),
1488	&uGTSectorLE, sizeof(uGTSectorLE));
1489	if (RT_FAILURE(rc))
1490	{
1491	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write new grain directory entry in '%s'"), pExtent->pszFullname);
1492	break;
1493	}
1494	uOffsetSectors += VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
1495	}
1496	}
1497	}
1498	}
1499	}
1500
1501	if (RT_FAILURE(rc))
1502	vmdkFreeGrainDirectory(pExtent);
1503	return rc;
1504	}
1505
1506	/**
1507	* Unquotes the given string returning the result in a separate buffer.
1508	*
1509	* @returns VBox status code.
1510	* @param pImage The VMDK image state.
1511	* @param pszStr The string to unquote.
1512	* @param ppszUnquoted Where to store the return value, use RTMemTmpFree to
1513	* free.
1514	* @param ppszNext Where to store the pointer to any character following
1515	* the quoted value, optional.
1516	*/
1517	static int vmdkStringUnquote(PVMDKIMAGE pImage, const char *pszStr,
1518	char ppszUnquoted, char ppszNext)
1519	{
1520	const char *pszStart = pszStr;
1521	char *pszQ;
1522	char *pszUnquoted;
1523
1524	/* Skip over whitespace. */
1525	while (pszStr == ' ' \|\| pszStr == '\t')
1526	pszStr++;
1527
1528	if (*pszStr != '"')
1529	{
1530	pszQ = (char *)pszStr;
1531	while (pszQ && pszQ != ' ' && *pszQ != '\t')
1532	pszQ++;
1533	}
1534	else
1535	{
1536	pszStr++;
1537	pszQ = (char *)strchr(pszStr, '"');
1538	if (pszQ == NULL)
1539	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrectly quoted value in descriptor in '%s' (raw value %s)"),
1540	pImage->pszFilename, pszStart);
1541	}
1542
1543	pszUnquoted = (char *)RTMemTmpAlloc(pszQ - pszStr + 1);
1544	if (!pszUnquoted)
1545	return VERR_NO_MEMORY;
1546	memcpy(pszUnquoted, pszStr, pszQ - pszStr);
1547	pszUnquoted[pszQ - pszStr] = '\0';
1548	*ppszUnquoted = pszUnquoted;
1549	if (ppszNext)
1550	*ppszNext = pszQ + 1;
1551	return VINF_SUCCESS;
1552	}
1553
1554	static int vmdkDescInitStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1555	const char *pszLine)
1556	{
1557	char *pEnd = pDescriptor->aLines[pDescriptor->cLines];
1558	ssize_t cbDiff = strlen(pszLine) + 1;
1559
1560	if ( pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1
1561	&& pEnd - pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff)
1562	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1563
1564	memcpy(pEnd, pszLine, cbDiff);
1565	pDescriptor->cLines++;
1566	pDescriptor->aLines[pDescriptor->cLines] = pEnd + cbDiff;
1567	pDescriptor->fDirty = true;
1568
1569	return VINF_SUCCESS;
1570	}
1571
1572	static bool vmdkDescGetStr(PVMDKDESCRIPTOR pDescriptor, unsigned uStart,
1573	const char pszKey, const char *ppszValue)
1574	{
1575	size_t cbKey = strlen(pszKey);
1576	const char *pszValue;
1577
1578	while (uStart != 0)
1579	{
1580	if (!strncmp(pDescriptor->aLines[uStart], pszKey, cbKey))
1581	{
1582	/* Key matches, check for a '=' (preceded by whitespace). */
1583	pszValue = pDescriptor->aLines[uStart] + cbKey;
1584	while (pszValue == ' ' \|\| pszValue == '\t')
1585	pszValue++;
1586	if (*pszValue == '=')
1587	{
1588	*ppszValue = pszValue + 1;
1589	break;
1590	}
1591	}
1592	uStart = pDescriptor->aNextLines[uStart];
1593	}
1594	return !!uStart;
1595	}
1596
1597	static int vmdkDescSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1598	unsigned uStart,
1599	const char pszKey, const char pszValue)
1600	{
1601	char pszTmp = NULL; / (MSC naturally cannot figure this isn't used uninitialized) */
1602	size_t cbKey = strlen(pszKey);
1603	unsigned uLast = 0;
1604
1605	while (uStart != 0)
1606	{
1607	if (!strncmp(pDescriptor->aLines[uStart], pszKey, cbKey))
1608	{
1609	/* Key matches, check for a '=' (preceded by whitespace). */
1610	pszTmp = pDescriptor->aLines[uStart] + cbKey;
1611	while (pszTmp == ' ' \|\| pszTmp == '\t')
1612	pszTmp++;
1613	if (*pszTmp == '=')
1614	{
1615	pszTmp++;
1616	/** @todo r=bird: Doesn't skipping trailing blanks here just cause unecessary
1617	* bloat and potentially out of space error? */
1618	while (pszTmp == ' ' \|\| pszTmp == '\t')
1619	pszTmp++;
1620	break;
1621	}
1622	}
1623	if (!pDescriptor->aNextLines[uStart])
1624	uLast = uStart;
1625	uStart = pDescriptor->aNextLines[uStart];
1626	}
1627	if (uStart)
1628	{
1629	if (pszValue)
1630	{
1631	/* Key already exists, replace existing value. */
1632	size_t cbOldVal = strlen(pszTmp);
1633	size_t cbNewVal = strlen(pszValue);
1634	ssize_t cbDiff = cbNewVal - cbOldVal;
1635	/* Check for buffer overflow. */
1636	if ( pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[0]
1637	> (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff)
1638	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1639
1640	memmove(pszTmp + cbNewVal, pszTmp + cbOldVal,
1641	pDescriptor->aLines[pDescriptor->cLines] - pszTmp - cbOldVal);
1642	memcpy(pszTmp, pszValue, cbNewVal + 1);
1643	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1644	pDescriptor->aLines[i] += cbDiff;
1645	}
1646	else
1647	{
1648	memmove(pDescriptor->aLines[uStart], pDescriptor->aLines[uStart+1],
1649	pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[uStart+1] + 1);
1650	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1651	{
1652	pDescriptor->aLines[i-1] = pDescriptor->aLines[i];
1653	if (pDescriptor->aNextLines[i])
1654	pDescriptor->aNextLines[i-1] = pDescriptor->aNextLines[i] - 1;
1655	else
1656	pDescriptor->aNextLines[i-1] = 0;
1657	}
1658	pDescriptor->cLines--;
1659	/* Adjust starting line numbers of following descriptor sections. */
1660	if (uStart < pDescriptor->uFirstExtent)
1661	pDescriptor->uFirstExtent--;
1662	if (uStart < pDescriptor->uFirstDDB)
1663	pDescriptor->uFirstDDB--;
1664	}
1665	}
1666	else
1667	{
1668	/* Key doesn't exist, append after the last entry in this category. */
1669	if (!pszValue)
1670	{
1671	/* Key doesn't exist, and it should be removed. Simply a no-op. */
1672	return VINF_SUCCESS;
1673	}
1674	cbKey = strlen(pszKey);
1675	size_t cbValue = strlen(pszValue);
1676	ssize_t cbDiff = cbKey + 1 + cbValue + 1;
1677	/* Check for buffer overflow. */
1678	if ( (pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1)
1679	\|\| ( pDescriptor->aLines[pDescriptor->cLines]
1680	- pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff))
1681	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1682	for (unsigned i = pDescriptor->cLines + 1; i > uLast + 1; i--)
1683	{
1684	pDescriptor->aLines[i] = pDescriptor->aLines[i - 1];
1685	if (pDescriptor->aNextLines[i - 1])
1686	pDescriptor->aNextLines[i] = pDescriptor->aNextLines[i - 1] + 1;
1687	else
1688	pDescriptor->aNextLines[i] = 0;
1689	}
1690	uStart = uLast + 1;
1691	pDescriptor->aNextLines[uLast] = uStart;
1692	pDescriptor->aNextLines[uStart] = 0;
1693	pDescriptor->cLines++;
1694	pszTmp = pDescriptor->aLines[uStart];
1695	memmove(pszTmp + cbDiff, pszTmp,
1696	pDescriptor->aLines[pDescriptor->cLines] - pszTmp);
1697	memcpy(pDescriptor->aLines[uStart], pszKey, cbKey);
1698	pDescriptor->aLines[uStart][cbKey] = '=';
1699	memcpy(pDescriptor->aLines[uStart] + cbKey + 1, pszValue, cbValue + 1);
1700	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1701	pDescriptor->aLines[i] += cbDiff;
1702
1703	/* Adjust starting line numbers of following descriptor sections. */
1704	if (uStart <= pDescriptor->uFirstExtent)
1705	pDescriptor->uFirstExtent++;
1706	if (uStart <= pDescriptor->uFirstDDB)
1707	pDescriptor->uFirstDDB++;
1708	}
1709	pDescriptor->fDirty = true;
1710	return VINF_SUCCESS;
1711	}
1712
1713	static int vmdkDescBaseGetU32(PVMDKDESCRIPTOR pDescriptor, const char *pszKey,
1714	uint32_t *puValue)
1715	{
1716	const char *pszValue;
1717
1718	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDesc, pszKey,
1719	&pszValue))
1720	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1721	return RTStrToUInt32Ex(pszValue, NULL, 10, puValue);
1722	}
1723
1724	/**
1725	* Returns the value of the given key as a string allocating the necessary memory.
1726	*
1727	* @returns VBox status code.
1728	* @retval VERR_VD_VMDK_VALUE_NOT_FOUND if the value could not be found.
1729	* @param pImage The VMDK image state.
1730	* @param pDescriptor The descriptor to fetch the value from.
1731	* @param pszKey The key to get the value from.
1732	* @param ppszValue Where to store the return value, use RTMemTmpFree to
1733	* free.
1734	*/
1735	static int vmdkDescBaseGetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1736	const char pszKey, char *ppszValue)
1737	{
1738	const char *pszValue;
1739	char *pszValueUnquoted;
1740
1741	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDesc, pszKey,
1742	&pszValue))
1743	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1744	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1745	if (RT_FAILURE(rc))
1746	return rc;
1747	*ppszValue = pszValueUnquoted;
1748	return rc;
1749	}
1750
1751	static int vmdkDescBaseSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1752	const char pszKey, const char pszValue)
1753	{
1754	char *pszValueQuoted;
1755
1756	RTStrAPrintf(&pszValueQuoted, "\"%s\"", pszValue);
1757	if (!pszValueQuoted)
1758	return VERR_NO_STR_MEMORY;
1759	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc, pszKey,
1760	pszValueQuoted);
1761	RTStrFree(pszValueQuoted);
1762	return rc;
1763	}
1764
1765	static void vmdkDescExtRemoveDummy(PVMDKIMAGE pImage,
1766	PVMDKDESCRIPTOR pDescriptor)
1767	{
1768	RT_NOREF1(pImage);
1769	unsigned uEntry = pDescriptor->uFirstExtent;
1770	ssize_t cbDiff;
1771
1772	if (!uEntry)
1773	return;
1774
1775	cbDiff = strlen(pDescriptor->aLines[uEntry]) + 1;
1776	/* Move everything including \0 in the entry marking the end of buffer. */
1777	memmove(pDescriptor->aLines[uEntry], pDescriptor->aLines[uEntry + 1],
1778	pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[uEntry + 1] + 1);
1779	for (unsigned i = uEntry + 1; i <= pDescriptor->cLines; i++)
1780	{
1781	pDescriptor->aLines[i - 1] = pDescriptor->aLines[i] - cbDiff;
1782	if (pDescriptor->aNextLines[i])
1783	pDescriptor->aNextLines[i - 1] = pDescriptor->aNextLines[i] - 1;
1784	else
1785	pDescriptor->aNextLines[i - 1] = 0;
1786	}
1787	pDescriptor->cLines--;
1788	if (pDescriptor->uFirstDDB)
1789	pDescriptor->uFirstDDB--;
1790
1791	return;
1792	}
1793
1794	static int vmdkDescExtInsert(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1795	VMDKACCESS enmAccess, uint64_t cNominalSectors,
1796	VMDKETYPE enmType, const char *pszBasename,
1797	uint64_t uSectorOffset)
1798	{
1799	static const char *apszAccess[] = { "NOACCESS", "RDONLY", "RW" };
1800	static const char *apszType[] = { "", "SPARSE", "FLAT", "ZERO", "VMFS" };
1801	char *pszTmp;
1802	unsigned uStart = pDescriptor->uFirstExtent, uLast = 0;
1803	char szExt[1024];
1804	ssize_t cbDiff;
1805
1806	Assert((unsigned)enmAccess < RT_ELEMENTS(apszAccess));
1807	Assert((unsigned)enmType < RT_ELEMENTS(apszType));
1808
1809	/* Find last entry in extent description. */
1810	while (uStart)
1811	{
1812	if (!pDescriptor->aNextLines[uStart])
1813	uLast = uStart;
1814	uStart = pDescriptor->aNextLines[uStart];
1815	}
1816
1817	if (enmType == VMDKETYPE_ZERO)
1818	{
1819	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s ", apszAccess[enmAccess],
1820	cNominalSectors, apszType[enmType]);
1821	}
1822	else if (enmType == VMDKETYPE_FLAT)
1823	{
1824	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s \"%s\" %llu",
1825	apszAccess[enmAccess], cNominalSectors,
1826	apszType[enmType], pszBasename, uSectorOffset);
1827	}
1828	else
1829	{
1830	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s \"%s\"",
1831	apszAccess[enmAccess], cNominalSectors,
1832	apszType[enmType], pszBasename);
1833	}
1834	cbDiff = strlen(szExt) + 1;
1835
1836	/* Check for buffer overflow. */
1837	if ( (pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1)
1838	\|\| ( pDescriptor->aLines[pDescriptor->cLines]
1839	- pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff))
1840	return vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1841
1842	for (unsigned i = pDescriptor->cLines + 1; i > uLast + 1; i--)
1843	{
1844	pDescriptor->aLines[i] = pDescriptor->aLines[i - 1];
1845	if (pDescriptor->aNextLines[i - 1])
1846	pDescriptor->aNextLines[i] = pDescriptor->aNextLines[i - 1] + 1;
1847	else
1848	pDescriptor->aNextLines[i] = 0;
1849	}
1850	uStart = uLast + 1;
1851	pDescriptor->aNextLines[uLast] = uStart;
1852	pDescriptor->aNextLines[uStart] = 0;
1853	pDescriptor->cLines++;
1854	pszTmp = pDescriptor->aLines[uStart];
1855	memmove(pszTmp + cbDiff, pszTmp,
1856	pDescriptor->aLines[pDescriptor->cLines] - pszTmp);
1857	memcpy(pDescriptor->aLines[uStart], szExt, cbDiff);
1858	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1859	pDescriptor->aLines[i] += cbDiff;
1860
1861	/* Adjust starting line numbers of following descriptor sections. */
1862	if (uStart <= pDescriptor->uFirstDDB)
1863	pDescriptor->uFirstDDB++;
1864
1865	pDescriptor->fDirty = true;
1866	return VINF_SUCCESS;
1867	}
1868
1869	/**
1870	* Returns the value of the given key from the DDB as a string allocating
1871	* the necessary memory.
1872	*
1873	* @returns VBox status code.
1874	* @retval VERR_VD_VMDK_VALUE_NOT_FOUND if the value could not be found.
1875	* @param pImage The VMDK image state.
1876	* @param pDescriptor The descriptor to fetch the value from.
1877	* @param pszKey The key to get the value from.
1878	* @param ppszValue Where to store the return value, use RTMemTmpFree to
1879	* free.
1880	*/
1881	static int vmdkDescDDBGetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1882	const char pszKey, char *ppszValue)
1883	{
1884	const char *pszValue;
1885	char *pszValueUnquoted;
1886
1887	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1888	&pszValue))
1889	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1890	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1891	if (RT_FAILURE(rc))
1892	return rc;
1893	*ppszValue = pszValueUnquoted;
1894	return rc;
1895	}
1896
1897	static int vmdkDescDDBGetU32(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1898	const char pszKey, uint32_t puValue)
1899	{
1900	const char *pszValue;
1901	char *pszValueUnquoted;
1902
1903	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1904	&pszValue))
1905	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1906	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1907	if (RT_FAILURE(rc))
1908	return rc;
1909	rc = RTStrToUInt32Ex(pszValueUnquoted, NULL, 10, puValue);
1910	RTMemTmpFree(pszValueUnquoted);
1911	return rc;
1912	}
1913
1914	static int vmdkDescDDBGetUuid(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1915	const char *pszKey, PRTUUID pUuid)
1916	{
1917	const char *pszValue;
1918	char *pszValueUnquoted;
1919
1920	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1921	&pszValue))
1922	return VERR_VD_VMDK_VALUE_NOT_FOUND;
1923	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1924	if (RT_FAILURE(rc))
1925	return rc;
1926	rc = RTUuidFromStr(pUuid, pszValueUnquoted);
1927	RTMemTmpFree(pszValueUnquoted);
1928	return rc;
1929	}
1930
1931	static int vmdkDescDDBSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1932	const char pszKey, const char pszVal)
1933	{
1934	int rc;
1935	char *pszValQuoted;
1936
1937	if (pszVal)
1938	{
1939	RTStrAPrintf(&pszValQuoted, "\"%s\"", pszVal);
1940	if (!pszValQuoted)
1941	return VERR_NO_STR_MEMORY;
1942	}
1943	else
1944	pszValQuoted = NULL;
1945	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1946	pszValQuoted);
1947	if (pszValQuoted)
1948	RTStrFree(pszValQuoted);
1949	return rc;
1950	}
1951
1952	static int vmdkDescDDBSetUuid(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1953	const char *pszKey, PCRTUUID pUuid)
1954	{
1955	char *pszUuid;
1956
1957	RTStrAPrintf(&pszUuid, "\"%RTuuid\"", pUuid);
1958	if (!pszUuid)
1959	return VERR_NO_STR_MEMORY;
1960	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1961	pszUuid);
1962	RTStrFree(pszUuid);
1963	return rc;
1964	}
1965
1966	static int vmdkDescDDBSetU32(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1967	const char *pszKey, uint32_t uValue)
1968	{
1969	char *pszValue;
1970
1971	RTStrAPrintf(&pszValue, "\"%d\"", uValue);
1972	if (!pszValue)
1973	return VERR_NO_STR_MEMORY;
1974	int rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1975	pszValue);
1976	RTStrFree(pszValue);
1977	return rc;
1978	}
1979
1980	/**
1981	* Splits the descriptor data into individual lines checking for correct line
1982	* endings and descriptor size.
1983	*
1984	* @returns VBox status code.
1985	* @param pImage The image instance.
1986	* @param pDesc The descriptor.
1987	* @param pszTmp The raw descriptor data from the image.
1988	*/
1989	static int vmdkDescSplitLines(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDesc, char *pszTmp)
1990	{
1991	unsigned cLine = 0;
1992	int rc = VINF_SUCCESS;
1993
1994	while ( RT_SUCCESS(rc)
1995	&& *pszTmp != '\0')
1996	{
1997	pDesc->aLines[cLine++] = pszTmp;
1998	if (cLine >= VMDK_DESCRIPTOR_LINES_MAX)
1999	{
2000	vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
2001	rc = VERR_VD_VMDK_INVALID_HEADER;
2002	break;
2003	}
2004
2005	while (pszTmp != '\0' && pszTmp != '\n')
2006	{
2007	if (*pszTmp == '\r')
2008	{
2009	if (*(pszTmp + 1) != '\n')
2010	{
2011	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: unsupported end of line in descriptor in '%s'"), pImage->pszFilename);
2012	break;
2013	}
2014	else
2015	{
2016	/* Get rid of CR character. */
2017	*pszTmp = '\0';
2018	}
2019	}
2020	pszTmp++;
2021	}
2022
2023	if (RT_FAILURE(rc))
2024	break;
2025
2026	/* Get rid of LF character. */
2027	if (*pszTmp == '\n')
2028	{
2029	*pszTmp = '\0';
2030	pszTmp++;
2031	}
2032	}
2033
2034	if (RT_SUCCESS(rc))
2035	{
2036	pDesc->cLines = cLine;
2037	/* Pointer right after the end of the used part of the buffer. */
2038	pDesc->aLines[cLine] = pszTmp;
2039	}
2040
2041	return rc;
2042	}
2043
2044	static int vmdkPreprocessDescriptor(PVMDKIMAGE pImage, char *pDescData,
2045	size_t cbDescData, PVMDKDESCRIPTOR pDescriptor)
2046	{
2047	pDescriptor->cbDescAlloc = cbDescData;
2048	int rc = vmdkDescSplitLines(pImage, pDescriptor, pDescData);
2049	if (RT_SUCCESS(rc))
2050	{
2051	if ( strcmp(pDescriptor->aLines[0], "# Disk DescriptorFile")
2052	&& strcmp(pDescriptor->aLines[0], "# Disk Descriptor File")
2053	&& strcmp(pDescriptor->aLines[0], "#Disk Descriptor File")
2054	&& strcmp(pDescriptor->aLines[0], "#Disk DescriptorFile"))
2055	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2056	N_("VMDK: descriptor does not start as expected in '%s'"), pImage->pszFilename);
2057	else
2058	{
2059	unsigned uLastNonEmptyLine = 0;
2060
2061	/* Initialize those, because we need to be able to reopen an image. */
2062	pDescriptor->uFirstDesc = 0;
2063	pDescriptor->uFirstExtent = 0;
2064	pDescriptor->uFirstDDB = 0;
2065	for (unsigned i = 0; i < pDescriptor->cLines; i++)
2066	{
2067	if (pDescriptor->aLines[i] != '#' && pDescriptor->aLines[i] != '\0')
2068	{
2069	if ( !strncmp(pDescriptor->aLines[i], "RW", 2)
2070	\|\| !strncmp(pDescriptor->aLines[i], "RDONLY", 6)
2071	\|\| !strncmp(pDescriptor->aLines[i], "NOACCESS", 8) )
2072	{
2073	/* An extent descriptor. */
2074	if (!pDescriptor->uFirstDesc \|\| pDescriptor->uFirstDDB)
2075	{
2076	/* Incorrect ordering of entries. */
2077	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2078	N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
2079	break;
2080	}
2081	if (!pDescriptor->uFirstExtent)
2082	{
2083	pDescriptor->uFirstExtent = i;
2084	uLastNonEmptyLine = 0;
2085	}
2086	}
2087	else if (!strncmp(pDescriptor->aLines[i], "ddb.", 4))
2088	{
2089	/* A disk database entry. */
2090	if (!pDescriptor->uFirstDesc \|\| !pDescriptor->uFirstExtent)
2091	{
2092	/* Incorrect ordering of entries. */
2093	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2094	N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
2095	break;
2096	}
2097	if (!pDescriptor->uFirstDDB)
2098	{
2099	pDescriptor->uFirstDDB = i;
2100	uLastNonEmptyLine = 0;
2101	}
2102	}
2103	else
2104	{
2105	/* A normal entry. */
2106	if (pDescriptor->uFirstExtent \|\| pDescriptor->uFirstDDB)
2107	{
2108	/* Incorrect ordering of entries. */
2109	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2110	N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
2111	break;
2112	}
2113	if (!pDescriptor->uFirstDesc)
2114	{
2115	pDescriptor->uFirstDesc = i;
2116	uLastNonEmptyLine = 0;
2117	}
2118	}
2119	if (uLastNonEmptyLine)
2120	pDescriptor->aNextLines[uLastNonEmptyLine] = i;
2121	uLastNonEmptyLine = i;
2122	}
2123	}
2124	}
2125	}
2126
2127	return rc;
2128	}
2129
2130	static int vmdkDescSetPCHSGeometry(PVMDKIMAGE pImage,
2131	PCVDGEOMETRY pPCHSGeometry)
2132	{
2133	int rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2134	VMDK_DDB_GEO_PCHS_CYLINDERS,
2135	pPCHSGeometry->cCylinders);
2136	if (RT_FAILURE(rc))
2137	return rc;
2138	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2139	VMDK_DDB_GEO_PCHS_HEADS,
2140	pPCHSGeometry->cHeads);
2141	if (RT_FAILURE(rc))
2142	return rc;
2143	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2144	VMDK_DDB_GEO_PCHS_SECTORS,
2145	pPCHSGeometry->cSectors);
2146	return rc;
2147	}
2148
2149	static int vmdkDescSetLCHSGeometry(PVMDKIMAGE pImage,
2150	PCVDGEOMETRY pLCHSGeometry)
2151	{
2152	int rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2153	VMDK_DDB_GEO_LCHS_CYLINDERS,
2154	pLCHSGeometry->cCylinders);
2155	if (RT_FAILURE(rc))
2156	return rc;
2157	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2158	VMDK_DDB_GEO_LCHS_HEADS,
2159
2160	pLCHSGeometry->cHeads);
2161	if (RT_FAILURE(rc))
2162	return rc;
2163	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
2164	VMDK_DDB_GEO_LCHS_SECTORS,
2165	pLCHSGeometry->cSectors);
2166	return rc;
2167	}
2168
2169	static int vmdkCreateDescriptor(PVMDKIMAGE pImage, char *pDescData,
2170	size_t cbDescData, PVMDKDESCRIPTOR pDescriptor)
2171	{
2172	pDescriptor->uFirstDesc = 0;
2173	pDescriptor->uFirstExtent = 0;
2174	pDescriptor->uFirstDDB = 0;
2175	pDescriptor->cLines = 0;
2176	pDescriptor->cbDescAlloc = cbDescData;
2177	pDescriptor->fDirty = false;
2178	pDescriptor->aLines[pDescriptor->cLines] = pDescData;
2179	memset(pDescriptor->aNextLines, '\0', sizeof(pDescriptor->aNextLines));
2180
2181	int rc = vmdkDescInitStr(pImage, pDescriptor, "# Disk DescriptorFile");
2182	if (RT_SUCCESS(rc))
2183	rc = vmdkDescInitStr(pImage, pDescriptor, "version=1");
2184	if (RT_SUCCESS(rc))
2185	{
2186	pDescriptor->uFirstDesc = pDescriptor->cLines - 1;
2187	rc = vmdkDescInitStr(pImage, pDescriptor, "");
2188	}
2189	if (RT_SUCCESS(rc))
2190	rc = vmdkDescInitStr(pImage, pDescriptor, "# Extent description");
2191	if (RT_SUCCESS(rc))
2192	rc = vmdkDescInitStr(pImage, pDescriptor, "NOACCESS 0 ZERO ");
2193	if (RT_SUCCESS(rc))
2194	{
2195	pDescriptor->uFirstExtent = pDescriptor->cLines - 1;
2196	rc = vmdkDescInitStr(pImage, pDescriptor, "");
2197	}
2198	if (RT_SUCCESS(rc))
2199	{
2200	/* The trailing space is created by VMware, too. */
2201	rc = vmdkDescInitStr(pImage, pDescriptor, "# The disk Data Base ");
2202	}
2203	if (RT_SUCCESS(rc))
2204	rc = vmdkDescInitStr(pImage, pDescriptor, "#DDB");
2205	if (RT_SUCCESS(rc))
2206	rc = vmdkDescInitStr(pImage, pDescriptor, "");
2207	if (RT_SUCCESS(rc))
2208	rc = vmdkDescInitStr(pImage, pDescriptor, "ddb.virtualHWVersion = \"4\"");
2209	if (RT_SUCCESS(rc))
2210	{
2211	pDescriptor->uFirstDDB = pDescriptor->cLines - 1;
2212
2213	/* Now that the framework is in place, use the normal functions to insert
2214	* the remaining keys. */
2215	char szBuf[9];
2216	RTStrPrintf(szBuf, sizeof(szBuf), "%08x", RTRandU32());
2217	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc,
2218	"CID", szBuf);
2219	}
2220	if (RT_SUCCESS(rc))
2221	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc,
2222	"parentCID", "ffffffff");
2223	if (RT_SUCCESS(rc))
2224	rc = vmdkDescDDBSetStr(pImage, pDescriptor, "ddb.adapterType", "ide");
2225
2226	return rc;
2227	}
2228
2229	static int vmdkParseDescriptor(PVMDKIMAGE pImage, char *pDescData, size_t cbDescData)
2230	{
2231	int rc;
2232	unsigned cExtents;
2233	unsigned uLine;
2234	unsigned i;
2235
2236	rc = vmdkPreprocessDescriptor(pImage, pDescData, cbDescData,
2237	&pImage->Descriptor);
2238	if (RT_FAILURE(rc))
2239	return rc;
2240
2241	/* Check version, must be 1. */
2242	uint32_t uVersion;
2243	rc = vmdkDescBaseGetU32(&pImage->Descriptor, "version", &uVersion);
2244	if (RT_FAILURE(rc))
2245	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error finding key 'version' in descriptor in '%s'"), pImage->pszFilename);
2246	if (uVersion != 1)
2247	return vdIfError(pImage->pIfError, VERR_VD_VMDK_UNSUPPORTED_VERSION, RT_SRC_POS, N_("VMDK: unsupported format version in descriptor in '%s'"), pImage->pszFilename);
2248
2249	/* Get image creation type and determine image flags. */
2250	char pszCreateType = NULL; / initialized to make gcc shut up */
2251	rc = vmdkDescBaseGetStr(pImage, &pImage->Descriptor, "createType",
2252	&pszCreateType);
2253	if (RT_FAILURE(rc))
2254	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot get image type from descriptor in '%s'"), pImage->pszFilename);
2255	if ( !strcmp(pszCreateType, "twoGbMaxExtentSparse")
2256	\|\| !strcmp(pszCreateType, "twoGbMaxExtentFlat"))
2257	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_SPLIT_2G;
2258	else if ( !strcmp(pszCreateType, "partitionedDevice")
2259	\|\| !strcmp(pszCreateType, "fullDevice"))
2260	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_RAWDISK;
2261	else if (!strcmp(pszCreateType, "streamOptimized"))
2262	pImage->uImageFlags \|= VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED;
2263	else if (!strcmp(pszCreateType, "vmfs"))
2264	pImage->uImageFlags \|= VD_IMAGE_FLAGS_FIXED \| VD_VMDK_IMAGE_FLAGS_ESX;
2265	RTMemTmpFree(pszCreateType);
2266
2267	/* Count the number of extent config entries. */
2268	for (uLine = pImage->Descriptor.uFirstExtent, cExtents = 0;
2269	uLine != 0;
2270	uLine = pImage->Descriptor.aNextLines[uLine], cExtents++)
2271	/* nothing */;
2272
2273	if (!pImage->pDescData && cExtents != 1)
2274	{
2275	/* Monolithic image, must have only one extent (already opened). */
2276	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);
2277	}
2278
2279	if (pImage->pDescData)
2280	{
2281	/* Non-monolithic image, extents need to be allocated. */
2282	rc = vmdkCreateExtents(pImage, cExtents);
2283	if (RT_FAILURE(rc))
2284	return rc;
2285	}
2286
2287	for (i = 0, uLine = pImage->Descriptor.uFirstExtent;
2288	i < cExtents; i++, uLine = pImage->Descriptor.aNextLines[uLine])
2289	{
2290	char *pszLine = pImage->Descriptor.aLines[uLine];
2291
2292	/* Access type of the extent. */
2293	if (!strncmp(pszLine, "RW", 2))
2294	{
2295	pImage->pExtents[i].enmAccess = VMDKACCESS_READWRITE;
2296	pszLine += 2;
2297	}
2298	else if (!strncmp(pszLine, "RDONLY", 6))
2299	{
2300	pImage->pExtents[i].enmAccess = VMDKACCESS_READONLY;
2301	pszLine += 6;
2302	}
2303	else if (!strncmp(pszLine, "NOACCESS", 8))
2304	{
2305	pImage->pExtents[i].enmAccess = VMDKACCESS_NOACCESS;
2306	pszLine += 8;
2307	}
2308	else
2309	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2310	if (*pszLine++ != ' ')
2311	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2312
2313	/* Nominal size of the extent. */
2314	rc = RTStrToUInt64Ex(pszLine, &pszLine, 10,
2315	&pImage->pExtents[i].cNominalSectors);
2316	if (RT_FAILURE(rc))
2317	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2318	if (*pszLine++ != ' ')
2319	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2320
2321	/* Type of the extent. */
2322	if (!strncmp(pszLine, "SPARSE", 6))
2323	{
2324	pImage->pExtents[i].enmType = VMDKETYPE_HOSTED_SPARSE;
2325	pszLine += 6;
2326	}
2327	else if (!strncmp(pszLine, "FLAT", 4))
2328	{
2329	pImage->pExtents[i].enmType = VMDKETYPE_FLAT;
2330	pszLine += 4;
2331	}
2332	else if (!strncmp(pszLine, "ZERO", 4))
2333	{
2334	pImage->pExtents[i].enmType = VMDKETYPE_ZERO;
2335	pszLine += 4;
2336	}
2337	else if (!strncmp(pszLine, "VMFS", 4))
2338	{
2339	pImage->pExtents[i].enmType = VMDKETYPE_VMFS;
2340	pszLine += 4;
2341	}
2342	else
2343	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2344
2345	if (pImage->pExtents[i].enmType == VMDKETYPE_ZERO)
2346	{
2347	/* This one has no basename or offset. */
2348	if (*pszLine == ' ')
2349	pszLine++;
2350	if (*pszLine != '\0')
2351	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2352	pImage->pExtents[i].pszBasename = NULL;
2353	}
2354	else
2355	{
2356	/* All other extent types have basename and optional offset. */
2357	if (*pszLine++ != ' ')
2358	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2359
2360	/* Basename of the image. Surrounded by quotes. */
2361	char *pszBasename;
2362	rc = vmdkStringUnquote(pImage, pszLine, &pszBasename, &pszLine);
2363	if (RT_FAILURE(rc))
2364	return rc;
2365	pImage->pExtents[i].pszBasename = pszBasename;
2366	if (*pszLine == ' ')
2367	{
2368	pszLine++;
2369	if (*pszLine != '\0')
2370	{
2371	/* Optional offset in extent specified. */
2372	rc = RTStrToUInt64Ex(pszLine, &pszLine, 10,
2373	&pImage->pExtents[i].uSectorOffset);
2374	if (RT_FAILURE(rc))
2375	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2376	}
2377	}
2378
2379	if (*pszLine != '\0')
2380	return vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
2381	}
2382	}
2383
2384	/* Determine PCHS geometry (autogenerate if necessary). */
2385	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2386	VMDK_DDB_GEO_PCHS_CYLINDERS,
2387	&pImage->PCHSGeometry.cCylinders);
2388	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2389	pImage->PCHSGeometry.cCylinders = 0;
2390	else if (RT_FAILURE(rc))
2391	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2392	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2393	VMDK_DDB_GEO_PCHS_HEADS,
2394	&pImage->PCHSGeometry.cHeads);
2395	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2396	pImage->PCHSGeometry.cHeads = 0;
2397	else if (RT_FAILURE(rc))
2398	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2399	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2400	VMDK_DDB_GEO_PCHS_SECTORS,
2401	&pImage->PCHSGeometry.cSectors);
2402	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2403	pImage->PCHSGeometry.cSectors = 0;
2404	else if (RT_FAILURE(rc))
2405	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
2406	if ( pImage->PCHSGeometry.cCylinders == 0
2407	\|\| pImage->PCHSGeometry.cHeads == 0
2408	\|\| pImage->PCHSGeometry.cHeads > 16
2409	\|\| pImage->PCHSGeometry.cSectors == 0
2410	\|\| pImage->PCHSGeometry.cSectors > 63)
2411	{
2412	/* Mark PCHS geometry as not yet valid (can't do the calculation here
2413	* as the total image size isn't known yet). */
2414	pImage->PCHSGeometry.cCylinders = 0;
2415	pImage->PCHSGeometry.cHeads = 16;
2416	pImage->PCHSGeometry.cSectors = 63;
2417	}
2418
2419	/* Determine LCHS geometry (set to 0 if not specified). */
2420	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2421	VMDK_DDB_GEO_LCHS_CYLINDERS,
2422	&pImage->LCHSGeometry.cCylinders);
2423	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2424	pImage->LCHSGeometry.cCylinders = 0;
2425	else if (RT_FAILURE(rc))
2426	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2427	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2428	VMDK_DDB_GEO_LCHS_HEADS,
2429	&pImage->LCHSGeometry.cHeads);
2430	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2431	pImage->LCHSGeometry.cHeads = 0;
2432	else if (RT_FAILURE(rc))
2433	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2434	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
2435	VMDK_DDB_GEO_LCHS_SECTORS,
2436	&pImage->LCHSGeometry.cSectors);
2437	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2438	pImage->LCHSGeometry.cSectors = 0;
2439	else if (RT_FAILURE(rc))
2440	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
2441	if ( pImage->LCHSGeometry.cCylinders == 0
2442	\|\| pImage->LCHSGeometry.cHeads == 0
2443	\|\| pImage->LCHSGeometry.cSectors == 0)
2444	{
2445	pImage->LCHSGeometry.cCylinders = 0;
2446	pImage->LCHSGeometry.cHeads = 0;
2447	pImage->LCHSGeometry.cSectors = 0;
2448	}
2449
2450	/* Get image UUID. */
2451	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor, VMDK_DDB_IMAGE_UUID,
2452	&pImage->ImageUuid);
2453	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2454	{
2455	/* Image without UUID. Probably created by VMware and not yet used
2456	* by VirtualBox. Can only be added for images opened in read/write
2457	* mode, so don't bother producing a sensible UUID otherwise. */
2458	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2459	RTUuidClear(&pImage->ImageUuid);
2460	else
2461	{
2462	rc = RTUuidCreate(&pImage->ImageUuid);
2463	if (RT_FAILURE(rc))
2464	return rc;
2465	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2466	VMDK_DDB_IMAGE_UUID, &pImage->ImageUuid);
2467	if (RT_FAILURE(rc))
2468	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image UUID in descriptor in '%s'"), pImage->pszFilename);
2469	}
2470	}
2471	else if (RT_FAILURE(rc))
2472	return rc;
2473
2474	/* Get image modification UUID. */
2475	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor,
2476	VMDK_DDB_MODIFICATION_UUID,
2477	&pImage->ModificationUuid);
2478	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2479	{
2480	/* Image without UUID. Probably created by VMware and not yet used
2481	* by VirtualBox. Can only be added for images opened in read/write
2482	* mode, so don't bother producing a sensible UUID otherwise. */
2483	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2484	RTUuidClear(&pImage->ModificationUuid);
2485	else
2486	{
2487	rc = RTUuidCreate(&pImage->ModificationUuid);
2488	if (RT_FAILURE(rc))
2489	return rc;
2490	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2491	VMDK_DDB_MODIFICATION_UUID,
2492	&pImage->ModificationUuid);
2493	if (RT_FAILURE(rc))
2494	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image modification UUID in descriptor in '%s'"), pImage->pszFilename);
2495	}
2496	}
2497	else if (RT_FAILURE(rc))
2498	return rc;
2499
2500	/* Get UUID of parent image. */
2501	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor, VMDK_DDB_PARENT_UUID,
2502	&pImage->ParentUuid);
2503	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2504	{
2505	/* Image without UUID. Probably created by VMware and not yet used
2506	* by VirtualBox. Can only be added for images opened in read/write
2507	* mode, so don't bother producing a sensible UUID otherwise. */
2508	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2509	RTUuidClear(&pImage->ParentUuid);
2510	else
2511	{
2512	rc = RTUuidClear(&pImage->ParentUuid);
2513	if (RT_FAILURE(rc))
2514	return rc;
2515	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2516	VMDK_DDB_PARENT_UUID, &pImage->ParentUuid);
2517	if (RT_FAILURE(rc))
2518	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent UUID in descriptor in '%s'"), pImage->pszFilename);
2519	}
2520	}
2521	else if (RT_FAILURE(rc))
2522	return rc;
2523
2524	/* Get parent image modification UUID. */
2525	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor,
2526	VMDK_DDB_PARENT_MODIFICATION_UUID,
2527	&pImage->ParentModificationUuid);
2528	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
2529	{
2530	/* Image without UUID. Probably created by VMware and not yet used
2531	* by VirtualBox. Can only be added for images opened in read/write
2532	* mode, so don't bother producing a sensible UUID otherwise. */
2533	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2534	RTUuidClear(&pImage->ParentModificationUuid);
2535	else
2536	{
2537	RTUuidClear(&pImage->ParentModificationUuid);
2538	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
2539	VMDK_DDB_PARENT_MODIFICATION_UUID,
2540	&pImage->ParentModificationUuid);
2541	if (RT_FAILURE(rc))
2542	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent modification UUID in descriptor in '%s'"), pImage->pszFilename);
2543	}
2544	}
2545	else if (RT_FAILURE(rc))
2546	return rc;
2547
2548	return VINF_SUCCESS;
2549	}
2550
2551	/**
2552	* Internal : Prepares the descriptor to write to the image.
2553	*/
2554	static int vmdkDescriptorPrepare(PVMDKIMAGE pImage, uint64_t cbLimit,
2555	void *ppvData, size_t pcbData)
2556	{
2557	int rc = VINF_SUCCESS;
2558
2559	/*
2560	* Allocate temporary descriptor buffer.
2561	* In case there is no limit allocate a default
2562	* and increase if required.
2563	*/
2564	size_t cbDescriptor = cbLimit ? cbLimit : 4 * _1K;
2565	char pszDescriptor = (char )RTMemAllocZ(cbDescriptor);
2566	size_t offDescriptor = 0;
2567
2568	if (!pszDescriptor)
2569	return VERR_NO_MEMORY;
2570
2571	for (unsigned i = 0; i < pImage->Descriptor.cLines; i++)
2572	{
2573	const char *psz = pImage->Descriptor.aLines[i];
2574	size_t cb = strlen(psz);
2575
2576	/*
2577	* Increase the descriptor if there is no limit and
2578	* there is not enough room left for this line.
2579	*/
2580	if (offDescriptor + cb + 1 > cbDescriptor)
2581	{
2582	if (cbLimit)
2583	{
2584	rc = vdIfError(pImage->pIfError, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too long in '%s'"), pImage->pszFilename);
2585	break;
2586	}
2587	else
2588	{
2589	char *pszDescriptorNew = NULL;
2590	LogFlow(("Increasing descriptor cache\n"));
2591
2592	pszDescriptorNew = (char )RTMemRealloc(pszDescriptor, cbDescriptor + cb + 4 _1K);
2593	if (!pszDescriptorNew)
2594	{
2595	rc = VERR_NO_MEMORY;
2596	break;
2597	}
2598	pszDescriptor = pszDescriptorNew;
2599	cbDescriptor += cb + 4 * _1K;
2600	}
2601	}
2602
2603	if (cb > 0)
2604	{
2605	memcpy(pszDescriptor + offDescriptor, psz, cb);
2606	offDescriptor += cb;
2607	}
2608
2609	memcpy(pszDescriptor + offDescriptor, "\n", 1);
2610	offDescriptor++;
2611	}
2612
2613	if (RT_SUCCESS(rc))
2614	{
2615	*ppvData = pszDescriptor;
2616	*pcbData = offDescriptor;
2617	}
2618	else if (pszDescriptor)
2619	RTMemFree(pszDescriptor);
2620
2621	return rc;
2622	}
2623
2624	/**
2625	* Internal: write/update the descriptor part of the image.
2626	*/
2627	static int vmdkWriteDescriptor(PVMDKIMAGE pImage, PVDIOCTX pIoCtx)
2628	{
2629	int rc = VINF_SUCCESS;
2630	uint64_t cbLimit;
2631	uint64_t uOffset;
2632	PVMDKFILE pDescFile;
2633	void *pvDescriptor = NULL;
2634	size_t cbDescriptor;
2635
2636	if (pImage->pDescData)
2637	{
2638	/* Separate descriptor file. */
2639	uOffset = 0;
2640	cbLimit = 0;
2641	pDescFile = pImage->pFile;
2642	}
2643	else
2644	{
2645	/* Embedded descriptor file. */
2646	uOffset = VMDK_SECTOR2BYTE(pImage->pExtents[0].uDescriptorSector);
2647	cbLimit = VMDK_SECTOR2BYTE(pImage->pExtents[0].cDescriptorSectors);
2648	pDescFile = pImage->pExtents[0].pFile;
2649	}
2650	/* Bail out if there is no file to write to. */
2651	if (pDescFile == NULL)
2652	return VERR_INVALID_PARAMETER;
2653
2654	rc = vmdkDescriptorPrepare(pImage, cbLimit, &pvDescriptor, &cbDescriptor);
2655	if (RT_SUCCESS(rc))
2656	{
2657	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pDescFile->pStorage,
2658	uOffset, pvDescriptor,
2659	cbLimit ? cbLimit : cbDescriptor,
2660	pIoCtx, NULL, NULL);
2661	if ( RT_FAILURE(rc)
2662	&& rc != VERR_VD_ASYNC_IO_IN_PROGRESS)
2663	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error writing descriptor in '%s'"), pImage->pszFilename);
2664	}
2665
2666	if (RT_SUCCESS(rc) && !cbLimit)
2667	{
2668	rc = vdIfIoIntFileSetSize(pImage->pIfIo, pDescFile->pStorage, cbDescriptor);
2669	if (RT_FAILURE(rc))
2670	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error truncating descriptor in '%s'"), pImage->pszFilename);
2671	}
2672
2673	if (RT_SUCCESS(rc))
2674	pImage->Descriptor.fDirty = false;
2675
2676	if (pvDescriptor)
2677	RTMemFree(pvDescriptor);
2678	return rc;
2679
2680	}
2681
2682	/**
2683	* Internal: validate the consistency check values in a binary header.
2684	*/
2685	static int vmdkValidateHeader(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, const SparseExtentHeader *pHeader)
2686	{
2687	int rc = VINF_SUCCESS;
2688	if (RT_LE2H_U32(pHeader->magicNumber) != VMDK_SPARSE_MAGICNUMBER)
2689	{
2690	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrect magic in sparse extent header in '%s'"), pExtent->pszFullname);
2691	return rc;
2692	}
2693	if (RT_LE2H_U32(pHeader->version) != 1 && RT_LE2H_U32(pHeader->version) != 3)
2694	{
2695	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);
2696	return rc;
2697	}
2698	if ( (RT_LE2H_U32(pHeader->flags) & 1)
2699	&& ( pHeader->singleEndLineChar != '\n'
2700	\|\| pHeader->nonEndLineChar != ' '
2701	\|\| pHeader->doubleEndLineChar1 != '\r'
2702	\|\| pHeader->doubleEndLineChar2 != '\n') )
2703	{
2704	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: corrupted by CR/LF translation in '%s'"), pExtent->pszFullname);
2705	return rc;
2706	}
2707	if (RT_LE2H_U64(pHeader->descriptorSize) > VMDK_SPARSE_DESCRIPTOR_SIZE_MAX)
2708	{
2709	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor size out of bounds (%llu vs %llu) '%s'"),
2710	pExtent->pszFullname, RT_LE2H_U64(pHeader->descriptorSize), VMDK_SPARSE_DESCRIPTOR_SIZE_MAX);
2711	return rc;
2712	}
2713	return rc;
2714	}
2715
2716	/**
2717	* Internal: read metadata belonging to an extent with binary header, i.e.
2718	* as found in monolithic files.
2719	*/
2720	static int vmdkReadBinaryMetaExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
2721	bool fMagicAlreadyRead)
2722	{
2723	SparseExtentHeader Header;
2724	int rc;
2725
2726	if (!fMagicAlreadyRead)
2727	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage, 0,
2728	&Header, sizeof(Header));
2729	else
2730	{
2731	Header.magicNumber = RT_H2LE_U32(VMDK_SPARSE_MAGICNUMBER);
2732	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
2733	RT_UOFFSETOF(SparseExtentHeader, version),
2734	&Header.version,
2735	sizeof(Header)
2736	- RT_UOFFSETOF(SparseExtentHeader, version));
2737	}
2738
2739	if (RT_SUCCESS(rc))
2740	{
2741	rc = vmdkValidateHeader(pImage, pExtent, &Header);
2742	if (RT_SUCCESS(rc))
2743	{
2744	uint64_t cbFile = 0;
2745
2746	if ( (RT_LE2H_U32(Header.flags) & RT_BIT(17))
2747	&& RT_LE2H_U64(Header.gdOffset) == VMDK_GD_AT_END)
2748	pExtent->fFooter = true;
2749
2750	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2751	\|\| ( pExtent->fFooter
2752	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2753	{
2754	rc = vdIfIoIntFileGetSize(pImage->pIfIo, pExtent->pFile->pStorage, &cbFile);
2755	if (RT_FAILURE(rc))
2756	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot get size of '%s'"), pExtent->pszFullname);
2757	}
2758
2759	if (RT_SUCCESS(rc))
2760	{
2761	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
2762	pExtent->uAppendPosition = RT_ALIGN_64(cbFile, 512);
2763
2764	if ( pExtent->fFooter
2765	&& ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2766	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2767	{
2768	/* Read the footer, which comes before the end-of-stream marker. */
2769	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
2770	cbFile - 2*512, &Header,
2771	sizeof(Header));
2772	if (RT_FAILURE(rc))
2773	{
2774	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading extent footer in '%s'"), pExtent->pszFullname);
2775	rc = VERR_VD_VMDK_INVALID_HEADER;
2776	}
2777
2778	if (RT_SUCCESS(rc))
2779	rc = vmdkValidateHeader(pImage, pExtent, &Header);
2780	/* Prohibit any writes to this extent. */
2781	pExtent->uAppendPosition = 0;
2782	}
2783
2784	if (RT_SUCCESS(rc))
2785	{
2786	pExtent->uVersion = RT_LE2H_U32(Header.version);
2787	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE; /* Just dummy value, changed later. */
2788	pExtent->cSectors = RT_LE2H_U64(Header.capacity);
2789	pExtent->cSectorsPerGrain = RT_LE2H_U64(Header.grainSize);
2790	pExtent->uDescriptorSector = RT_LE2H_U64(Header.descriptorOffset);
2791	pExtent->cDescriptorSectors = RT_LE2H_U64(Header.descriptorSize);
2792	pExtent->cGTEntries = RT_LE2H_U32(Header.numGTEsPerGT);
2793	pExtent->cOverheadSectors = RT_LE2H_U64(Header.overHead);
2794	pExtent->fUncleanShutdown = !!Header.uncleanShutdown;
2795	pExtent->uCompression = RT_LE2H_U16(Header.compressAlgorithm);
2796	if (RT_LE2H_U32(Header.flags) & RT_BIT(1))
2797	{
2798	pExtent->uSectorRGD = RT_LE2H_U64(Header.rgdOffset);
2799	pExtent->uSectorGD = RT_LE2H_U64(Header.gdOffset);
2800	}
2801	else
2802	{
2803	pExtent->uSectorGD = RT_LE2H_U64(Header.gdOffset);
2804	pExtent->uSectorRGD = 0;
2805	}
2806
2807	if (pExtent->uDescriptorSector && !pExtent->cDescriptorSectors)
2808	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2809	N_("VMDK: inconsistent embedded descriptor config in '%s'"), pExtent->pszFullname);
2810
2811	if ( RT_SUCCESS(rc)
2812	&& ( pExtent->uSectorGD == VMDK_GD_AT_END
2813	\|\| pExtent->uSectorRGD == VMDK_GD_AT_END)
2814	&& ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2815	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL)))
2816	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2817	N_("VMDK: cannot resolve grain directory offset in '%s'"), pExtent->pszFullname);
2818
2819	if (RT_SUCCESS(rc))
2820	{
2821	uint64_t cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
2822	if (!cSectorsPerGDE \|\| cSectorsPerGDE > UINT32_MAX)
2823	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2824	N_("VMDK: incorrect grain directory size in '%s'"), pExtent->pszFullname);
2825	else
2826	{
2827	pExtent->cSectorsPerGDE = cSectorsPerGDE;
2828	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
2829
2830	/* Fix up the number of descriptor sectors, as some flat images have
2831	* really just one, and this causes failures when inserting the UUID
2832	* values and other extra information. */
2833	if (pExtent->cDescriptorSectors != 0 && pExtent->cDescriptorSectors < 4)
2834	{
2835	/* Do it the easy way - just fix it for flat images which have no
2836	* other complicated metadata which needs space too. */
2837	if ( pExtent->uDescriptorSector + 4 < pExtent->cOverheadSectors
2838	&& pExtent->cGTEntries * pExtent->cGDEntries == 0)
2839	pExtent->cDescriptorSectors = 4;
2840	}
2841	}
2842	}
2843	}
2844	}
2845	}
2846	}
2847	else
2848	{
2849	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading extent header in '%s'"), pExtent->pszFullname);
2850	rc = VERR_VD_VMDK_INVALID_HEADER;
2851	}
2852
2853	if (RT_FAILURE(rc))
2854	vmdkFreeExtentData(pImage, pExtent, false);
2855
2856	return rc;
2857	}
2858
2859	/**
2860	* Internal: read additional metadata belonging to an extent. For those
2861	* extents which have no additional metadata just verify the information.
2862	*/
2863	static int vmdkReadMetaExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
2864	{
2865	int rc = VINF_SUCCESS;
2866
2867	/* disabled the check as there are too many truncated vmdk images out there */
2868	#ifdef VBOX_WITH_VMDK_STRICT_SIZE_CHECK
2869	uint64_t cbExtentSize;
2870	/* The image must be a multiple of a sector in size and contain the data
2871	* area (flat images only). If not, it means the image is at least
2872	* truncated, or even seriously garbled. */
2873	rc = vdIfIoIntFileGetSize(pImage->pIfIo, pExtent->pFile->pStorage, &cbExtentSize);
2874	if (RT_FAILURE(rc))
2875	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error getting size in '%s'"), pExtent->pszFullname);
2876	else if ( cbExtentSize != RT_ALIGN_64(cbExtentSize, 512)
2877	&& (pExtent->enmType != VMDKETYPE_FLAT \|\| pExtent->cNominalSectors + pExtent->uSectorOffset > VMDK_BYTE2SECTOR(cbExtentSize)))
2878	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2879	N_("VMDK: file size is not a multiple of 512 in '%s', file is truncated or otherwise garbled"), pExtent->pszFullname);
2880	#endif /* VBOX_WITH_VMDK_STRICT_SIZE_CHECK */
2881	if ( RT_SUCCESS(rc)
2882	&& pExtent->enmType == VMDKETYPE_HOSTED_SPARSE)
2883	{
2884	/* The spec says that this must be a power of two and greater than 8,
2885	* but probably they meant not less than 8. */
2886	if ( (pExtent->cSectorsPerGrain & (pExtent->cSectorsPerGrain - 1))
2887	\|\| pExtent->cSectorsPerGrain < 8)
2888	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2889	N_("VMDK: invalid extent grain size %u in '%s'"), pExtent->cSectorsPerGrain, pExtent->pszFullname);
2890	else
2891	{
2892	/* This code requires that a grain table must hold a power of two multiple
2893	* of the number of entries per GT cache entry. */
2894	if ( (pExtent->cGTEntries & (pExtent->cGTEntries - 1))
2895	\|\| pExtent->cGTEntries < VMDK_GT_CACHELINE_SIZE)
2896	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS,
2897	N_("VMDK: grain table cache size problem in '%s'"), pExtent->pszFullname);
2898	else
2899	{
2900	rc = vmdkAllocStreamBuffers(pImage, pExtent);
2901	if (RT_SUCCESS(rc))
2902	{
2903	/* Prohibit any writes to this streamOptimized extent. */
2904	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2905	pExtent->uAppendPosition = 0;
2906
2907	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2908	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
2909	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
2910	rc = vmdkReadGrainDirectory(pImage, pExtent);
2911	else
2912	{
2913	pExtent->uGrainSectorAbs = pExtent->cOverheadSectors;
2914	pExtent->cbGrainStreamRead = 0;
2915	}
2916	}
2917	}
2918	}
2919	}
2920
2921	if (RT_FAILURE(rc))
2922	vmdkFreeExtentData(pImage, pExtent, false);
2923
2924	return rc;
2925	}
2926
2927	/**
2928	* Internal: write/update the metadata for a sparse extent.
2929	*/
2930	static int vmdkWriteMetaSparseExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
2931	uint64_t uOffset, PVDIOCTX pIoCtx)
2932	{
2933	SparseExtentHeader Header;
2934
2935	memset(&Header, '\0', sizeof(Header));
2936	Header.magicNumber = RT_H2LE_U32(VMDK_SPARSE_MAGICNUMBER);
2937	Header.version = RT_H2LE_U32(pExtent->uVersion);
2938	Header.flags = RT_H2LE_U32(RT_BIT(0));
2939	if (pExtent->pRGD)
2940	Header.flags \|= RT_H2LE_U32(RT_BIT(1));
2941	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
2942	Header.flags \|= RT_H2LE_U32(RT_BIT(16) \| RT_BIT(17));
2943	Header.capacity = RT_H2LE_U64(pExtent->cSectors);
2944	Header.grainSize = RT_H2LE_U64(pExtent->cSectorsPerGrain);
2945	Header.descriptorOffset = RT_H2LE_U64(pExtent->uDescriptorSector);
2946	Header.descriptorSize = RT_H2LE_U64(pExtent->cDescriptorSectors);
2947	Header.numGTEsPerGT = RT_H2LE_U32(pExtent->cGTEntries);
2948	if (pExtent->fFooter && uOffset == 0)
2949	{
2950	if (pExtent->pRGD)
2951	{
2952	Assert(pExtent->uSectorRGD);
2953	Header.rgdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2954	Header.gdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2955	}
2956	else
2957	Header.gdOffset = RT_H2LE_U64(VMDK_GD_AT_END);
2958	}
2959	else
2960	{
2961	if (pExtent->pRGD)
2962	{
2963	Assert(pExtent->uSectorRGD);
2964	Header.rgdOffset = RT_H2LE_U64(pExtent->uSectorRGD);
2965	Header.gdOffset = RT_H2LE_U64(pExtent->uSectorGD);
2966	}
2967	else
2968	Header.gdOffset = RT_H2LE_U64(pExtent->uSectorGD);
2969	}
2970	Header.overHead = RT_H2LE_U64(pExtent->cOverheadSectors);
2971	Header.uncleanShutdown = pExtent->fUncleanShutdown;
2972	Header.singleEndLineChar = '\n';
2973	Header.nonEndLineChar = ' ';
2974	Header.doubleEndLineChar1 = '\r';
2975	Header.doubleEndLineChar2 = '\n';
2976	Header.compressAlgorithm = RT_H2LE_U16(pExtent->uCompression);
2977
2978	int rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
2979	uOffset, &Header, sizeof(Header),
2980	pIoCtx, NULL, NULL);
2981	if (RT_FAILURE(rc) && (rc != VERR_VD_ASYNC_IO_IN_PROGRESS))
2982	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error writing extent header in '%s'"), pExtent->pszFullname);
2983	return rc;
2984	}
2985
2986	/**
2987	* Internal: free the buffers used for streamOptimized images.
2988	*/
2989	static void vmdkFreeStreamBuffers(PVMDKEXTENT pExtent)
2990	{
2991	if (pExtent->pvCompGrain)
2992	{
2993	RTMemFree(pExtent->pvCompGrain);
2994	pExtent->pvCompGrain = NULL;
2995	}
2996	if (pExtent->pvGrain)
2997	{
2998	RTMemFree(pExtent->pvGrain);
2999	pExtent->pvGrain = NULL;
3000	}
3001	}
3002
3003	/**
3004	* Internal: free the memory used by the extent data structure, optionally
3005	* deleting the referenced files.
3006	*
3007	* @returns VBox status code.
3008	* @param pImage Pointer to the image instance data.
3009	* @param pExtent The extent to free.
3010	* @param fDelete Flag whether to delete the backing storage.
3011	*/
3012	static int vmdkFreeExtentData(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
3013	bool fDelete)
3014	{
3015	int rc = VINF_SUCCESS;
3016
3017	vmdkFreeGrainDirectory(pExtent);
3018	if (pExtent->pDescData)
3019	{
3020	RTMemFree(pExtent->pDescData);
3021	pExtent->pDescData = NULL;
3022	}
3023	if (pExtent->pFile != NULL)
3024	{
3025	/* Do not delete raw extents, these have full and base names equal. */
3026	rc = vmdkFileClose(pImage, &pExtent->pFile,
3027	fDelete
3028	&& pExtent->pszFullname
3029	&& pExtent->pszBasename
3030	&& strcmp(pExtent->pszFullname, pExtent->pszBasename));
3031	}
3032	if (pExtent->pszBasename)
3033	{
3034	RTMemTmpFree((void *)pExtent->pszBasename);
3035	pExtent->pszBasename = NULL;
3036	}
3037	if (pExtent->pszFullname)
3038	{
3039	RTStrFree((char )(void )pExtent->pszFullname);
3040	pExtent->pszFullname = NULL;
3041	}
3042	vmdkFreeStreamBuffers(pExtent);
3043
3044	return rc;
3045	}
3046
3047	/**
3048	* Internal: allocate grain table cache if necessary for this image.
3049	*/
3050	static int vmdkAllocateGrainTableCache(PVMDKIMAGE pImage)
3051	{
3052	PVMDKEXTENT pExtent;
3053
3054	/* Allocate grain table cache if any sparse extent is present. */
3055	for (unsigned i = 0; i < pImage->cExtents; i++)
3056	{
3057	pExtent = &pImage->pExtents[i];
3058	if (pExtent->enmType == VMDKETYPE_HOSTED_SPARSE)
3059	{
3060	/* Allocate grain table cache. */
3061	pImage->pGTCache = (PVMDKGTCACHE)RTMemAllocZ(sizeof(VMDKGTCACHE));
3062	if (!pImage->pGTCache)
3063	return VERR_NO_MEMORY;
3064	for (unsigned j = 0; j < VMDK_GT_CACHE_SIZE; j++)
3065	{
3066	PVMDKGTCACHEENTRY pGCE = &pImage->pGTCache->aGTCache[j];
3067	pGCE->uExtent = UINT32_MAX;
3068	}
3069	pImage->pGTCache->cEntries = VMDK_GT_CACHE_SIZE;
3070	break;
3071	}
3072	}
3073
3074	return VINF_SUCCESS;
3075	}
3076
3077	/**
3078	* Internal: allocate the given number of extents.
3079	*/
3080	static int vmdkCreateExtents(PVMDKIMAGE pImage, unsigned cExtents)
3081	{
3082	int rc = VINF_SUCCESS;
3083	PVMDKEXTENT pExtents = (PVMDKEXTENT)RTMemAllocZ(cExtents * sizeof(VMDKEXTENT));
3084	if (pExtents)
3085	{
3086	for (unsigned i = 0; i < cExtents; i++)
3087	{
3088	pExtents[i].pFile = NULL;
3089	pExtents[i].pszBasename = NULL;
3090	pExtents[i].pszFullname = NULL;
3091	pExtents[i].pGD = NULL;
3092	pExtents[i].pRGD = NULL;
3093	pExtents[i].pDescData = NULL;
3094	pExtents[i].uVersion = 1;
3095	pExtents[i].uCompression = VMDK_COMPRESSION_NONE;
3096	pExtents[i].uExtent = i;
3097	pExtents[i].pImage = pImage;
3098	}
3099	pImage->pExtents = pExtents;
3100	pImage->cExtents = cExtents;
3101	}
3102	else
3103	rc = VERR_NO_MEMORY;
3104
3105	return rc;
3106	}
3107
3108	/**
3109	* Internal: allocate and describes an additional, file-backed extent
3110	* for the given size. Preserves original extents.
3111	*/
3112	static int vmdkAddFileBackedExtent(PVMDKIMAGE pImage, uint64_t cbSize)
3113	{
3114	int rc = VINF_SUCCESS;
3115	PVMDKEXTENT pNewExtents = (PVMDKEXTENT)RTMemAllocZ((pImage->cExtents + 1) * sizeof(VMDKEXTENT));
3116	if (pNewExtents)
3117	{
3118	memcpy(pNewExtents, pImage->pExtents, pImage->cExtents * sizeof(VMDKEXTENT));
3119	PVMDKEXTENT pExtent = &pNewExtents[pImage->cExtents];
3120
3121	pExtent->pFile = NULL;
3122	pExtent->pszBasename = NULL;
3123	pExtent->pszFullname = NULL;
3124	pExtent->pGD = NULL;
3125	pExtent->pRGD = NULL;
3126	pExtent->pDescData = NULL;
3127	pExtent->uVersion = 1;
3128	pExtent->uCompression = VMDK_COMPRESSION_NONE;
3129	pExtent->uExtent = pImage->cExtents;
3130	pExtent->pImage = pImage;
3131	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
3132	pExtent->enmType = VMDKETYPE_FLAT;
3133	pExtent->enmAccess = VMDKACCESS_READWRITE;
3134	pExtent->uSectorOffset = 0;
3135
3136	char *pszBasenameSubstr = RTPathFilename(pImage->pszFilename);
3137	AssertPtr(pszBasenameSubstr);
3138
3139	char *pszBasenameSuff = RTPathSuffix(pszBasenameSubstr);
3140	char *pszBasenameBase = RTStrDup(pszBasenameSubstr);
3141	RTPathStripSuffix(pszBasenameBase);
3142	char *pszTmp;
3143	size_t cbTmp;
3144
3145	if (pImage->uImageFlags & VD_IMAGE_FLAGS_FIXED)
3146	RTStrAPrintf(&pszTmp, "%s-f%03d%s", pszBasenameBase,
3147	pExtent->uExtent + 1, pszBasenameSuff);
3148	else
3149	RTStrAPrintf(&pszTmp, "%s-s%03d%s", pszBasenameBase, pExtent->uExtent + 1,
3150	pszBasenameSuff);
3151
3152	RTStrFree(pszBasenameBase);
3153	if (!pszTmp)
3154	return VERR_NO_STR_MEMORY;
3155	cbTmp = strlen(pszTmp) + 1;
3156	char pszBasename = (char )RTMemTmpAlloc(cbTmp);
3157	if (!pszBasename)
3158	{
3159	RTStrFree(pszTmp);
3160	return VERR_NO_MEMORY;
3161	}
3162
3163	memcpy(pszBasename, pszTmp, cbTmp);
3164	RTStrFree(pszTmp);
3165
3166	pExtent->pszBasename = pszBasename;
3167
3168	char *pszBasedirectory = RTStrDup(pImage->pszFilename);
3169	if (!pszBasedirectory)
3170	return VERR_NO_STR_MEMORY;
3171	RTPathStripFilename(pszBasedirectory);
3172	char *pszFullname = RTPathJoinA(pszBasedirectory, pExtent->pszBasename);
3173	RTStrFree(pszBasedirectory);
3174	if (!pszFullname)
3175	return VERR_NO_STR_MEMORY;
3176	pExtent->pszFullname = pszFullname;
3177
3178	/* Create file for extent. */
3179	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
3180	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
3181	true /* fCreate */));
3182	if (RT_FAILURE(rc))
3183	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pExtent->pszFullname);
3184
3185	rc = vmdkDescExtInsert(pImage, &pImage->Descriptor, pExtent->enmAccess,
3186	pExtent->cNominalSectors, pExtent->enmType,
3187	pExtent->pszBasename, pExtent->uSectorOffset);
3188	if (RT_FAILURE(rc))
3189	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not insert the extent list into descriptor in '%s'"), pImage->pszFilename);
3190
3191	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage, cbSize,
3192	0 /* fFlags */, NULL, 0, 0);
3193
3194	if (RT_FAILURE(rc))
3195	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
3196
3197	pImage->pExtents = pNewExtents;
3198	pImage->cExtents++;
3199	}
3200	else
3201	rc = VERR_NO_MEMORY;
3202	return rc;
3203	}
3204	/**
3205	* Reads and processes the descriptor embedded in sparse images.
3206	*
3207	* @returns VBox status code.
3208	* @param pImage VMDK image instance.
3209	* @param pFile The sparse file handle.
3210	*/
3211	static int vmdkDescriptorReadSparse(PVMDKIMAGE pImage, PVMDKFILE pFile)
3212	{
3213	/* It's a hosted single-extent image. */
3214	int rc = vmdkCreateExtents(pImage, 1);
3215	if (RT_SUCCESS(rc))
3216	{
3217	/* The opened file is passed to the extent. No separate descriptor
3218	* file, so no need to keep anything open for the image. */
3219	PVMDKEXTENT pExtent = &pImage->pExtents[0];
3220	pExtent->pFile = pFile;
3221	pImage->pFile = NULL;
3222	pExtent->pszFullname = RTPathAbsDup(pImage->pszFilename);
3223	if (RT_LIKELY(pExtent->pszFullname))
3224	{
3225	/* As we're dealing with a monolithic image here, there must
3226	* be a descriptor embedded in the image file. */
3227	rc = vmdkReadBinaryMetaExtent(pImage, pExtent, true /* fMagicAlreadyRead */);
3228	if ( RT_SUCCESS(rc)
3229	&& pExtent->uDescriptorSector
3230	&& pExtent->cDescriptorSectors)
3231	{
3232	/* HACK: extend the descriptor if it is unusually small and it fits in
3233	* the unused space after the image header. Allows opening VMDK files
3234	* with extremely small descriptor in read/write mode.
3235	*
3236	* The previous version introduced a possible regression for VMDK stream
3237	* optimized images from VMware which tend to have only a single sector sized
3238	* descriptor. Increasing the descriptor size resulted in adding the various uuid
3239	* entries required to make it work with VBox but for stream optimized images
3240	* the updated binary header wasn't written to the disk creating a mismatch
3241	* between advertised and real descriptor size.
3242	*
3243	* The descriptor size will be increased even if opened readonly now if there
3244	* enough room but the new value will not be written back to the image.
3245	*/
3246	if ( pExtent->cDescriptorSectors < 3
3247	&& (int64_t)pExtent->uSectorGD - pExtent->uDescriptorSector >= 4
3248	&& (!pExtent->uSectorRGD \|\| (int64_t)pExtent->uSectorRGD - pExtent->uDescriptorSector >= 4))
3249	{
3250	uint64_t cDescriptorSectorsOld = pExtent->cDescriptorSectors;
3251
3252	pExtent->cDescriptorSectors = 4;
3253	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
3254	{
3255	/*
3256	* Update the on disk number now to make sure we don't introduce inconsistencies
3257	* in case of stream optimized images from VMware where the descriptor is just
3258	* one sector big (the binary header is not written to disk for complete
3259	* stream optimized images in vmdkFlushImage()).
3260	*/
3261	uint64_t u64DescSizeNew = RT_H2LE_U64(pExtent->cDescriptorSectors);
3262	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pFile->pStorage,
3263	RT_UOFFSETOF(SparseExtentHeader, descriptorSize),
3264	&u64DescSizeNew, sizeof(u64DescSizeNew));
3265	if (RT_FAILURE(rc))
3266	{
3267	LogFlowFunc(("Increasing the descriptor size failed with %Rrc\n", rc));
3268	/* Restore the old size and carry on. */
3269	pExtent->cDescriptorSectors = cDescriptorSectorsOld;
3270	}
3271	}
3272	}
3273	/* Read the descriptor from the extent. */
3274	pExtent->pDescData = (char *)RTMemAllocZ(VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
3275	if (RT_LIKELY(pExtent->pDescData))
3276	{
3277	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
3278	VMDK_SECTOR2BYTE(pExtent->uDescriptorSector),
3279	pExtent->pDescData,
3280	VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
3281	if (RT_SUCCESS(rc))
3282	{
3283	rc = vmdkParseDescriptor(pImage, pExtent->pDescData,
3284	VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
3285	if ( RT_SUCCESS(rc)
3286	&& ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
3287	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_ASYNC_IO)))
3288	{
3289	rc = vmdkReadMetaExtent(pImage, pExtent);
3290	if (RT_SUCCESS(rc))
3291	{
3292	/* Mark the extent as unclean if opened in read-write mode. */
3293	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3294	&& !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
3295	{
3296	pExtent->fUncleanShutdown = true;
3297	pExtent->fMetaDirty = true;
3298	}
3299	}
3300	}
3301	else if (RT_SUCCESS(rc))
3302	rc = VERR_NOT_SUPPORTED;
3303	}
3304	else
3305	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: read error for descriptor in '%s'"), pExtent->pszFullname);
3306	}
3307	else
3308	rc = VERR_NO_MEMORY;
3309	}
3310	else if (RT_SUCCESS(rc))
3311	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: monolithic image without descriptor in '%s'"), pImage->pszFilename);
3312	}
3313	else
3314	rc = VERR_NO_MEMORY;
3315	}
3316
3317	return rc;
3318	}
3319
3320	/**
3321	* Reads the descriptor from a pure text file.
3322	*
3323	* @returns VBox status code.
3324	* @param pImage VMDK image instance.
3325	* @param pFile The descriptor file handle.
3326	*/
3327	static int vmdkDescriptorReadAscii(PVMDKIMAGE pImage, PVMDKFILE pFile)
3328	{
3329	/* Allocate at least 10K, and make sure that there is 5K free space
3330	* in case new entries need to be added to the descriptor. Never
3331	* allocate more than 128K, because that's no valid descriptor file
3332	* and will result in the correct "truncated read" error handling. */
3333	uint64_t cbFileSize;
3334	int rc = vdIfIoIntFileGetSize(pImage->pIfIo, pFile->pStorage, &cbFileSize);
3335	if ( RT_SUCCESS(rc)
3336	&& cbFileSize >= 50)
3337	{
3338	uint64_t cbSize = cbFileSize;
3339	if (cbSize % VMDK_SECTOR2BYTE(10))
3340	cbSize += VMDK_SECTOR2BYTE(20) - cbSize % VMDK_SECTOR2BYTE(10);
3341	else
3342	cbSize += VMDK_SECTOR2BYTE(10);
3343	cbSize = RT_MIN(cbSize, _128K);
3344	pImage->cbDescAlloc = RT_MAX(VMDK_SECTOR2BYTE(20), cbSize);
3345	pImage->pDescData = (char *)RTMemAllocZ(pImage->cbDescAlloc);
3346	if (RT_LIKELY(pImage->pDescData))
3347	{
3348	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pFile->pStorage, 0, pImage->pDescData,
3349	RT_MIN(pImage->cbDescAlloc, cbFileSize));
3350	if (RT_SUCCESS(rc))
3351	{
3352	#if 0 /** @todo Revisit */
3353	cbRead += sizeof(u32Magic);
3354	if (cbRead == pImage->cbDescAlloc)
3355	{
3356	/* Likely the read is truncated. Better fail a bit too early
3357	* (normally the descriptor is much smaller than our buffer). */
3358	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: cannot read descriptor in '%s'"), pImage->pszFilename);
3359	goto out;
3360	}
3361	#endif
3362	rc = vmdkParseDescriptor(pImage, pImage->pDescData,
3363	pImage->cbDescAlloc);
3364	if (RT_SUCCESS(rc))
3365	{
3366	for (unsigned i = 0; i < pImage->cExtents && RT_SUCCESS(rc); i++)
3367	{
3368	PVMDKEXTENT pExtent = &pImage->pExtents[i];
3369	if (pExtent->pszBasename)
3370	{
3371	/* Hack to figure out whether the specified name in the
3372	* extent descriptor is absolute. Doesn't always work, but
3373	* should be good enough for now. */
3374	char *pszFullname;
3375	/** @todo implement proper path absolute check. */
3376	if (pExtent->pszBasename[0] == RTPATH_SLASH)
3377	{
3378	pszFullname = RTStrDup(pExtent->pszBasename);
3379	if (!pszFullname)
3380	{
3381	rc = VERR_NO_MEMORY;
3382	break;
3383	}
3384	}
3385	else
3386	{
3387	char *pszDirname = RTStrDup(pImage->pszFilename);
3388	if (!pszDirname)
3389	{
3390	rc = VERR_NO_MEMORY;
3391	break;
3392	}
3393	RTPathStripFilename(pszDirname);
3394	pszFullname = RTPathJoinA(pszDirname, pExtent->pszBasename);
3395	RTStrFree(pszDirname);
3396	if (!pszFullname)
3397	{
3398	rc = VERR_NO_STR_MEMORY;
3399	break;
3400	}
3401	}
3402	pExtent->pszFullname = pszFullname;
3403	}
3404	else
3405	pExtent->pszFullname = NULL;
3406
3407	unsigned uOpenFlags = pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0);
3408	switch (pExtent->enmType)
3409	{
3410	case VMDKETYPE_HOSTED_SPARSE:
3411	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
3412	VDOpenFlagsToFileOpenFlags(uOpenFlags, false /* fCreate */));
3413	if (RT_FAILURE(rc))
3414	{
3415	/* Do NOT signal an appropriate error here, as the VD
3416	* layer has the choice of retrying the open if it
3417	* failed. */
3418	break;
3419	}
3420	rc = vmdkReadBinaryMetaExtent(pImage, pExtent,
3421	false /* fMagicAlreadyRead */);
3422	if (RT_FAILURE(rc))
3423	break;
3424	rc = vmdkReadMetaExtent(pImage, pExtent);
3425	if (RT_FAILURE(rc))
3426	break;
3427
3428	/* Mark extent as unclean if opened in read-write mode. */
3429	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
3430	{
3431	pExtent->fUncleanShutdown = true;
3432	pExtent->fMetaDirty = true;
3433	}
3434	break;
3435	case VMDKETYPE_VMFS:
3436	case VMDKETYPE_FLAT:
3437	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
3438	VDOpenFlagsToFileOpenFlags(uOpenFlags, false /* fCreate */));
3439	if (RT_FAILURE(rc))
3440	{
3441	/* Do NOT signal an appropriate error here, as the VD
3442	* layer has the choice of retrying the open if it
3443	* failed. */
3444	break;
3445	}
3446	break;
3447	case VMDKETYPE_ZERO:
3448	/* Nothing to do. */
3449	break;
3450	default:
3451	AssertMsgFailed(("unknown vmdk extent type %d\n", pExtent->enmType));
3452	}
3453	}
3454	}
3455	}
3456	else
3457	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: read error for descriptor in '%s'"), pImage->pszFilename);
3458	}
3459	else
3460	rc = VERR_NO_MEMORY;
3461	}
3462	else if (RT_SUCCESS(rc))
3463	rc = vdIfError(pImage->pIfError, VERR_VD_VMDK_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor in '%s' is too short"), pImage->pszFilename);
3464
3465	return rc;
3466	}
3467
3468	/**
3469	* Read and process the descriptor based on the image type.
3470	*
3471	* @returns VBox status code.
3472	* @param pImage VMDK image instance.
3473	* @param pFile VMDK file handle.
3474	*/
3475	static int vmdkDescriptorRead(PVMDKIMAGE pImage, PVMDKFILE pFile)
3476	{
3477	uint32_t u32Magic;
3478
3479	/* Read magic (if present). */
3480	int rc = vdIfIoIntFileReadSync(pImage->pIfIo, pFile->pStorage, 0,
3481	&u32Magic, sizeof(u32Magic));
3482	if (RT_SUCCESS(rc))
3483	{
3484	/* Handle the file according to its magic number. */
3485	if (RT_LE2H_U32(u32Magic) == VMDK_SPARSE_MAGICNUMBER)
3486	rc = vmdkDescriptorReadSparse(pImage, pFile);
3487	else
3488	rc = vmdkDescriptorReadAscii(pImage, pFile);
3489	}
3490	else
3491	{
3492	vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error reading the magic number in '%s'"), pImage->pszFilename);
3493	rc = VERR_VD_VMDK_INVALID_HEADER;
3494	}
3495
3496	return rc;
3497	}
3498
3499	/**
3500	* Internal: Open an image, constructing all necessary data structures.
3501	*/
3502	static int vmdkOpenImage(PVMDKIMAGE pImage, unsigned uOpenFlags)
3503	{
3504	pImage->uOpenFlags = uOpenFlags;
3505	pImage->pIfError = VDIfErrorGet(pImage->pVDIfsDisk);
3506	pImage->pIfIo = VDIfIoIntGet(pImage->pVDIfsImage);
3507	AssertPtrReturn(pImage->pIfIo, VERR_INVALID_PARAMETER);
3508
3509	/*
3510	* Open the image.
3511	* We don't have to check for asynchronous access because
3512	* we only support raw access and the opened file is a description
3513	* file were no data is stored.
3514	*/
3515	PVMDKFILE pFile;
3516	int rc = vmdkFileOpen(pImage, &pFile, NULL, pImage->pszFilename,
3517	VDOpenFlagsToFileOpenFlags(uOpenFlags, false /* fCreate */));
3518	if (RT_SUCCESS(rc))
3519	{
3520	pImage->pFile = pFile;
3521
3522	rc = vmdkDescriptorRead(pImage, pFile);
3523	if (RT_SUCCESS(rc))
3524	{
3525	/* Determine PCHS geometry if not set. */
3526	if (pImage->PCHSGeometry.cCylinders == 0)
3527	{
3528	uint64_t cCylinders = VMDK_BYTE2SECTOR(pImage->cbSize)
3529	/ pImage->PCHSGeometry.cHeads
3530	/ pImage->PCHSGeometry.cSectors;
3531	pImage->PCHSGeometry.cCylinders = (unsigned)RT_MIN(cCylinders, 16383);
3532	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3533	&& !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
3534	{
3535	rc = vmdkDescSetPCHSGeometry(pImage, &pImage->PCHSGeometry);
3536	AssertRC(rc);
3537	}
3538	}
3539
3540	/* Update the image metadata now in case has changed. */
3541	rc = vmdkFlushImage(pImage, NULL);
3542	if (RT_SUCCESS(rc))
3543	{
3544	/* Figure out a few per-image constants from the extents. */
3545	pImage->cbSize = 0;
3546	for (unsigned i = 0; i < pImage->cExtents; i++)
3547	{
3548	PVMDKEXTENT pExtent = &pImage->pExtents[i];
3549	if (pExtent->enmType == VMDKETYPE_HOSTED_SPARSE)
3550	{
3551	/* Here used to be a check whether the nominal size of an extent
3552	* is a multiple of the grain size. The spec says that this is
3553	* always the case, but unfortunately some files out there in the
3554	* wild violate the spec (e.g. ReactOS 0.3.1). */
3555	}
3556	else if ( pExtent->enmType == VMDKETYPE_FLAT
3557	\|\| pExtent->enmType == VMDKETYPE_ZERO)
3558	pImage->uImageFlags \|= VD_IMAGE_FLAGS_FIXED;
3559
3560	pImage->cbSize += VMDK_SECTOR2BYTE(pExtent->cNominalSectors);
3561	}
3562
3563	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
3564	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3565	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
3566	rc = vmdkAllocateGrainTableCache(pImage);
3567	}
3568	}
3569	}
3570	/* else: Do NOT signal an appropriate error here, as the VD layer has the
3571	* choice of retrying the open if it failed. */
3572
3573	if (RT_SUCCESS(rc))
3574	{
3575	PVDREGIONDESC pRegion = &pImage->RegionList.aRegions[0];
3576	pImage->RegionList.fFlags = 0;
3577	pImage->RegionList.cRegions = 1;
3578
3579	pRegion->offRegion = 0; /* Disk start. */
3580	pRegion->cbBlock = 512;
3581	pRegion->enmDataForm = VDREGIONDATAFORM_RAW;
3582	pRegion->enmMetadataForm = VDREGIONMETADATAFORM_NONE;
3583	pRegion->cbData = 512;
3584	pRegion->cbMetadata = 0;
3585	pRegion->cRegionBlocksOrBytes = pImage->cbSize;
3586	}
3587	else
3588	vmdkFreeImage(pImage, false, false /fFlush/); /* Don't try to flush anything if opening failed. */
3589	return rc;
3590	}
3591
3592	/**
3593	* Frees a raw descriptor.
3594	* @internal
3595	*/
3596	static int vmdkRawDescFree(PVDISKRAW pRawDesc)
3597	{
3598	if (!pRawDesc)
3599	return VINF_SUCCESS;
3600
3601	RTStrFree(pRawDesc->pszRawDisk);
3602	pRawDesc->pszRawDisk = NULL;
3603
3604	/* Partitions: */
3605	for (unsigned i = 0; i < pRawDesc->cPartDescs; i++)
3606	{
3607	RTStrFree(pRawDesc->pPartDescs[i].pszRawDevice);
3608	pRawDesc->pPartDescs[i].pszRawDevice = NULL;
3609
3610	RTMemFree(pRawDesc->pPartDescs[i].pvPartitionData);
3611	pRawDesc->pPartDescs[i].pvPartitionData = NULL;
3612	}
3613
3614	RTMemFree(pRawDesc->pPartDescs);
3615	pRawDesc->pPartDescs = NULL;
3616
3617	RTMemFree(pRawDesc);
3618	return VINF_SUCCESS;
3619	}
3620
3621	/**
3622	* Helper that grows the raw partition descriptor table by @a cToAdd entries,
3623	* returning the pointer to the first new entry.
3624	* @internal
3625	*/
3626	static int vmdkRawDescAppendPartDesc(PVMDKIMAGE pImage, PVDISKRAW pRawDesc, uint32_t cToAdd, PVDISKRAWPARTDESC *ppRet)
3627	{
3628	uint32_t const cOld = pRawDesc->cPartDescs;
3629	uint32_t const cNew = cOld + cToAdd;
3630	PVDISKRAWPARTDESC paNew = (PVDISKRAWPARTDESC)RTMemReallocZ(pRawDesc->pPartDescs,
3631	cOld * sizeof(pRawDesc->pPartDescs[0]),
3632	cNew * sizeof(pRawDesc->pPartDescs[0]));
3633	if (paNew)
3634	{
3635	pRawDesc->cPartDescs = cNew;
3636	pRawDesc->pPartDescs = paNew;
3637
3638	*ppRet = &paNew[cOld];
3639	return VINF_SUCCESS;
3640	}
3641	*ppRet = NULL;
3642	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS,
3643	N_("VMDK: Image path: '%s'. Out of memory growing the partition descriptors (%u -> %u)."),
3644	pImage->pszFilename, cOld, cNew);
3645	}
3646
3647	/**
3648	* @callback_method_impl{FNRTSORTCMP}
3649	*/
3650	static DECLCALLBACK(int) vmdkRawDescPartComp(void const pvElement1, void const pvElement2, void *pvUser)
3651	{
3652	RT_NOREF(pvUser);
3653	int64_t const iDelta = ((PVDISKRAWPARTDESC)pvElement1)->offStartInVDisk - ((PVDISKRAWPARTDESC)pvElement2)->offStartInVDisk;
3654	return iDelta < 0 ? -1 : iDelta > 0 ? 1 : 0;
3655	}
3656
3657	/**
3658	* Post processes the partition descriptors.
3659	*
3660	* Sorts them and check that they don't overlap.
3661	*/
3662	static int vmdkRawDescPostProcessPartitions(PVMDKIMAGE pImage, PVDISKRAW pRawDesc, uint64_t cbSize)
3663	{
3664	/*
3665	* Sort data areas in ascending order of start.
3666	*/
3667	RTSortShell(pRawDesc->pPartDescs, pRawDesc->cPartDescs, sizeof(pRawDesc->pPartDescs[0]), vmdkRawDescPartComp, NULL);
3668
3669	/*
3670	* Check that we don't have overlapping descriptors. If we do, that's an
3671	* indication that the drive is corrupt or that the RTDvm code is buggy.
3672	*/
3673	VDISKRAWPARTDESC const *paPartDescs = pRawDesc->pPartDescs;
3674	for (uint32_t i = 0; i < pRawDesc->cPartDescs; i++)
3675	{
3676	uint64_t offLast = paPartDescs[i].offStartInVDisk + paPartDescs[i].cbData;
3677	if (offLast <= paPartDescs[i].offStartInVDisk)
3678	return vdIfError(pImage->pIfError, VERR_FILESYSTEM_CORRUPT /?/, RT_SRC_POS,
3679	N_("VMDK: Image path: '%s'. Bogus partition descriptor #%u (%#RX64 LB %#RX64%s): Wrap around or zero"),
3680	pImage->pszFilename, i, paPartDescs[i].offStartInVDisk, paPartDescs[i].cbData,
3681	paPartDescs[i].pvPartitionData ? " (data)" : "");
3682	offLast -= 1;
3683
3684	if (i + 1 < pRawDesc->cPartDescs && offLast >= paPartDescs[i + 1].offStartInVDisk)
3685	return vdIfError(pImage->pIfError, VERR_FILESYSTEM_CORRUPT /?/, RT_SRC_POS,
3686	N_("VMDK: Image path: '%s'. Partition descriptor #%u (%#RX64 LB %#RX64%s) overlaps with the next (%#RX64 LB %#RX64%s)"),
3687	pImage->pszFilename, i, paPartDescs[i].offStartInVDisk, paPartDescs[i].cbData,
3688	paPartDescs[i].pvPartitionData ? " (data)" : "", paPartDescs[i + 1].offStartInVDisk,
3689	paPartDescs[i + 1].cbData, paPartDescs[i + 1].pvPartitionData ? " (data)" : "");
3690	if (offLast >= cbSize)
3691	return vdIfError(pImage->pIfError, VERR_FILESYSTEM_CORRUPT /?/, RT_SRC_POS,
3692	N_("VMDK: Image path: '%s'. Partition descriptor #%u (%#RX64 LB %#RX64%s) goes beyond the end of the drive (%#RX64)"),
3693	pImage->pszFilename, i, paPartDescs[i].offStartInVDisk, paPartDescs[i].cbData,
3694	paPartDescs[i].pvPartitionData ? " (data)" : "", cbSize);
3695	}
3696
3697	return VINF_SUCCESS;
3698	}
3699
3700
3701	#ifdef RT_OS_LINUX
3702	/**
3703	* Searches the dir specified in @a pszBlockDevDir for subdirectories with a
3704	* 'dev' file matching @a uDevToLocate.
3705	*
3706	* This is used both
3707	*
3708	* @returns IPRT status code, errors have been reported properly.
3709	* @param pImage For error reporting.
3710	* @param pszBlockDevDir Input: Path to the directory search under.
3711	* Output: Path to the directory containing information
3712	* for @a uDevToLocate.
3713	* @param cbBlockDevDir The size of the buffer @a pszBlockDevDir points to.
3714	* @param uDevToLocate The device number of the block device info dir to
3715	* locate.
3716	* @param pszDevToLocate For error reporting.
3717	*/
3718	static int vmdkFindSysBlockDevPath(PVMDKIMAGE pImage, char *pszBlockDevDir, size_t cbBlockDevDir,
3719	dev_t uDevToLocate, const char *pszDevToLocate)
3720	{
3721	size_t const cchDir = RTPathEnsureTrailingSeparator(pszBlockDevDir, cbBlockDevDir);
3722	AssertReturn(cchDir > 0, VERR_BUFFER_OVERFLOW);
3723
3724	RTDIR hDir = NIL_RTDIR;
3725	int rc = RTDirOpen(&hDir, pszBlockDevDir);
3726	if (RT_SUCCESS(rc))
3727	{
3728	for (;;)
3729	{
3730	RTDIRENTRY Entry;
3731	rc = RTDirRead(hDir, &Entry, NULL);
3732	if (RT_SUCCESS(rc))
3733	{
3734	/* We're interested in directories and symlinks. */
3735	if ( Entry.enmType == RTDIRENTRYTYPE_DIRECTORY
3736	\|\| Entry.enmType == RTDIRENTRYTYPE_SYMLINK
3737	\|\| Entry.enmType == RTDIRENTRYTYPE_UNKNOWN)
3738	{
3739	rc = RTStrCopy(&pszBlockDevDir[cchDir], cbBlockDevDir - cchDir, Entry.szName);
3740	AssertContinue(RT_SUCCESS(rc)); /* should not happen! */
3741
3742	dev_t uThisDevNo = ~uDevToLocate;
3743	rc = RTLinuxSysFsReadDevNumFile(&uThisDevNo, "%s/dev", pszBlockDevDir);
3744	if (RT_SUCCESS(rc) && uThisDevNo == uDevToLocate)
3745	break;
3746	}
3747	}
3748	else
3749	{
3750	pszBlockDevDir[cchDir] = '\0';
3751	if (rc == VERR_NO_MORE_FILES)
3752	rc = vdIfError(pImage->pIfError, VERR_NOT_FOUND, RT_SRC_POS,
3753	N_("VMDK: Image path: '%s'. Failed to locate device corresponding to '%s' under '%s'"),
3754	pImage->pszFilename, pszDevToLocate, pszBlockDevDir);
3755	else
3756	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
3757	N_("VMDK: Image path: '%s'. RTDirRead failed enumerating '%s': %Rrc"),
3758	pImage->pszFilename, pszBlockDevDir, rc);
3759	break;
3760	}
3761	}
3762	RTDirClose(hDir);
3763	}
3764	else
3765	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
3766	N_("VMDK: Image path: '%s'. Failed to open dir '%s' for listing: %Rrc"),
3767	pImage->pszFilename, pszBlockDevDir, rc);
3768	return rc;
3769	}
3770	#endif /* RT_OS_LINUX */
3771
3772	#ifdef RT_OS_FREEBSD
3773
3774
3775	/**
3776	* Reads the config data from the provider and returns offset and size
3777	*
3778	* @return IPRT status code
3779	* @param pProvider GEOM provider representing partition
3780	* @param pcbOffset Placeholder for the offset of the partition
3781	* @param pcbSize Placeholder for the size of the partition
3782	*/
3783	static int vmdkReadPartitionsParamsFromProvider(gprovider pProvider, uint64_t pcbOffset, uint64_t *pcbSize)
3784	{
3785	gconfig *pConfEntry;
3786	int rc = VERR_NOT_FOUND;
3787
3788	/*
3789	* Required parameters are located in the list containing key/value pairs.
3790	* Both key and value are in text form. Manuals tells nothing about the fact
3791	* that the both parameters should be present in the list. Thus, there are
3792	* cases when only one parameter is presented. To handle such cases we treat
3793	* absent params as zero allowing the caller decide the case is either correct
3794	* or an error.
3795	*/
3796	uint64_t cbOffset = 0;
3797	uint64_t cbSize = 0;
3798	LIST_FOREACH(pConfEntry, &pProvider->lg_config, lg_config)
3799	{
3800	if (RTStrCmp(pConfEntry->lg_name, "offset") == 0)
3801	{
3802	cbOffset = RTStrToUInt64(pConfEntry->lg_val);
3803	rc = VINF_SUCCESS;
3804	}
3805	else if (RTStrCmp(pConfEntry->lg_name, "length") == 0)
3806	{
3807	cbSize = RTStrToUInt64(pConfEntry->lg_val);
3808	rc = VINF_SUCCESS;
3809	}
3810	}
3811	if (RT_SUCCESS(rc))
3812	{
3813	*pcbOffset = cbOffset;
3814	*pcbSize = cbSize;
3815	}
3816	return rc;
3817	}
3818
3819
3820	/**
3821	* Searches the partition specified by name and calculates its size and absolute offset.
3822	*
3823	* @return IPRT status code.
3824	* @param pParentClass Class containing pParentGeom
3825	* @param pszParentGeomName Name of the parent geom where we are looking for provider
3826	* @param pszProviderName Name of the provider we are looking for
3827	* @param pcbAbsoluteOffset Placeholder for the absolute offset of the partition, i.e. offset from the beginning of the disk
3828	* @param psbSize Placeholder for the size of the partition.
3829	*/
3830	static int vmdkFindPartitionParamsByName(gclass pParentClass, const char pszParentGeomName, const char *pszProviderName,
3831	uint64_t pcbAbsoluteOffset, uint64_t pcbSize)
3832	{
3833	AssertReturn(pParentClass, VERR_INVALID_PARAMETER);
3834	AssertReturn(pszParentGeomName, VERR_INVALID_PARAMETER);
3835	AssertReturn(pszProviderName, VERR_INVALID_PARAMETER);
3836	AssertReturn(pcbAbsoluteOffset, VERR_INVALID_PARAMETER);
3837	AssertReturn(pcbSize, VERR_INVALID_PARAMETER);
3838
3839	ggeom *pParentGeom;
3840	int rc = VERR_NOT_FOUND;
3841	LIST_FOREACH(pParentGeom, &pParentClass->lg_geom, lg_geom)
3842	{
3843	if (RTStrCmp(pParentGeom->lg_name, pszParentGeomName) == 0)
3844	{
3845	rc = VINF_SUCCESS;
3846	break;
3847	}
3848	}
3849	if (RT_FAILURE(rc))
3850	return rc;
3851
3852	gprovider *pProvider;
3853	/*
3854	* First, go over providers without handling EBR or BSDLabel
3855	* partitions for case when looking provider is child
3856	* of the givng geom, to reduce searching time
3857	*/
3858	LIST_FOREACH(pProvider, &pParentGeom->lg_provider, lg_provider)
3859	{
3860	if (RTStrCmp(pProvider->lg_name, pszProviderName) == 0)
3861	return vmdkReadPartitionsParamsFromProvider(pProvider, pcbAbsoluteOffset, pcbSize);
3862	}
3863
3864	/*
3865	* No provider found. Go over the parent geom again
3866	* and make recursions if geom represents EBR or BSDLabel.
3867	* In this case given parent geom contains only EBR or BSDLabel
3868	* partition itself and their own partitions are in the separate
3869	* geoms. Also, partition offsets are relative to geom, so
3870	* we have to add offset from child provider with parent geoms
3871	* provider
3872	*/
3873
3874	LIST_FOREACH(pProvider, &pParentGeom->lg_provider, lg_provider)
3875	{
3876	uint64_t cbOffset = 0;
3877	uint64_t cbSize = 0;
3878	rc = vmdkReadPartitionsParamsFromProvider(pProvider, &cbOffset, &cbSize);
3879	if (RT_FAILURE(rc))
3880	return rc;
3881
3882	uint64_t cbProviderOffset = 0;
3883	uint64_t cbProviderSize = 0;
3884	rc = vmdkFindPartitionParamsByName(pParentClass, pProvider->lg_name, pszProviderName, &cbProviderOffset, &cbProviderSize);
3885	if (RT_SUCCESS(rc))
3886	{
3887	*pcbAbsoluteOffset = cbOffset + cbProviderOffset;
3888	*pcbSize = cbProviderSize;
3889	return rc;
3890	}
3891	}
3892
3893	return VERR_NOT_FOUND;
3894	}
3895	#endif
3896
3897
3898	/**
3899	* Attempts to verify the raw partition path.
3900	*
3901	* We don't want to trust RTDvm and the partition device node morphing blindly.
3902	*/
3903	static int vmdkRawDescVerifyPartitionPath(PVMDKIMAGE pImage, PVDISKRAWPARTDESC pPartDesc, uint32_t idxPartition,
3904	const char *pszRawDrive, RTFILE hRawDrive, uint32_t cbSector, RTDVMVOLUME hVol)
3905	{
3906	RT_NOREF(pImage, pPartDesc, idxPartition, pszRawDrive, hRawDrive, cbSector, hVol);
3907
3908	/*
3909	* Try open the raw partition device.
3910	*/
3911	RTFILE hRawPart = NIL_RTFILE;
3912	int rc = RTFileOpen(&hRawPart, pPartDesc->pszRawDevice, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE);
3913	if (RT_FAILURE(rc))
3914	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
3915	N_("VMDK: Image path: '%s'. Failed to open partition #%u on '%s' via '%s' (%Rrc)"),
3916	pImage->pszFilename, idxPartition, pszRawDrive, pPartDesc->pszRawDevice, rc);
3917
3918	/*
3919	* Compare the partition UUID if we can get it.
3920	*/
3921	#ifdef RT_OS_WINDOWS
3922	DWORD cbReturned;
3923
3924	/* 1. Get the device numbers for both handles, they should have the same disk. */
3925	STORAGE_DEVICE_NUMBER DevNum1;
3926	RT_ZERO(DevNum1);
3927	if (!DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_STORAGE_GET_DEVICE_NUMBER,
3928	NULL /pvInBuffer/, 0 /cbInBuffer/, &DevNum1, sizeof(DevNum1), &cbReturned, NULL /pOverlapped/))
3929	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
3930	N_("VMDK: Image path: '%s'. IOCTL_STORAGE_GET_DEVICE_NUMBER failed on '%s': %u"),
3931	pImage->pszFilename, pszRawDrive, GetLastError());
3932
3933	STORAGE_DEVICE_NUMBER DevNum2;
3934	RT_ZERO(DevNum2);
3935	if (!DeviceIoControl((HANDLE)RTFileToNative(hRawPart), IOCTL_STORAGE_GET_DEVICE_NUMBER,
3936	NULL /pvInBuffer/, 0 /cbInBuffer/, &DevNum2, sizeof(DevNum2), &cbReturned, NULL /pOverlapped/))
3937	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
3938	N_("VMDK: Image path: '%s'. IOCTL_STORAGE_GET_DEVICE_NUMBER failed on '%s': %u"),
3939	pImage->pszFilename, pPartDesc->pszRawDevice, GetLastError());
3940	if ( RT_SUCCESS(rc)
3941	&& ( DevNum1.DeviceNumber != DevNum2.DeviceNumber
3942	\|\| DevNum1.DeviceType != DevNum2.DeviceType))
3943	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3944	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s' (%#x != %#x \|\| %#x != %#x)"),
3945	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
3946	DevNum1.DeviceNumber, DevNum2.DeviceNumber, DevNum1.DeviceType, DevNum2.DeviceType);
3947	if (RT_SUCCESS(rc))
3948	{
3949	/* Get the partitions from the raw drive and match up with the volume info
3950	from RTDvm. The partition number is found in DevNum2. */
3951	DWORD cbNeeded = 0;
3952	if ( DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_DISK_GET_DRIVE_LAYOUT_EX,
3953	NULL /pvInBuffer/, 0 /cbInBuffer/, NULL, 0, &cbNeeded, NULL /pOverlapped/)
3954	\|\| cbNeeded < RT_UOFFSETOF_DYN(DRIVE_LAYOUT_INFORMATION_EX, PartitionEntry[1]))
3955	cbNeeded = RT_UOFFSETOF_DYN(DRIVE_LAYOUT_INFORMATION_EX, PartitionEntry[64]);
3956	cbNeeded += sizeof(PARTITION_INFORMATION_EX) * 2; /* just in case */
3957	DRIVE_LAYOUT_INFORMATION_EX pLayout = (DRIVE_LAYOUT_INFORMATION_EX )RTMemTmpAllocZ(cbNeeded);
3958	if (pLayout)
3959	{
3960	cbReturned = 0;
3961	if (DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_DISK_GET_DRIVE_LAYOUT_EX,
3962	NULL /pvInBuffer/, 0 /cbInBuffer/, pLayout, cbNeeded, &cbReturned, NULL /pOverlapped/))
3963	{
3964	/* Find the entry with the given partition number (it's not an index, array contains empty MBR entries ++). */
3965	unsigned iEntry = 0;
3966	while ( iEntry < pLayout->PartitionCount
3967	&& pLayout->PartitionEntry[iEntry].PartitionNumber != DevNum2.PartitionNumber)
3968	iEntry++;
3969	if (iEntry < pLayout->PartitionCount)
3970	{
3971	/* Compare the basics */
3972	PARTITION_INFORMATION_EX const * const pLayoutEntry = &pLayout->PartitionEntry[iEntry];
3973	if (pLayoutEntry->StartingOffset.QuadPart != (int64_t)pPartDesc->offStartInVDisk)
3974	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3975	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': StartingOffset %RU64, expected %RU64"),
3976	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
3977	pLayoutEntry->StartingOffset.QuadPart, pPartDesc->offStartInVDisk);
3978	else if (pLayoutEntry->PartitionLength.QuadPart != (int64_t)pPartDesc->cbData)
3979	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3980	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': PartitionLength %RU64, expected %RU64"),
3981	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
3982	pLayoutEntry->PartitionLength.QuadPart, pPartDesc->cbData);
3983	/** @todo We could compare the MBR type, GPT type and ID. */
3984	RT_NOREF(hVol);
3985	}
3986	else
3987	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
3988	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': PartitionCount (%#x vs %#x)"),
3989	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
3990	DevNum2.PartitionNumber, pLayout->PartitionCount);
3991	# ifndef LOG_ENABLED
3992	if (RT_FAILURE(rc))
3993	# endif
3994	{
3995	LogRel(("VMDK: Windows reports %u partitions for '%s':\n", pLayout->PartitionCount, pszRawDrive));
3996	PARTITION_INFORMATION_EX const *pEntry = &pLayout->PartitionEntry[0];
3997	for (DWORD i = 0; i < pLayout->PartitionCount; i++, pEntry++)
3998	{
3999	LogRel(("VMDK: #%u/%u: %016RU64 LB %016RU64 style=%d rewrite=%d",
4000	i, pEntry->PartitionNumber, pEntry->StartingOffset.QuadPart, pEntry->PartitionLength.QuadPart,
4001	pEntry->PartitionStyle, pEntry->RewritePartition));
4002	if (pEntry->PartitionStyle == PARTITION_STYLE_MBR)
4003	LogRel((" type=%#x boot=%d rec=%d hidden=%u\n", pEntry->Mbr.PartitionType, pEntry->Mbr.BootIndicator,
4004	pEntry->Mbr.RecognizedPartition, pEntry->Mbr.HiddenSectors));
4005	else if (pEntry->PartitionStyle == PARTITION_STYLE_GPT)
4006	LogRel((" type=%RTuuid id=%RTuuid aatrib=%RX64 name=%.36ls\n", &pEntry->Gpt.PartitionType,
4007	&pEntry->Gpt.PartitionId, pEntry->Gpt.Attributes, &pEntry->Gpt.Name[0]));
4008	else
4009	LogRel(("\n"));
4010	}
4011	LogRel(("VMDK: Looked for partition #%u (%u, '%s') at %RU64 LB %RU64\n", DevNum2.PartitionNumber,
4012	idxPartition, pPartDesc->pszRawDevice, pPartDesc->offStartInVDisk, pPartDesc->cbData));
4013	}
4014	}
4015	else
4016	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
4017	N_("VMDK: Image path: '%s'. IOCTL_DISK_GET_DRIVE_LAYOUT_EX failed on '%s': %u (cb %u, cbRet %u)"),
4018	pImage->pszFilename, pPartDesc->pszRawDevice, GetLastError(), cbNeeded, cbReturned);
4019	RTMemTmpFree(pLayout);
4020	}
4021	else
4022	rc = VERR_NO_TMP_MEMORY;
4023	}
4024
4025	#elif defined(RT_OS_LINUX)
4026	RT_NOREF(hVol);
4027
4028	/* Stat the two devices first to get their device numbers. (We probably
4029	could make some assumptions here about the major & minor number assignments
4030	for legacy nodes, but it doesn't hold up for nvme, so we'll skip that.) */
4031	struct stat StDrive, StPart;
4032	if (fstat((int)RTFileToNative(hRawDrive), &StDrive) != 0)
4033	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4034	N_("VMDK: Image path: '%s'. fstat failed on '%s': %d"), pImage->pszFilename, pszRawDrive, errno);
4035	else if (fstat((int)RTFileToNative(hRawPart), &StPart) != 0)
4036	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4037	N_("VMDK: Image path: '%s'. fstat failed on '%s': %d"), pImage->pszFilename, pPartDesc->pszRawDevice, errno);
4038	else
4039	{
4040	/* Scan the directories immediately under /sys/block/ for one with a
4041	'dev' file matching the drive's device number: */
4042	char szSysPath[RTPATH_MAX];
4043	rc = RTLinuxConstructPath(szSysPath, sizeof(szSysPath), "block/");
4044	AssertRCReturn(rc, rc); /* this shall not fail */
4045	if (RTDirExists(szSysPath))
4046	{
4047	rc = vmdkFindSysBlockDevPath(pImage, szSysPath, sizeof(szSysPath), StDrive.st_rdev, pszRawDrive);
4048
4049	/* Now, scan the directories under that again for a partition device
4050	matching the hRawPart device's number: */
4051	if (RT_SUCCESS(rc))
4052	rc = vmdkFindSysBlockDevPath(pImage, szSysPath, sizeof(szSysPath), StPart.st_rdev, pPartDesc->pszRawDevice);
4053
4054	/* Having found the /sys/block/device/partition/ path, we can finally
4055	read the partition attributes and compare with hVol. */
4056	if (RT_SUCCESS(rc))
4057	{
4058	/* partition number: */
4059	int64_t iLnxPartition = 0;
4060	rc = RTLinuxSysFsReadIntFile(10, &iLnxPartition, "%s/partition", szSysPath);
4061	if (RT_SUCCESS(rc) && iLnxPartition != idxPartition)
4062	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4063	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Partition number %RI64, expected %RU32"),
4064	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, iLnxPartition, idxPartition);
4065	/* else: ignore failure? */
4066
4067	/* start offset: */
4068	uint32_t const cbLnxSector = 512; /* It's hardcoded in the Linux kernel */
4069	if (RT_SUCCESS(rc))
4070	{
4071	int64_t offLnxStart = -1;
4072	rc = RTLinuxSysFsReadIntFile(10, &offLnxStart, "%s/start", szSysPath);
4073	offLnxStart *= cbLnxSector;
4074	if (RT_SUCCESS(rc) && offLnxStart != (int64_t)pPartDesc->offStartInVDisk)
4075	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4076	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RI64, expected %RU64"),
4077	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, offLnxStart, pPartDesc->offStartInVDisk);
4078	/* else: ignore failure? */
4079	}
4080
4081	/* the size: */
4082	if (RT_SUCCESS(rc))
4083	{
4084	int64_t cbLnxData = -1;
4085	rc = RTLinuxSysFsReadIntFile(10, &cbLnxData, "%s/size", szSysPath);
4086	cbLnxData *= cbLnxSector;
4087	if (RT_SUCCESS(rc) && cbLnxData != (int64_t)pPartDesc->cbData)
4088	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4089	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RI64, expected %RU64"),
4090	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbLnxData, pPartDesc->cbData);
4091	/* else: ignore failure? */
4092	}
4093	}
4094	}
4095	/* else: We've got nothing to work on, so only do content comparison. */
4096	}
4097
4098	#elif defined(RT_OS_FREEBSD)
4099	char szDriveDevName[256];
4100	char* pszDevName = fdevname_r(RTFileToNative(hRawDrive), szDriveDevName, 256);
4101	if (pszDevName == NULL)
4102	rc = vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4103	N_("VMDK: Image path: '%s'. '%s' is not a drive path"), pImage->pszFilename, pszRawDrive);
4104	char szPartDevName[256];
4105	if (RT_SUCCESS(rc))
4106	{
4107	pszDevName = fdevname_r(RTFileToNative(hRawPart), szPartDevName, 256);
4108	if (pszDevName == NULL)
4109	rc = vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4110	N_("VMDK: Image path: '%s'. '%s' is not a partition path"), pImage->pszFilename, pPartDesc->pszRawDevice);
4111	}
4112	if (RT_SUCCESS(rc))
4113	{
4114	gmesh geomMesh;
4115	int err = geom_gettree(&geomMesh);
4116	if (err == 0)
4117	{
4118	/* Find root class containg partitions info */
4119	gclass* pPartClass;
4120	LIST_FOREACH(pPartClass, &geomMesh.lg_class, lg_class)
4121	{
4122	if (RTStrCmp(pPartClass->lg_name, "PART") == 0)
4123	break;
4124	}
4125	if (pPartClass == NULL \|\| RTStrCmp(pPartClass->lg_name, "PART") != 0)
4126	rc = vdIfError(pImage->pIfError, VERR_GENERAL_FAILURE, RT_SRC_POS,
4127	N_("VMDK: Image path: '%s'. 'PART' class not found in the GEOM tree"), pImage->pszFilename);
4128
4129
4130	if (RT_SUCCESS(rc))
4131	{
4132	/* Find provider representing partition device */
4133	uint64_t cbOffset;
4134	uint64_t cbSize;
4135	rc = vmdkFindPartitionParamsByName(pPartClass, szDriveDevName, szPartDevName, &cbOffset, &cbSize);
4136	if (RT_SUCCESS(rc))
4137	{
4138	if (cbOffset != pPartDesc->offStartInVDisk)
4139	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4140	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RU64, expected %RU64"),
4141	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbOffset, pPartDesc->offStartInVDisk);
4142	if (cbSize != pPartDesc->cbData)
4143	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4144	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RU64, expected %RU64"),
4145	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbSize, pPartDesc->cbData);
4146	}
4147	else
4148	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4149	N_("VMDK: Image path: '%s'. Error getting geom provider for the partition '%s' of the drive '%s' in the GEOM tree: %Rrc"),
4150	pImage->pszFilename, pPartDesc->pszRawDevice, pszRawDrive, rc);
4151	}
4152
4153	geom_deletetree(&geomMesh);
4154	}
4155	else
4156	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(err), RT_SRC_POS,
4157	N_("VMDK: Image path: '%s'. geom_gettree failed: %d"), pImage->pszFilename, err);
4158	}
4159
4160	#elif defined(RT_OS_SOLARIS)
4161	RT_NOREF(hVol);
4162
4163	dk_cinfo dkiDriveInfo;
4164	dk_cinfo dkiPartInfo;
4165	if (ioctl(RTFileToNative(hRawDrive), DKIOCINFO, (caddr_t)&dkiDriveInfo) == -1)
4166	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4167	N_("VMDK: Image path: '%s'. DKIOCINFO failed on '%s': %d"), pImage->pszFilename, pszRawDrive, errno);
4168	else if (ioctl(RTFileToNative(hRawPart), DKIOCINFO, (caddr_t)&dkiPartInfo) == -1)
4169	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4170	N_("VMDK: Image path: '%s'. DKIOCINFO failed on '%s': %d"), pImage->pszFilename, pszRawDrive, errno);
4171	else if ( dkiDriveInfo.dki_ctype != dkiPartInfo.dki_ctype
4172	\|\| dkiDriveInfo.dki_cnum != dkiPartInfo.dki_cnum
4173	\|\| dkiDriveInfo.dki_addr != dkiPartInfo.dki_addr
4174	\|\| dkiDriveInfo.dki_unit != dkiPartInfo.dki_unit
4175	\|\| dkiDriveInfo.dki_slave != dkiPartInfo.dki_slave)
4176	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4177	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s' (%#x != %#x \|\| %#x != %#x \|\| %#x != %#x \|\| %#x != %#x \|\| %#x != %#x)"),
4178	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4179	dkiDriveInfo.dki_ctype, dkiPartInfo.dki_ctype, dkiDriveInfo.dki_cnum, dkiPartInfo.dki_cnum,
4180	dkiDriveInfo.dki_addr, dkiPartInfo.dki_addr, dkiDriveInfo.dki_unit, dkiPartInfo.dki_unit,
4181	dkiDriveInfo.dki_slave, dkiPartInfo.dki_slave);
4182	else
4183	{
4184	uint64_t cbOffset = 0;
4185	uint64_t cbSize = 0;
4186	dk_gpt *pEfi = NULL;
4187	int idxEfiPart = efi_alloc_and_read(RTFileToNative(hRawPart), &pEfi);
4188	if (idxEfiPart >= 0)
4189	{
4190	if ((uint32_t)dkiPartInfo.dki_partition + 1 == idxPartition)
4191	{
4192	cbOffset = pEfi->efi_parts[idxEfiPart].p_start * pEfi->efi_lbasize;
4193	cbSize = pEfi->efi_parts[idxEfiPart].p_size * pEfi->efi_lbasize;
4194	}
4195	else
4196	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4197	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s' (%#x != %#x)"),
4198	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4199	idxPartition, (uint32_t)dkiPartInfo.dki_partition + 1);
4200	efi_free(pEfi);
4201	}
4202	else
4203	{
4204	/*
4205	* Manual says the efi_alloc_and_read returns VT_EINVAL if no EFI partition table found.
4206	* Actually, the function returns any error, e.g. VT_ERROR. Thus, we are not sure, is it
4207	* real error or just no EFI table found. Therefore, let's try to obtain partition info
4208	* using another way. If there is an error, it returns errno which will be handled below.
4209	*/
4210
4211	uint32_t numPartition = (uint32_t)dkiPartInfo.dki_partition;
4212	if (numPartition > NDKMAP)
4213	numPartition -= NDKMAP;
4214	if (numPartition != idxPartition)
4215	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4216	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s' (%#x != %#x)"),
4217	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4218	idxPartition, numPartition);
4219	else
4220	{
4221	dk_minfo_ext mediaInfo;
4222	if (ioctl(RTFileToNative(hRawPart), DKIOCGMEDIAINFOEXT, (caddr_t)&mediaInfo) == -1)
4223	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4224	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s'. Can not obtain partition info: %d"),
4225	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4226	else
4227	{
4228	extpart_info extPartInfo;
4229	if (ioctl(RTFileToNative(hRawPart), DKIOCEXTPARTINFO, (caddr_t)&extPartInfo) != -1)
4230	{
4231	cbOffset = (uint64_t)extPartInfo.p_start * mediaInfo.dki_lbsize;
4232	cbSize = (uint64_t)extPartInfo.p_length * mediaInfo.dki_lbsize;
4233	}
4234	else
4235	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4236	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s'. Can not obtain partition info: %d"),
4237	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4238	}
4239	}
4240	}
4241	if (RT_SUCCESS(rc) && cbOffset != pPartDesc->offStartInVDisk)
4242	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4243	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RI64, expected %RU64"),
4244	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbOffset, pPartDesc->offStartInVDisk);
4245
4246	if (RT_SUCCESS(rc) && cbSize != pPartDesc->cbData)
4247	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4248	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RI64, expected %RU64"),
4249	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbSize, pPartDesc->cbData);
4250	}
4251
4252	#elif defined(RT_OS_DARWIN)
4253	/* Stat the drive get its device number. */
4254	struct stat StDrive;
4255	if (fstat((int)RTFileToNative(hRawDrive), &StDrive) != 0)
4256	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4257	N_("VMDK: Image path: '%s'. fstat failed on '%s' (errno=%d)"), pImage->pszFilename, pszRawDrive, errno);
4258	else
4259	{
4260	if (ioctl(RTFileToNative(hRawPart), DKIOCLOCKPHYSICALEXTENTS, NULL) == -1)
4261	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4262	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to lock the partition (errno=%d)"),
4263	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4264	else
4265	{
4266	uint32_t cbBlockSize = 0;
4267	uint64_t cbOffset = 0;
4268	uint64_t cbSize = 0;
4269	if (ioctl(RTFileToNative(hRawPart), DKIOCGETBLOCKSIZE, (caddr_t)&cbBlockSize) == -1)
4270	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4271	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain the sector size of the partition (errno=%d)"),
4272	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4273	else if (ioctl(RTFileToNative(hRawPart), DKIOCGETBASE, (caddr_t)&cbOffset) == -1)
4274	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4275	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain the start offset of the partition (errno=%d)"),
4276	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4277	else if (ioctl(RTFileToNative(hRawPart), DKIOCGETBLOCKCOUNT, (caddr_t)&cbSize) == -1)
4278	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4279	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain the size of the partition (errno=%d)"),
4280	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4281	else
4282	{
4283	cbSize *= (uint64_t)cbBlockSize;
4284	dk_physical_extent_t dkPartExtent = {0};
4285	dkPartExtent.offset = 0;
4286	dkPartExtent.length = cbSize;
4287	if (ioctl(RTFileToNative(hRawPart), DKIOCGETPHYSICALEXTENT, (caddr_t)&dkPartExtent) == -1)
4288	rc = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4289	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to obtain partition info (errno=%d)"),
4290	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4291	else
4292	{
4293	if (dkPartExtent.dev != StDrive.st_rdev)
4294	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4295	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Drive does not contain the partition"),
4296	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive);
4297	else if (cbOffset != pPartDesc->offStartInVDisk)
4298	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4299	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Start offset %RU64, expected %RU64"),
4300	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbOffset, pPartDesc->offStartInVDisk);
4301	else if (cbSize != pPartDesc->cbData)
4302	rc = vdIfError(pImage->pIfError, VERR_MISMATCH, RT_SRC_POS,
4303	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s': Size %RU64, expected %RU64"),
4304	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cbSize, pPartDesc->cbData);
4305	}
4306	}
4307
4308	if (ioctl(RTFileToNative(hRawPart), DKIOCUNLOCKPHYSICALEXTENTS, NULL) == -1)
4309	{
4310	int rc2 = vdIfError(pImage->pIfError, RTErrConvertFromErrno(errno), RT_SRC_POS,
4311	N_("VMDK: Image path: '%s'. Partition #%u number ('%s') verification failed on '%s': Unable to unlock the partition (errno=%d)"),
4312	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, errno);
4313	if (RT_SUCCESS(rc))
4314	rc = rc2;
4315	}
4316	}
4317	}
4318
4319	#else
4320	RT_NOREF(hVol); /* PORTME */
4321	rc = VERR_NOT_SUPPORTED;
4322	#endif
4323	if (RT_SUCCESS(rc))
4324	{
4325	/*
4326	* Compare the first 32 sectors of the partition.
4327	*
4328	* This might not be conclusive, but for partitions formatted with the more
4329	* common file systems it should be as they have a superblock copy at or near
4330	* the start of the partition (fat, fat32, ntfs, and ext4 does at least).
4331	*/
4332	size_t const cbToCompare = (size_t)RT_MIN(pPartDesc->cbData / cbSector, 32) * cbSector;
4333	uint8_t pbSector1 = (uint8_t )RTMemTmpAlloc(cbToCompare * 2);
4334	if (pbSector1 != NULL)
4335	{
4336	uint8_t *pbSector2 = pbSector1 + cbToCompare;
4337
4338	/* Do the comparing, we repeat if it fails and the data might be volatile. */
4339	uint64_t uPrevCrc1 = 0;
4340	uint64_t uPrevCrc2 = 0;
4341	uint32_t cStable = 0;
4342	for (unsigned iTry = 0; iTry < 256; iTry++)
4343	{
4344	rc = RTFileReadAt(hRawDrive, pPartDesc->offStartInVDisk, pbSector1, cbToCompare, NULL);
4345	if (RT_SUCCESS(rc))
4346	{
4347	rc = RTFileReadAt(hRawPart, pPartDesc->offStartInDevice, pbSector2, cbToCompare, NULL);
4348	if (RT_SUCCESS(rc))
4349	{
4350	if (memcmp(pbSector1, pbSector2, cbToCompare) != 0)
4351	{
4352	rc = VERR_MISMATCH;
4353
4354	/* Do data stability checks before repeating: */
4355	uint64_t const uCrc1 = RTCrc64(pbSector1, cbToCompare);
4356	uint64_t const uCrc2 = RTCrc64(pbSector2, cbToCompare);
4357	if ( uPrevCrc1 != uCrc1
4358	\|\| uPrevCrc2 != uCrc2)
4359	cStable = 0;
4360	else if (++cStable > 4)
4361	break;
4362	uPrevCrc1 = uCrc1;
4363	uPrevCrc2 = uCrc2;
4364	continue;
4365	}
4366	rc = VINF_SUCCESS;
4367	}
4368	else
4369	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4370	N_("VMDK: Image path: '%s'. Error reading %zu bytes from '%s' at offset %RU64 (%Rrc)"),
4371	pImage->pszFilename, cbToCompare, pPartDesc->pszRawDevice, pPartDesc->offStartInDevice, rc);
4372	}
4373	else
4374	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4375	N_("VMDK: Image path: '%s'. Error reading %zu bytes from '%s' at offset %RU64 (%Rrc)"),
4376	pImage->pszFilename, cbToCompare, pszRawDrive, pPartDesc->offStartInVDisk, rc);
4377	break;
4378	}
4379	if (rc == VERR_MISMATCH)
4380	{
4381	/* Find the first mismatching bytes: */
4382	size_t offMissmatch = 0;
4383	while (offMissmatch < cbToCompare && pbSector1[offMissmatch] == pbSector2[offMissmatch])
4384	offMissmatch++;
4385	int cbSample = (int)RT_MIN(cbToCompare - offMissmatch, 16);
4386
4387	if (cStable > 0)
4388	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4389	N_("VMDK: Image path: '%s'. Partition #%u path ('%s') verification failed on '%s' (cStable=%d @%#zx: %.Rhxs vs %.Rhxs)"),
4390	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive, cStable,
4391	offMissmatch, cbSample, &pbSector1[offMissmatch], cbSample, &pbSector2[offMissmatch]);
4392	else
4393	{
4394	LogRel(("VMDK: Image path: '%s'. Partition #%u path ('%s') verification undecided on '%s' because of unstable data! (@%#zx: %.Rhxs vs %.Rhxs)\n",
4395	pImage->pszFilename, idxPartition, pPartDesc->pszRawDevice, pszRawDrive,
4396	offMissmatch, cbSample, &pbSector1[offMissmatch], cbSample, &pbSector2[offMissmatch]));
4397	rc = -rc;
4398	}
4399	}
4400
4401	RTMemTmpFree(pbSector1);
4402	}
4403	else
4404	rc = vdIfError(pImage->pIfError, VERR_NO_TMP_MEMORY, RT_SRC_POS,
4405	N_("VMDK: Image path: '%s'. Failed to allocate %zu bytes for a temporary read buffer\n"),
4406	pImage->pszFilename, cbToCompare * 2);
4407	}
4408	RTFileClose(hRawPart);
4409	return rc;
4410	}
4411
4412	#ifdef RT_OS_WINDOWS
4413	/**
4414	* Construct the device name for the given partition number.
4415	*/
4416	static int vmdkRawDescWinMakePartitionName(PVMDKIMAGE pImage, const char *pszRawDrive, RTFILE hRawDrive, uint32_t idxPartition,
4417	char **ppszRawPartition)
4418	{
4419	int rc = VINF_SUCCESS;
4420	DWORD cbReturned = 0;
4421	STORAGE_DEVICE_NUMBER DevNum;
4422	RT_ZERO(DevNum);
4423	if (DeviceIoControl((HANDLE)RTFileToNative(hRawDrive), IOCTL_STORAGE_GET_DEVICE_NUMBER,
4424	NULL /pvInBuffer/, 0 /cbInBuffer/, &DevNum, sizeof(DevNum), &cbReturned, NULL /pOverlapped/))
4425	RTStrAPrintf(ppszRawPartition, "\\\\.\\Harddisk%uPartition%u", DevNum.DeviceNumber, idxPartition);
4426	else
4427	rc = vdIfError(pImage->pIfError, RTErrConvertFromWin32(GetLastError()), RT_SRC_POS,
4428	N_("VMDK: Image path: '%s'. IOCTL_STORAGE_GET_DEVICE_NUMBER failed on '%s': %u"),
4429	pImage->pszFilename, pszRawDrive, GetLastError());
4430	return rc;
4431	}
4432	#endif /* RT_OS_WINDOWS */
4433
4434	/**
4435	* Worker for vmdkMakeRawDescriptor that adds partition descriptors when the
4436	* 'Partitions' configuration value is present.
4437	*
4438	* @returns VBox status code, error message has been set on failure.
4439	*
4440	* @note Caller is assumed to clean up @a pRawDesc and release
4441	* @a *phVolToRelease.
4442	* @internal
4443	*/
4444	static int vmdkRawDescDoPartitions(PVMDKIMAGE pImage, RTDVM hVolMgr, PVDISKRAW pRawDesc,
4445	RTFILE hRawDrive, const char *pszRawDrive, uint32_t cbSector,
4446	uint32_t fPartitions, uint32_t fPartitionsReadOnly, bool fRelative,
4447	PRTDVMVOLUME phVolToRelease)
4448	{
4449	*phVolToRelease = NIL_RTDVMVOLUME;
4450
4451	/* Check sanity/understanding. */
4452	Assert(fPartitions);
4453	Assert((fPartitions & fPartitionsReadOnly) == fPartitionsReadOnly); /* RO should be a sub-set */
4454
4455	/*
4456	* Allocate on descriptor for each volume up front.
4457	*/
4458	uint32_t const cVolumes = RTDvmMapGetValidVolumes(hVolMgr);
4459
4460	PVDISKRAWPARTDESC paPartDescs = NULL;
4461	int rc = vmdkRawDescAppendPartDesc(pImage, pRawDesc, cVolumes, &paPartDescs);
4462	AssertRCReturn(rc, rc);
4463
4464	/*
4465	* Enumerate the partitions (volumes) on the disk and create descriptors for each of them.
4466	*/
4467	uint32_t fPartitionsLeft = fPartitions;
4468	RTDVMVOLUME hVol = NIL_RTDVMVOLUME; /* the current volume, needed for getting the next. */
4469	for (uint32_t i = 0; i < cVolumes; i++)
4470	{
4471	/*
4472	* Get the next/first volume and release the current.
4473	*/
4474	RTDVMVOLUME hVolNext = NIL_RTDVMVOLUME;
4475	if (i == 0)
4476	rc = RTDvmMapQueryFirstVolume(hVolMgr, &hVolNext);
4477	else
4478	rc = RTDvmMapQueryNextVolume(hVolMgr, hVol, &hVolNext);
4479	if (RT_FAILURE(rc))
4480	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4481	N_("VMDK: Image path: '%s'. Volume enumeration failed at volume #%u on '%s' (%Rrc)"),
4482	pImage->pszFilename, i, pszRawDrive, rc);
4483	uint32_t cRefs = RTDvmVolumeRelease(hVol);
4484	Assert(cRefs != UINT32_MAX); RT_NOREF(cRefs);
4485	*phVolToRelease = hVol = hVolNext;
4486
4487	/*
4488	* Depending on the fPartitions selector and associated read-only mask,
4489	* the guest either gets read-write or read-only access (bits set)
4490	* or no access (selector bit clear, access directed to the VMDK).
4491	*/
4492	paPartDescs[i].cbData = RTDvmVolumeGetSize(hVol);
4493
4494	uint64_t offVolumeEndIgnored = 0;
4495	rc = RTDvmVolumeQueryRange(hVol, &paPartDescs[i].offStartInVDisk, &offVolumeEndIgnored);
4496	if (RT_FAILURE(rc))
4497	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4498	N_("VMDK: Image path: '%s'. Failed to get location of volume #%u on '%s' (%Rrc)"),
4499	pImage->pszFilename, i, pszRawDrive, rc);
4500	Assert(paPartDescs[i].cbData == offVolumeEndIgnored + 1 - paPartDescs[i].offStartInVDisk);
4501
4502	/* Note! The index must match IHostDrivePartition::number. */
4503	uint32_t idxPartition = RTDvmVolumeGetIndex(hVol, RTDVMVOLIDX_HOST);
4504	if ( idxPartition < 32
4505	&& (fPartitions & RT_BIT_32(idxPartition)))
4506	{
4507	fPartitionsLeft &= ~RT_BIT_32(idxPartition);
4508	if (fPartitionsReadOnly & RT_BIT_32(idxPartition))
4509	paPartDescs[i].uFlags \|= VDISKRAW_READONLY;
4510
4511	if (!fRelative)
4512	{
4513	/*
4514	* Accessing the drive thru the main device node (pRawDesc->pszRawDisk).
4515	*/
4516	paPartDescs[i].offStartInDevice = paPartDescs[i].offStartInVDisk;
4517	paPartDescs[i].pszRawDevice = RTStrDup(pszRawDrive);
4518	AssertPtrReturn(paPartDescs[i].pszRawDevice, VERR_NO_STR_MEMORY);
4519	}
4520	else
4521	{
4522	/*
4523	* Relative means access the partition data via the device node for that
4524	* partition, allowing the sysadmin/OS to allow a user access to individual
4525	* partitions without necessarily being able to compromise the host OS.
4526	* Obviously, the creation of the VMDK requires read access to the main
4527	* device node for the drive, but that's a one-time thing and can be done
4528	* by the sysadmin. Here data starts at offset zero in the device node.
4529	*/
4530	paPartDescs[i].offStartInDevice = 0;
4531
4532	#if defined(RT_OS_DARWIN) \|\| defined(RT_OS_FREEBSD)
4533	/* /dev/rdisk1 -> /dev/rdisk1s2 (s=slice) */
4534	RTStrAPrintf(&paPartDescs[i].pszRawDevice, "%ss%u", pszRawDrive, idxPartition);
4535	#elif defined(RT_OS_LINUX)
4536	/* Two naming schemes here: /dev/nvme0n1 -> /dev/nvme0n1p1; /dev/sda -> /dev/sda1 */
4537	RTStrAPrintf(&paPartDescs[i].pszRawDevice,
4538	RT_C_IS_DIGIT(pszRawDrive[strlen(pszRawDrive) - 1]) ? "%sp%u" : "%s%u", pszRawDrive, idxPartition);
4539	#elif defined(RT_OS_WINDOWS)
4540	rc = vmdkRawDescWinMakePartitionName(pImage, pszRawDrive, hRawDrive, idxPartition, &paPartDescs[i].pszRawDevice);
4541	AssertRCReturn(rc, rc);
4542	#elif defined(RT_OS_SOLARIS)
4543	if (pRawDesc->enmPartitioningType == VDISKPARTTYPE_MBR)
4544	{
4545	/*
4546	* MBR partitions have device nodes in form /dev/(r)dsk/cXtYdZpK
4547	* where X is the controller,
4548	* Y is target (SCSI device number),
4549	* Z is disk number,
4550	* K is partition number,
4551	* where p0 is the whole disk
4552	* p1-pN are the partitions of the disk
4553	*/
4554	const char *pszRawDrivePath = pszRawDrive;
4555	char szDrivePath[RTPATH_MAX];
4556	size_t cbRawDrive = strlen(pszRawDrive);
4557	if ( cbRawDrive > 1 && strcmp(&pszRawDrive[cbRawDrive - 2], "p0") == 0)
4558	{
4559	memcpy(szDrivePath, pszRawDrive, cbRawDrive - 2);
4560	szDrivePath[cbRawDrive - 2] = '\0';
4561	pszRawDrivePath = szDrivePath;
4562	}
4563	RTStrAPrintf(&paPartDescs[i].pszRawDevice, "%sp%u", pszRawDrivePath, idxPartition);
4564	}
4565	else /* GPT */
4566	{
4567	/*
4568	* GPT partitions have device nodes in form /dev/(r)dsk/cXtYdZsK
4569	* where X is the controller,
4570	* Y is target (SCSI device number),
4571	* Z is disk number,
4572	* K is partition number, zero based. Can be only from 0 to 6.
4573	* Thus, only partitions numbered 0 through 6 have device nodes.
4574	*/
4575	if (idxPartition > 7)
4576	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4577	N_("VMDK: Image path: '%s'. the partition #%u on '%s' has no device node and can not be specified with 'Relative' property"),
4578	pImage->pszFilename, idxPartition, pszRawDrive);
4579	RTStrAPrintf(&paPartDescs[i].pszRawDevice, "%ss%u", pszRawDrive, idxPartition - 1);
4580	}
4581	#else
4582	AssertFailedReturn(VERR_INTERNAL_ERROR_4); /* The option parsing code should have prevented this - PORTME */
4583	#endif
4584	AssertPtrReturn(paPartDescs[i].pszRawDevice, VERR_NO_STR_MEMORY);
4585
4586	rc = vmdkRawDescVerifyPartitionPath(pImage, &paPartDescs[i], idxPartition, pszRawDrive, hRawDrive, cbSector, hVol);
4587	AssertRCReturn(rc, rc);
4588	}
4589	}
4590	else
4591	{
4592	/* Not accessible to the guest. */
4593	paPartDescs[i].offStartInDevice = 0;
4594	paPartDescs[i].pszRawDevice = NULL;
4595	}
4596	} /* for each volume */
4597
4598	RTDvmVolumeRelease(hVol);
4599	*phVolToRelease = NIL_RTDVMVOLUME;
4600
4601	/*
4602	* Check that we found all the partitions the user selected.
4603	*/
4604	if (fPartitionsLeft)
4605	{
4606	char szLeft[3 * sizeof(fPartitions) * 8];
4607	size_t cchLeft = 0;
4608	for (unsigned i = 0; i < sizeof(fPartitions) * 8; i++)
4609	if (fPartitionsLeft & RT_BIT_32(i))
4610	cchLeft += RTStrPrintf(&szLeft[cchLeft], sizeof(szLeft) - cchLeft, cchLeft ? "%u" : ",%u", i);
4611	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4612	N_("VMDK: Image path: '%s'. Not all the specified partitions for drive '%s' was found: %s"),
4613	pImage->pszFilename, pszRawDrive, szLeft);
4614	}
4615
4616	return VINF_SUCCESS;
4617	}
4618
4619	/**
4620	* Worker for vmdkMakeRawDescriptor that adds partition descriptors with copies
4621	* of the partition tables and associated padding areas when the 'Partitions'
4622	* configuration value is present.
4623	*
4624	* The guest is not allowed access to the partition tables, however it needs
4625	* them to be able to access the drive. So, create descriptors for each of the
4626	* tables and attach the current disk content. vmdkCreateRawImage() will later
4627	* write the content to the VMDK. Any changes the guest later makes to the
4628	* partition tables will then go to the VMDK copy, rather than the host drive.
4629	*
4630	* @returns VBox status code, error message has been set on failure.
4631	*
4632	* @note Caller is assumed to clean up @a pRawDesc
4633	* @internal
4634	*/
4635	static int vmdkRawDescDoCopyPartitionTables(PVMDKIMAGE pImage, RTDVM hVolMgr, PVDISKRAW pRawDesc,
4636	const char pszRawDrive, RTFILE hRawDrive, void pvBootSector, size_t cbBootSector)
4637	{
4638	/*
4639	* Query the locations.
4640	*/
4641	/* Determin how many locations there are: */
4642	size_t cLocations = 0;
4643	int rc = RTDvmMapQueryTableLocations(hVolMgr, RTDVMMAPQTABLOC_F_INCLUDE_LEGACY, NULL, 0, &cLocations);
4644	if (rc != VERR_BUFFER_OVERFLOW)
4645	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4646	N_("VMDK: Image path: '%s'. RTDvmMapQueryTableLocations failed on '%s' (%Rrc)"),
4647	pImage->pszFilename, pszRawDrive, rc);
4648	AssertReturn(cLocations > 0 && cLocations < _16M, VERR_INTERNAL_ERROR_5);
4649
4650	/* We can allocate the partition descriptors here to save an intentation level. */
4651	PVDISKRAWPARTDESC paPartDescs = NULL;
4652	rc = vmdkRawDescAppendPartDesc(pImage, pRawDesc, (uint32_t)cLocations, &paPartDescs);
4653	AssertRCReturn(rc, rc);
4654
4655	/* Allocate the result table and repeat the location table query: */
4656	PRTDVMTABLELOCATION paLocations = (PRTDVMTABLELOCATION)RTMemAllocZ(sizeof(paLocations[0]) * cLocations);
4657	if (!paLocations)
4658	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS, N_("VMDK: Image path: '%s'. Failed to allocate %zu bytes"),
4659	pImage->pszFilename, sizeof(paLocations[0]) * cLocations);
4660	rc = RTDvmMapQueryTableLocations(hVolMgr, RTDVMMAPQTABLOC_F_INCLUDE_LEGACY, paLocations, cLocations, NULL);
4661	if (RT_SUCCESS(rc))
4662	{
4663	/*
4664	* Translate them into descriptors.
4665	*
4666	* We restrict the amount of partition alignment padding to 4MiB as more
4667	* will just be a waste of space. The use case for including the padding
4668	* are older boot loaders and boot manager (including one by a team member)
4669	* that put data and code in the 62 sectors between the MBR and the first
4670	* partition (total of 63). Later CHS was abandond and partition started
4671	* being aligned on power of two sector boundraries (typically 64KiB or
4672	* 1MiB depending on the media size).
4673	*/
4674	for (size_t i = 0; i < cLocations && RT_SUCCESS(rc); i++)
4675	{
4676	Assert(paLocations[i].cb > 0);
4677	if (paLocations[i].cb <= _64M)
4678	{
4679	/* Create the partition descriptor entry: */
4680	//paPartDescs[i].pszRawDevice = NULL;
4681	//paPartDescs[i].offStartInDevice = 0;
4682	//paPartDescs[i].uFlags = 0;
4683	paPartDescs[i].offStartInVDisk = paLocations[i].off;
4684	paPartDescs[i].cbData = paLocations[i].cb;
4685	if (paPartDescs[i].cbData < _4M)
4686	paPartDescs[i].cbData = RT_MIN(paPartDescs[i].cbData + paLocations[i].cbPadding, _4M);
4687	paPartDescs[i].pvPartitionData = RTMemAllocZ((size_t)paPartDescs[i].cbData);
4688	if (paPartDescs[i].pvPartitionData)
4689	{
4690	/* Read the content from the drive: */
4691	rc = RTFileReadAt(hRawDrive, paPartDescs[i].offStartInVDisk, paPartDescs[i].pvPartitionData,
4692	(size_t)paPartDescs[i].cbData, NULL);
4693	if (RT_SUCCESS(rc))
4694	{
4695	/* Do we have custom boot sector code? */
4696	if (pvBootSector && cbBootSector && paPartDescs[i].offStartInVDisk == 0)
4697	{
4698	/* Note! Old code used to quietly drop the bootsector if it was considered too big.
4699	Instead we fail as we weren't able to do what the user requested us to do.
4700	Better if the user knows than starts questioning why the guest isn't
4701	booting as expected. */
4702	if (cbBootSector <= paPartDescs[i].cbData)
4703	memcpy(paPartDescs[i].pvPartitionData, pvBootSector, cbBootSector);
4704	else
4705	rc = vdIfError(pImage->pIfError, VERR_TOO_MUCH_DATA, RT_SRC_POS,
4706	N_("VMDK: Image path: '%s'. The custom boot sector is too big: %zu bytes, %RU64 bytes available"),
4707	pImage->pszFilename, cbBootSector, paPartDescs[i].cbData);
4708	}
4709	}
4710	else
4711	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4712	N_("VMDK: Image path: '%s'. Failed to read partition at off %RU64 length %zu from '%s' (%Rrc)"),
4713	pImage->pszFilename, paPartDescs[i].offStartInVDisk,
4714	(size_t)paPartDescs[i].cbData, pszRawDrive, rc);
4715	}
4716	else
4717	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4718	N_("VMDK: Image path: '%s'. Failed to allocate %zu bytes for copying the partition table at off %RU64"),
4719	pImage->pszFilename, (size_t)paPartDescs[i].cbData, paPartDescs[i].offStartInVDisk);
4720	}
4721	else
4722	rc = vdIfError(pImage->pIfError, VERR_TOO_MUCH_DATA, RT_SRC_POS,
4723	N_("VMDK: Image path: '%s'. Partition table #%u at offset %RU64 in '%s' is to big: %RU64 bytes"),
4724	pImage->pszFilename, i, paLocations[i].off, pszRawDrive, paLocations[i].cb);
4725	}
4726	}
4727	else
4728	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4729	N_("VMDK: Image path: '%s'. RTDvmMapQueryTableLocations failed on '%s' (%Rrc)"),
4730	pImage->pszFilename, pszRawDrive, rc);
4731	RTMemFree(paLocations);
4732	return rc;
4733	}
4734
4735	/**
4736	* Opens the volume manager for the raw drive when in selected-partition mode.
4737	*
4738	* @param pImage The VMDK image (for errors).
4739	* @param hRawDrive The raw drive handle.
4740	* @param pszRawDrive The raw drive device path (for errors).
4741	* @param cbSector The sector size.
4742	* @param phVolMgr Where to return the handle to the volume manager on
4743	* success.
4744	* @returns VBox status code, errors have been reported.
4745	* @internal
4746	*/
4747	static int vmdkRawDescOpenVolMgr(PVMDKIMAGE pImage, RTFILE hRawDrive, const char *pszRawDrive, uint32_t cbSector, PRTDVM phVolMgr)
4748	{
4749	*phVolMgr = NIL_RTDVM;
4750
4751	RTVFSFILE hVfsFile = NIL_RTVFSFILE;
4752	int rc = RTVfsFileFromRTFile(hRawDrive, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE, true /fLeaveOpen/, &hVfsFile);
4753	if (RT_FAILURE(rc))
4754	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4755	N_("VMDK: Image path: '%s'. RTVfsFileFromRTFile failed for '%s' handle (%Rrc)"),
4756	pImage->pszFilename, pszRawDrive, rc);
4757
4758	RTDVM hVolMgr = NIL_RTDVM;
4759	rc = RTDvmCreate(&hVolMgr, hVfsFile, cbSector, 0 /fFlags/);
4760
4761	RTVfsFileRelease(hVfsFile);
4762
4763	if (RT_FAILURE(rc))
4764	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4765	N_("VMDK: Image path: '%s'. Failed to create volume manager instance for '%s' (%Rrc)"),
4766	pImage->pszFilename, pszRawDrive, rc);
4767
4768	rc = RTDvmMapOpen(hVolMgr);
4769	if (RT_SUCCESS(rc))
4770	{
4771	*phVolMgr = hVolMgr;
4772	return VINF_SUCCESS;
4773	}
4774	RTDvmRelease(hVolMgr);
4775	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: Image path: '%s'. RTDvmMapOpen failed for '%s' (%Rrc)"),
4776	pImage->pszFilename, pszRawDrive, rc);
4777	}
4778
4779	/**
4780	* Opens the raw drive device and get the sizes for it.
4781	*
4782	* @param pImage The image (for error reporting).
4783	* @param pszRawDrive The device/whatever to open.
4784	* @param phRawDrive Where to return the file handle.
4785	* @param pcbRawDrive Where to return the size.
4786	* @param pcbSector Where to return the sector size.
4787	* @returns IPRT status code, errors have been reported.
4788	* @internal
4789	*/
4790	static int vmkdRawDescOpenDevice(PVMDKIMAGE pImage, const char *pszRawDrive,
4791	PRTFILE phRawDrive, uint64_t pcbRawDrive, uint32_t pcbSector)
4792	{
4793	/*
4794	* Open the device for the raw drive.
4795	*/
4796	RTFILE hRawDrive = NIL_RTFILE;
4797	int rc = RTFileOpen(&hRawDrive, pszRawDrive, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE);
4798	if (RT_FAILURE(rc))
4799	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4800	N_("VMDK: Image path: '%s'. Failed to open the raw drive '%s' for reading (%Rrc)"),
4801	pImage->pszFilename, pszRawDrive, rc);
4802
4803	/*
4804	* Get the sector size.
4805	*/
4806	uint32_t cbSector = 0;
4807	rc = RTFileQuerySectorSize(hRawDrive, &cbSector);
4808	if (RT_SUCCESS(rc))
4809	{
4810	/* sanity checks */
4811	if ( cbSector >= 512
4812	&& cbSector <= _64K
4813	&& RT_IS_POWER_OF_TWO(cbSector))
4814	{
4815	/*
4816	* Get the size.
4817	*/
4818	uint64_t cbRawDrive = 0;
4819	rc = RTFileQuerySize(hRawDrive, &cbRawDrive);
4820	if (RT_SUCCESS(rc))
4821	{
4822	/* Check whether cbSize is actually sensible. */
4823	if (cbRawDrive > cbSector && (cbRawDrive % cbSector) == 0)
4824	{
4825	*phRawDrive = hRawDrive;
4826	*pcbRawDrive = cbRawDrive;
4827	*pcbSector = cbSector;
4828	return VINF_SUCCESS;
4829	}
4830	rc = vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4831	N_("VMDK: Image path: '%s'. Got a bogus size for the raw drive '%s': %RU64 (sector size %u)"),
4832	pImage->pszFilename, pszRawDrive, cbRawDrive, cbSector);
4833	}
4834	else
4835	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4836	N_("VMDK: Image path: '%s'. Failed to query size of the drive '%s' (%Rrc)"),
4837	pImage->pszFilename, pszRawDrive, rc);
4838	}
4839	else
4840	rc = vdIfError(pImage->pIfError, VERR_OUT_OF_RANGE, RT_SRC_POS,
4841	N_("VMDK: Image path: '%s'. Unsupported sector size for '%s': %u (%#x)"),
4842	pImage->pszFilename, pszRawDrive, cbSector, cbSector);
4843	}
4844	else
4845	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4846	N_("VMDK: Image path: '%s'. Failed to get the sector size for '%s' (%Rrc)"),
4847	pImage->pszFilename, pszRawDrive, rc);
4848	RTFileClose(hRawDrive);
4849	return rc;
4850	}
4851
4852	/**
4853	* Reads the raw disk configuration, leaving initalization and cleanup to the
4854	* caller (regardless of return status).
4855	*
4856	* @returns VBox status code, errors properly reported.
4857	* @internal
4858	*/
4859	static int vmdkRawDescParseConfig(PVMDKIMAGE pImage, char **ppszRawDrive,
4860	uint32_t pfPartitions, uint32_t pfPartitionsReadOnly,
4861	void *ppvBootSector, size_t pcbBootSector, bool *pfRelative,
4862	char **ppszFreeMe)
4863	{
4864	PVDINTERFACECONFIG pImgCfg = VDIfConfigGet(pImage->pVDIfsImage);
4865	if (!pImgCfg)
4866	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4867	N_("VMDK: Image path: '%s'. Getting config interface failed"), pImage->pszFilename);
4868
4869	/*
4870	* RawDrive = path
4871	*/
4872	int rc = VDCFGQueryStringAlloc(pImgCfg, "RawDrive", ppszRawDrive);
4873	if (RT_FAILURE(rc))
4874	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4875	N_("VMDK: Image path: '%s'. Getting 'RawDrive' configuration failed (%Rrc)"), pImage->pszFilename, rc);
4876	AssertPtrReturn(*ppszRawDrive, VERR_INTERNAL_ERROR_3);
4877
4878	/*
4879	* Partitions=n[r][,...]
4880	*/
4881	uint32_t const cMaxPartitionBits = sizeof(pfPartitions) 8 /* ASSUMES 8 bits per char */;
4882	pfPartitions = pfPartitionsReadOnly = 0;
4883
4884	rc = VDCFGQueryStringAlloc(pImgCfg, "Partitions", ppszFreeMe);
4885	if (RT_SUCCESS(rc))
4886	{
4887	char psz = ppszFreeMe;
4888	while (*psz != '\0')
4889	{
4890	char *pszNext;
4891	uint32_t u32;
4892	rc = RTStrToUInt32Ex(psz, &pszNext, 0, &u32);
4893	if (rc == VWRN_NUMBER_TOO_BIG \|\| rc == VWRN_NEGATIVE_UNSIGNED)
4894	rc = -rc;
4895	if (RT_FAILURE(rc))
4896	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4897	N_("VMDK: Image path: '%s'. Parsing 'Partitions' config value failed. Incorrect value (%Rrc): %s"),
4898	pImage->pszFilename, rc, psz);
4899	if (u32 >= cMaxPartitionBits)
4900	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4901	N_("VMDK: Image path: '%s'. 'Partitions' config sub-value out of range: %RU32, max %RU32"),
4902	pImage->pszFilename, u32, cMaxPartitionBits);
4903	*pfPartitions \|= RT_BIT_32(u32);
4904	psz = pszNext;
4905	if (*psz == 'r')
4906	{
4907	*pfPartitionsReadOnly \|= RT_BIT_32(u32);
4908	psz++;
4909	}
4910	if (*psz == ',')
4911	psz++;
4912	else if (*psz != '\0')
4913	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4914	N_("VMDK: Image path: '%s'. Malformed 'Partitions' config value, expected separator: %s"),
4915	pImage->pszFilename, psz);
4916	}
4917
4918	RTStrFree(*ppszFreeMe);
4919	*ppszFreeMe = NULL;
4920	}
4921	else if (rc != VERR_CFGM_VALUE_NOT_FOUND)
4922	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4923	N_("VMDK: Image path: '%s'. Getting 'Partitions' configuration failed (%Rrc)"), pImage->pszFilename, rc);
4924
4925	/*
4926	* BootSector=base64
4927	*/
4928	rc = VDCFGQueryStringAlloc(pImgCfg, "BootSector", ppszFreeMe);
4929	if (RT_SUCCESS(rc))
4930	{
4931	ssize_t cbBootSector = RTBase64DecodedSize(*ppszFreeMe, NULL);
4932	if (cbBootSector < 0)
4933	return vdIfError(pImage->pIfError, VERR_INVALID_BASE64_ENCODING, RT_SRC_POS,
4934	N_("VMDK: Image path: '%s'. BASE64 decoding failed on the custom bootsector for '%s'"),
4935	pImage->pszFilename, *ppszRawDrive);
4936	if (cbBootSector == 0)
4937	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4938	N_("VMDK: Image path: '%s'. Custom bootsector for '%s' is zero bytes big"),
4939	pImage->pszFilename, *ppszRawDrive);
4940	if (cbBootSector > _4M) /* this is just a preliminary max */
4941	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4942	N_("VMDK: Image path: '%s'. Custom bootsector for '%s' is way too big: %zu bytes, max 4MB"),
4943	pImage->pszFilename, *ppszRawDrive, cbBootSector);
4944
4945	/* Refuse the boot sector if whole-drive. This used to be done quietly,
4946	however, bird disagrees and thinks the user should be told that what
4947	he/she/it tries to do isn't possible. There should be less head
4948	scratching this way when the guest doesn't do the expected thing. */
4949	if (!*pfPartitions)
4950	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4951	N_("VMDK: Image path: '%s'. Custom bootsector for '%s' is not supported for whole-drive configurations, only when selecting partitions"),
4952	pImage->pszFilename, *ppszRawDrive);
4953
4954	*pcbBootSector = (size_t)cbBootSector;
4955	*ppvBootSector = RTMemAlloc((size_t)cbBootSector);
4956	if (!*ppvBootSector)
4957	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS,
4958	N_("VMDK: Image path: '%s'. Failed to allocate %zd bytes for the custom bootsector for '%s'"),
4959	pImage->pszFilename, cbBootSector, *ppszRawDrive);
4960
4961	rc = RTBase64Decode(ppszFreeMe, ppvBootSector, cbBootSector, NULL /pcbActual/, NULL /ppszEnd/);
4962	if (RT_FAILURE(rc))
4963	return vdIfError(pImage->pIfError, VERR_NO_MEMORY, RT_SRC_POS,
4964	N_("VMDK: Image path: '%s'. Base64 decoding of the custom boot sector for '%s' failed (%Rrc)"),
4965	pImage->pszFilename, *ppszRawDrive, rc);
4966
4967	RTStrFree(*ppszFreeMe);
4968	*ppszFreeMe = NULL;
4969	}
4970	else if (rc != VERR_CFGM_VALUE_NOT_FOUND)
4971	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4972	N_("VMDK: Image path: '%s'. Getting 'BootSector' configuration failed (%Rrc)"), pImage->pszFilename, rc);
4973
4974	/*
4975	* Relative=0/1
4976	*/
4977	*pfRelative = false;
4978	rc = VDCFGQueryBool(pImgCfg, "Relative", pfRelative);
4979	if (RT_SUCCESS(rc))
4980	{
4981	if (!pfPartitions && pfRelative != false)
4982	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4983	N_("VMDK: Image path: '%s'. The 'Relative' option is not supported for whole-drive configurations, only when selecting partitions"),
4984	pImage->pszFilename);
4985	#if !defined(RT_OS_DARWIN) && !defined(RT_OS_LINUX) && !defined(RT_OS_FREEBSD) && !defined(RT_OS_WINDOWS) && !defined(RT_OS_SOLARIS) /* PORTME */
4986	if (*pfRelative == true)
4987	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
4988	N_("VMDK: Image path: '%s'. The 'Relative' option is not supported on this host OS"),
4989	pImage->pszFilename);
4990	#endif
4991	}
4992	else if (rc != VERR_CFGM_VALUE_NOT_FOUND)
4993	return vdIfError(pImage->pIfError, rc, RT_SRC_POS,
4994	N_("VMDK: Image path: '%s'. Getting 'Relative' configuration failed (%Rrc)"), pImage->pszFilename, rc);
4995	else
4996	#ifdef RT_OS_DARWIN /* different default on macOS, see ticketref:1461 (comment 20). */
4997	*pfRelative = true;
4998	#else
4999	*pfRelative = false;
5000	#endif
5001
5002	return VINF_SUCCESS;
5003	}
5004
5005	/**
5006	* Creates a raw drive (nee disk) descriptor.
5007	*
5008	* This was originally done in VBoxInternalManage.cpp, but was copied (not move)
5009	* here much later. That's one of the reasons why we produce a descriptor just
5010	* like it does, rather than mixing directly into the vmdkCreateRawImage code.
5011	*
5012	* @returns VBox status code.
5013	* @param pImage The image.
5014	* @param ppRaw Where to return the raw drive descriptor. Caller must
5015	* free it using vmdkRawDescFree regardless of the status
5016	* code.
5017	* @internal
5018	*/
5019	static int vmdkMakeRawDescriptor(PVMDKIMAGE pImage, PVDISKRAW *ppRaw)
5020	{
5021	/* Make sure it's NULL. */
5022	*ppRaw = NULL;
5023
5024	/*
5025	* Read the configuration.
5026	*/
5027	char *pszRawDrive = NULL;
5028	uint32_t fPartitions = 0; /* zero if whole-drive */
5029	uint32_t fPartitionsReadOnly = 0; /* (subset of fPartitions) */
5030	void *pvBootSector = NULL;
5031	size_t cbBootSector = 0;
5032	bool fRelative = false;
5033	char pszFreeMe = NULL; / lazy bird cleanup. */
5034	int rc = vmdkRawDescParseConfig(pImage, &pszRawDrive, &fPartitions, &fPartitionsReadOnly,
5035	&pvBootSector, &cbBootSector, &fRelative, &pszFreeMe);
5036	RTStrFree(pszFreeMe);
5037	if (RT_SUCCESS(rc))
5038	{
5039	/*
5040	* Open the device, getting the sector size and drive size.
5041	*/
5042	uint64_t cbSize = 0;
5043	uint32_t cbSector = 0;
5044	RTFILE hRawDrive = NIL_RTFILE;
5045	rc = vmkdRawDescOpenDevice(pImage, pszRawDrive, &hRawDrive, &cbSize, &cbSector);
5046	if (RT_SUCCESS(rc))
5047	{
5048	pImage->cbSize = cbSize;
5049	/*
5050	* Create the raw-drive descriptor
5051	*/
5052	PVDISKRAW pRawDesc = (PVDISKRAW)RTMemAllocZ(sizeof(*pRawDesc));
5053	if (pRawDesc)
5054	{
5055	pRawDesc->szSignature[0] = 'R';
5056	pRawDesc->szSignature[1] = 'A';
5057	pRawDesc->szSignature[2] = 'W';
5058	//pRawDesc->szSignature[3] = '\0';
5059	if (!fPartitions)
5060	{
5061	/*
5062	* It's simple for when doing the whole drive.
5063	*/
5064	pRawDesc->uFlags = VDISKRAW_DISK;
5065	rc = RTStrDupEx(&pRawDesc->pszRawDisk, pszRawDrive);
5066	}
5067	else
5068	{
5069	/*
5070	* In selected partitions mode we've got a lot more work ahead of us.
5071	*/
5072	pRawDesc->uFlags = VDISKRAW_NORMAL;
5073	//pRawDesc->pszRawDisk = NULL;
5074	//pRawDesc->cPartDescs = 0;
5075	//pRawDesc->pPartDescs = NULL;
5076
5077	/* We need to parse the partition map to complete the descriptor: */
5078	RTDVM hVolMgr = NIL_RTDVM;
5079	rc = vmdkRawDescOpenVolMgr(pImage, hRawDrive, pszRawDrive, cbSector, &hVolMgr);
5080	if (RT_SUCCESS(rc))
5081	{
5082	RTDVMFORMATTYPE enmFormatType = RTDvmMapGetFormatType(hVolMgr);
5083	if ( enmFormatType == RTDVMFORMATTYPE_MBR
5084	\|\| enmFormatType == RTDVMFORMATTYPE_GPT)
5085	{
5086	pRawDesc->enmPartitioningType = enmFormatType == RTDVMFORMATTYPE_MBR
5087	? VDISKPARTTYPE_MBR : VDISKPARTTYPE_GPT;
5088
5089	/* Add copies of the partition tables: */
5090	rc = vmdkRawDescDoCopyPartitionTables(pImage, hVolMgr, pRawDesc, pszRawDrive, hRawDrive,
5091	pvBootSector, cbBootSector);
5092	if (RT_SUCCESS(rc))
5093	{
5094	/* Add descriptors for the partitions/volumes, indicating which
5095	should be accessible and how to access them: */
5096	RTDVMVOLUME hVolRelease = NIL_RTDVMVOLUME;
5097	rc = vmdkRawDescDoPartitions(pImage, hVolMgr, pRawDesc, hRawDrive, pszRawDrive, cbSector,
5098	fPartitions, fPartitionsReadOnly, fRelative, &hVolRelease);
5099	RTDvmVolumeRelease(hVolRelease);
5100
5101	/* Finally, sort the partition and check consistency (overlaps, etc): */
5102	if (RT_SUCCESS(rc))
5103	rc = vmdkRawDescPostProcessPartitions(pImage, pRawDesc, cbSize);
5104	}
5105	}
5106	else
5107	rc = vdIfError(pImage->pIfError, VERR_NOT_SUPPORTED, RT_SRC_POS,
5108	N_("VMDK: Image path: '%s'. Unsupported partitioning for the disk '%s': %s"),
5109	pImage->pszFilename, pszRawDrive, RTDvmMapGetFormatType(hVolMgr));
5110	RTDvmRelease(hVolMgr);
5111	}
5112	}
5113	if (RT_SUCCESS(rc))
5114	{
5115	/*
5116	* We succeeded.
5117	*/
5118	*ppRaw = pRawDesc;
5119	Log(("vmdkMakeRawDescriptor: fFlags=%#x enmPartitioningType=%d cPartDescs=%u pszRawDisk=%s\n",
5120	pRawDesc->uFlags, pRawDesc->enmPartitioningType, pRawDesc->cPartDescs, pRawDesc->pszRawDisk));
5121	if (pRawDesc->cPartDescs)
5122	{
5123	Log(("# VMDK offset Length Device offset PartDataPtr Device\n"));
5124	for (uint32_t i = 0; i < pRawDesc->cPartDescs; i++)
5125	Log(("%2u %14RU64 %14RU64 %14RU64 %#18p %s\n", i, pRawDesc->pPartDescs[i].offStartInVDisk,
5126	pRawDesc->pPartDescs[i].cbData, pRawDesc->pPartDescs[i].offStartInDevice,
5127	pRawDesc->pPartDescs[i].pvPartitionData, pRawDesc->pPartDescs[i].pszRawDevice));
5128	}
5129	}
5130	else
5131	vmdkRawDescFree(pRawDesc);
5132	}
5133	else
5134	rc = vdIfError(pImage->pIfError, VERR_NOT_SUPPORTED, RT_SRC_POS,
5135	N_("VMDK: Image path: '%s'. Failed to allocate %u bytes for the raw drive descriptor"),
5136	pImage->pszFilename, sizeof(*pRawDesc));
5137	RTFileClose(hRawDrive);
5138	}
5139	}
5140	RTStrFree(pszRawDrive);
5141	RTMemFree(pvBootSector);
5142	return rc;
5143	}
5144
5145	/**
5146	* Internal: create VMDK images for raw disk/partition access.
5147	*/
5148	static int vmdkCreateRawImage(PVMDKIMAGE pImage, const PVDISKRAW pRaw,
5149	uint64_t cbSize)
5150	{
5151	int rc = VINF_SUCCESS;
5152	PVMDKEXTENT pExtent;
5153
5154	if (pRaw->uFlags & VDISKRAW_DISK)
5155	{
5156	/* Full raw disk access. This requires setting up a descriptor
5157	* file and open the (flat) raw disk. */
5158	rc = vmdkCreateExtents(pImage, 1);
5159	if (RT_FAILURE(rc))
5160	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
5161	pExtent = &pImage->pExtents[0];
5162	/* Create raw disk descriptor file. */
5163	rc = vmdkFileOpen(pImage, &pImage->pFile, NULL, pImage->pszFilename,
5164	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5165	true /* fCreate */));
5166	if (RT_FAILURE(rc))
5167	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pImage->pszFilename);
5168
5169	/* Set up basename for extent description. Cannot use StrDup. */
5170	size_t cbBasename = strlen(pRaw->pszRawDisk) + 1;
5171	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
5172	if (!pszBasename)
5173	return VERR_NO_MEMORY;
5174	memcpy(pszBasename, pRaw->pszRawDisk, cbBasename);
5175	pExtent->pszBasename = pszBasename;
5176	/* For raw disks the full name is identical to the base name. */
5177	pExtent->pszFullname = RTStrDup(pszBasename);
5178	if (!pExtent->pszFullname)
5179	return VERR_NO_MEMORY;
5180	pExtent->enmType = VMDKETYPE_FLAT;
5181	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
5182	pExtent->uSectorOffset = 0;
5183	pExtent->enmAccess = (pRaw->uFlags & VDISKRAW_READONLY) ? VMDKACCESS_READONLY : VMDKACCESS_READWRITE;
5184	pExtent->fMetaDirty = false;
5185
5186	/* Open flat image, the raw disk. */
5187	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5188	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
5189	false /* fCreate */));
5190	if (RT_FAILURE(rc))
5191	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not open raw disk file '%s'"), pExtent->pszFullname);
5192	}
5193	else
5194	{
5195	/* Raw partition access. This requires setting up a descriptor
5196	* file, write the partition information to a flat extent and
5197	* open all the (flat) raw disk partitions. */
5198
5199	/* First pass over the partition data areas to determine how many
5200	* extents we need. One data area can require up to 2 extents, as
5201	* it might be necessary to skip over unpartitioned space. */
5202	unsigned cExtents = 0;
5203	uint64_t uStart = 0;
5204	for (unsigned i = 0; i < pRaw->cPartDescs; i++)
5205	{
5206	PVDISKRAWPARTDESC pPart = &pRaw->pPartDescs[i];
5207	if (uStart > pPart->offStartInVDisk)
5208	return vdIfError(pImage->pIfError, VERR_INVALID_PARAMETER, RT_SRC_POS,
5209	N_("VMDK: incorrect partition data area ordering set up by the caller in '%s'"), pImage->pszFilename);
5210
5211	if (uStart < pPart->offStartInVDisk)
5212	cExtents++;
5213	uStart = pPart->offStartInVDisk + pPart->cbData;
5214	cExtents++;
5215	}
5216	/* Another extent for filling up the rest of the image. */
5217	if (uStart != cbSize)
5218	cExtents++;
5219
5220	rc = vmdkCreateExtents(pImage, cExtents);
5221	if (RT_FAILURE(rc))
5222	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
5223
5224	/* Create raw partition descriptor file. */
5225	rc = vmdkFileOpen(pImage, &pImage->pFile, NULL, pImage->pszFilename,
5226	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5227	true /* fCreate */));
5228	if (RT_FAILURE(rc))
5229	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pImage->pszFilename);
5230
5231	/* Create base filename for the partition table extent. */
5232	/** @todo remove fixed buffer without creating memory leaks. */
5233	char pszPartition[1024];
5234	const char *pszBase = RTPathFilename(pImage->pszFilename);
5235	const char *pszSuff = RTPathSuffix(pszBase);
5236	if (pszSuff == NULL)
5237	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: invalid filename '%s'"), pImage->pszFilename);
5238	char *pszBaseBase = RTStrDup(pszBase);
5239	if (!pszBaseBase)
5240	return VERR_NO_MEMORY;
5241	RTPathStripSuffix(pszBaseBase);
5242	RTStrPrintf(pszPartition, sizeof(pszPartition), "%s-pt%s",
5243	pszBaseBase, pszSuff);
5244	RTStrFree(pszBaseBase);
5245
5246	/* Second pass over the partitions, now define all extents. */
5247	uint64_t uPartOffset = 0;
5248	cExtents = 0;
5249	uStart = 0;
5250	for (unsigned i = 0; i < pRaw->cPartDescs; i++)
5251	{
5252	PVDISKRAWPARTDESC pPart = &pRaw->pPartDescs[i];
5253	pExtent = &pImage->pExtents[cExtents++];
5254
5255	if (uStart < pPart->offStartInVDisk)
5256	{
5257	pExtent->pszBasename = NULL;
5258	pExtent->pszFullname = NULL;
5259	pExtent->enmType = VMDKETYPE_ZERO;
5260	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->offStartInVDisk - uStart);
5261	pExtent->uSectorOffset = 0;
5262	pExtent->enmAccess = VMDKACCESS_READWRITE;
5263	pExtent->fMetaDirty = false;
5264	/* go to next extent */
5265	pExtent = &pImage->pExtents[cExtents++];
5266	}
5267	uStart = pPart->offStartInVDisk + pPart->cbData;
5268
5269	if (pPart->pvPartitionData)
5270	{
5271	/* Set up basename for extent description. Can't use StrDup. */
5272	size_t cbBasename = strlen(pszPartition) + 1;
5273	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
5274	if (!pszBasename)
5275	return VERR_NO_MEMORY;
5276	memcpy(pszBasename, pszPartition, cbBasename);
5277	pExtent->pszBasename = pszBasename;
5278
5279	/* Set up full name for partition extent. */
5280	char *pszDirname = RTStrDup(pImage->pszFilename);
5281	if (!pszDirname)
5282	return VERR_NO_STR_MEMORY;
5283	RTPathStripFilename(pszDirname);
5284	char *pszFullname = RTPathJoinA(pszDirname, pExtent->pszBasename);
5285	RTStrFree(pszDirname);
5286	if (!pszFullname)
5287	return VERR_NO_STR_MEMORY;
5288	pExtent->pszFullname = pszFullname;
5289	pExtent->enmType = VMDKETYPE_FLAT;
5290	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
5291	pExtent->uSectorOffset = uPartOffset;
5292	pExtent->enmAccess = VMDKACCESS_READWRITE;
5293	pExtent->fMetaDirty = false;
5294
5295	/* Create partition table flat image. */
5296	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5297	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
5298	true /* fCreate */));
5299	if (RT_FAILURE(rc))
5300	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new partition data file '%s'"), pExtent->pszFullname);
5301	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
5302	VMDK_SECTOR2BYTE(uPartOffset),
5303	pPart->pvPartitionData,
5304	pPart->cbData);
5305	if (RT_FAILURE(rc))
5306	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not write partition data to '%s'"), pExtent->pszFullname);
5307	uPartOffset += VMDK_BYTE2SECTOR(pPart->cbData);
5308	}
5309	else
5310	{
5311	if (pPart->pszRawDevice)
5312	{
5313	/* Set up basename for extent descr. Can't use StrDup. */
5314	size_t cbBasename = strlen(pPart->pszRawDevice) + 1;
5315	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
5316	if (!pszBasename)
5317	return VERR_NO_MEMORY;
5318	memcpy(pszBasename, pPart->pszRawDevice, cbBasename);
5319	pExtent->pszBasename = pszBasename;
5320	/* For raw disks full name is identical to base name. */
5321	pExtent->pszFullname = RTStrDup(pszBasename);
5322	if (!pExtent->pszFullname)
5323	return VERR_NO_MEMORY;
5324	pExtent->enmType = VMDKETYPE_FLAT;
5325	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
5326	pExtent->uSectorOffset = VMDK_BYTE2SECTOR(pPart->offStartInDevice);
5327	pExtent->enmAccess = (pPart->uFlags & VDISKRAW_READONLY) ? VMDKACCESS_READONLY : VMDKACCESS_READWRITE;
5328	pExtent->fMetaDirty = false;
5329
5330	/* Open flat image, the raw partition. */
5331	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5332	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags \| ((pExtent->enmAccess == VMDKACCESS_READONLY) ? VD_OPEN_FLAGS_READONLY : 0),
5333	false /* fCreate */));
5334	if (RT_FAILURE(rc))
5335	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not open raw partition file '%s'"), pExtent->pszFullname);
5336	}
5337	else
5338	{
5339	pExtent->pszBasename = NULL;
5340	pExtent->pszFullname = NULL;
5341	pExtent->enmType = VMDKETYPE_ZERO;
5342	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbData);
5343	pExtent->uSectorOffset = 0;
5344	pExtent->enmAccess = VMDKACCESS_READWRITE;
5345	pExtent->fMetaDirty = false;
5346	}
5347	}
5348	}
5349	/* Another extent for filling up the rest of the image. */
5350	if (uStart != cbSize)
5351	{
5352	pExtent = &pImage->pExtents[cExtents++];
5353	pExtent->pszBasename = NULL;
5354	pExtent->pszFullname = NULL;
5355	pExtent->enmType = VMDKETYPE_ZERO;
5356	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize - uStart);
5357	pExtent->uSectorOffset = 0;
5358	pExtent->enmAccess = VMDKACCESS_READWRITE;
5359	pExtent->fMetaDirty = false;
5360	}
5361	}
5362
5363	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
5364	(pRaw->uFlags & VDISKRAW_DISK) ?
5365	"fullDevice" : "partitionedDevice");
5366	if (RT_FAILURE(rc))
5367	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
5368	return rc;
5369	}
5370
5371	/**
5372	* Internal: create a regular (i.e. file-backed) VMDK image.
5373	*/
5374	static int vmdkCreateRegularImage(PVMDKIMAGE pImage, uint64_t cbSize,
5375	unsigned uImageFlags, PVDINTERFACEPROGRESS pIfProgress,
5376	unsigned uPercentStart, unsigned uPercentSpan)
5377	{
5378	int rc = VINF_SUCCESS;
5379	unsigned cExtents = 1;
5380	uint64_t cbOffset = 0;
5381	uint64_t cbRemaining = cbSize;
5382
5383	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
5384	{
5385	cExtents = cbSize / VMDK_2G_SPLIT_SIZE;
5386	/* Do proper extent computation: need one smaller extent if the total
5387	* size isn't evenly divisible by the split size. */
5388	if (cbSize % VMDK_2G_SPLIT_SIZE)
5389	cExtents++;
5390	}
5391	rc = vmdkCreateExtents(pImage, cExtents);
5392	if (RT_FAILURE(rc))
5393	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
5394
5395	/* Basename strings needed for constructing the extent names. */
5396	char *pszBasenameSubstr = RTPathFilename(pImage->pszFilename);
5397	AssertPtr(pszBasenameSubstr);
5398	size_t cbBasenameSubstr = strlen(pszBasenameSubstr) + 1;
5399
5400	/* Create separate descriptor file if necessary. */
5401	if (cExtents != 1 \|\| (uImageFlags & VD_IMAGE_FLAGS_FIXED))
5402	{
5403	rc = vmdkFileOpen(pImage, &pImage->pFile, NULL, pImage->pszFilename,
5404	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5405	true /* fCreate */));
5406	if (RT_FAILURE(rc))
5407	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new sparse descriptor file '%s'"), pImage->pszFilename);
5408	}
5409	else
5410	pImage->pFile = NULL;
5411
5412	/* Set up all extents. */
5413	for (unsigned i = 0; i < cExtents; i++)
5414	{
5415	PVMDKEXTENT pExtent = &pImage->pExtents[i];
5416	uint64_t cbExtent = cbRemaining;
5417
5418	/* Set up fullname/basename for extent description. Cannot use StrDup
5419	* for basename, as it is not guaranteed that the memory can be freed
5420	* with RTMemTmpFree, which must be used as in other code paths
5421	* StrDup is not usable. */
5422	if (cExtents == 1 && !(uImageFlags & VD_IMAGE_FLAGS_FIXED))
5423	{
5424	char pszBasename = (char )RTMemTmpAlloc(cbBasenameSubstr);
5425	if (!pszBasename)
5426	return VERR_NO_MEMORY;
5427	memcpy(pszBasename, pszBasenameSubstr, cbBasenameSubstr);
5428	pExtent->pszBasename = pszBasename;
5429	}
5430	else
5431	{
5432	char *pszBasenameSuff = RTPathSuffix(pszBasenameSubstr);
5433	char *pszBasenameBase = RTStrDup(pszBasenameSubstr);
5434	RTPathStripSuffix(pszBasenameBase);
5435	char *pszTmp;
5436	size_t cbTmp;
5437	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
5438	{
5439	if (cExtents == 1)
5440	RTStrAPrintf(&pszTmp, "%s-flat%s", pszBasenameBase,
5441	pszBasenameSuff);
5442	else
5443	RTStrAPrintf(&pszTmp, "%s-f%03d%s", pszBasenameBase,
5444	i+1, pszBasenameSuff);
5445	}
5446	else
5447	RTStrAPrintf(&pszTmp, "%s-s%03d%s", pszBasenameBase, i+1,
5448	pszBasenameSuff);
5449	RTStrFree(pszBasenameBase);
5450	if (!pszTmp)
5451	return VERR_NO_STR_MEMORY;
5452	cbTmp = strlen(pszTmp) + 1;
5453	char pszBasename = (char )RTMemTmpAlloc(cbTmp);
5454	if (!pszBasename)
5455	{
5456	RTStrFree(pszTmp);
5457	return VERR_NO_MEMORY;
5458	}
5459	memcpy(pszBasename, pszTmp, cbTmp);
5460	RTStrFree(pszTmp);
5461	pExtent->pszBasename = pszBasename;
5462	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
5463	cbExtent = RT_MIN(cbRemaining, VMDK_2G_SPLIT_SIZE);
5464	}
5465	char *pszBasedirectory = RTStrDup(pImage->pszFilename);
5466	if (!pszBasedirectory)
5467	return VERR_NO_STR_MEMORY;
5468	RTPathStripFilename(pszBasedirectory);
5469	char *pszFullname = RTPathJoinA(pszBasedirectory, pExtent->pszBasename);
5470	RTStrFree(pszBasedirectory);
5471	if (!pszFullname)
5472	return VERR_NO_STR_MEMORY;
5473	pExtent->pszFullname = pszFullname;
5474
5475	/* Create file for extent. */
5476	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5477	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5478	true /* fCreate */));
5479	if (RT_FAILURE(rc))
5480	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pExtent->pszFullname);
5481	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
5482	{
5483	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage, cbExtent,
5484	0 /* fFlags */, pIfProgress,
5485	uPercentStart + cbOffset * uPercentSpan / cbSize,
5486	cbExtent * uPercentSpan / cbSize);
5487	if (RT_FAILURE(rc))
5488	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
5489	}
5490
5491	/* Place descriptor file information (where integrated). */
5492	if (cExtents == 1 && !(uImageFlags & VD_IMAGE_FLAGS_FIXED))
5493	{
5494	pExtent->uDescriptorSector = 1;
5495	pExtent->cDescriptorSectors = VMDK_BYTE2SECTOR(pImage->cbDescAlloc);
5496	/* The descriptor is part of the (only) extent. */
5497	pExtent->pDescData = pImage->pDescData;
5498	pImage->pDescData = NULL;
5499	}
5500
5501	if (!(uImageFlags & VD_IMAGE_FLAGS_FIXED))
5502	{
5503	uint64_t cSectorsPerGDE, cSectorsPerGD;
5504	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE;
5505	pExtent->cSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64(cbExtent, _64K));
5506	pExtent->cSectorsPerGrain = VMDK_BYTE2SECTOR(_64K);
5507	pExtent->cGTEntries = 512;
5508	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
5509	pExtent->cSectorsPerGDE = cSectorsPerGDE;
5510	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
5511	cSectorsPerGD = (pExtent->cGDEntries + (512 / sizeof(uint32_t) - 1)) / (512 / sizeof(uint32_t));
5512	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5513	{
5514	/* The spec says version is 1 for all VMDKs, but the vast
5515	* majority of streamOptimized VMDKs actually contain
5516	* version 3 - so go with the majority. Both are accepted. */
5517	pExtent->uVersion = 3;
5518	pExtent->uCompression = VMDK_COMPRESSION_DEFLATE;
5519	}
5520	}
5521	else
5522	{
5523	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
5524	pExtent->enmType = VMDKETYPE_VMFS;
5525	else
5526	pExtent->enmType = VMDKETYPE_FLAT;
5527	}
5528
5529	pExtent->enmAccess = VMDKACCESS_READWRITE;
5530	pExtent->fUncleanShutdown = true;
5531	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbExtent);
5532	pExtent->uSectorOffset = 0;
5533	pExtent->fMetaDirty = true;
5534
5535	if (!(uImageFlags & VD_IMAGE_FLAGS_FIXED))
5536	{
5537	/* fPreAlloc should never be false because VMware can't use such images. */
5538	rc = vmdkCreateGrainDirectory(pImage, pExtent,
5539	RT_MAX( pExtent->uDescriptorSector
5540	+ pExtent->cDescriptorSectors,
5541	1),
5542	true /* fPreAlloc */);
5543	if (RT_FAILURE(rc))
5544	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new grain directory in '%s'"), pExtent->pszFullname);
5545	}
5546
5547	cbOffset += cbExtent;
5548
5549	if (RT_SUCCESS(rc))
5550	vdIfProgress(pIfProgress, uPercentStart + cbOffset * uPercentSpan / cbSize);
5551
5552	cbRemaining -= cbExtent;
5553	}
5554
5555	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
5556	{
5557	/* VirtualBox doesn't care, but VMWare ESX freaks out if the wrong
5558	* controller type is set in an image. */
5559	rc = vmdkDescDDBSetStr(pImage, &pImage->Descriptor, "ddb.adapterType", "lsilogic");
5560	if (RT_FAILURE(rc))
5561	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set controller type to lsilogic in '%s'"), pImage->pszFilename);
5562	}
5563
5564	const char *pszDescType = NULL;
5565	if (uImageFlags & VD_IMAGE_FLAGS_FIXED)
5566	{
5567	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX)
5568	pszDescType = "vmfs";
5569	else
5570	pszDescType = (cExtents == 1)
5571	? "monolithicFlat" : "twoGbMaxExtentFlat";
5572	}
5573	else
5574	{
5575	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5576	pszDescType = "streamOptimized";
5577	else
5578	{
5579	pszDescType = (cExtents == 1)
5580	? "monolithicSparse" : "twoGbMaxExtentSparse";
5581	}
5582	}
5583	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
5584	pszDescType);
5585	if (RT_FAILURE(rc))
5586	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
5587	return rc;
5588	}
5589
5590	/**
5591	* Internal: Create a real stream optimized VMDK using only linear writes.
5592	*/
5593	static int vmdkCreateStreamImage(PVMDKIMAGE pImage, uint64_t cbSize)
5594	{
5595	int rc = vmdkCreateExtents(pImage, 1);
5596	if (RT_FAILURE(rc))
5597	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
5598
5599	/* Basename strings needed for constructing the extent names. */
5600	const char *pszBasenameSubstr = RTPathFilename(pImage->pszFilename);
5601	AssertPtr(pszBasenameSubstr);
5602	size_t cbBasenameSubstr = strlen(pszBasenameSubstr) + 1;
5603
5604	/* No separate descriptor file. */
5605	pImage->pFile = NULL;
5606
5607	/* Set up all extents. */
5608	PVMDKEXTENT pExtent = &pImage->pExtents[0];
5609
5610	/* Set up fullname/basename for extent description. Cannot use StrDup
5611	* for basename, as it is not guaranteed that the memory can be freed
5612	* with RTMemTmpFree, which must be used as in other code paths
5613	* StrDup is not usable. */
5614	char pszBasename = (char )RTMemTmpAlloc(cbBasenameSubstr);
5615	if (!pszBasename)
5616	return VERR_NO_MEMORY;
5617	memcpy(pszBasename, pszBasenameSubstr, cbBasenameSubstr);
5618	pExtent->pszBasename = pszBasename;
5619
5620	char *pszBasedirectory = RTStrDup(pImage->pszFilename);
5621	RTPathStripFilename(pszBasedirectory);
5622	char *pszFullname = RTPathJoinA(pszBasedirectory, pExtent->pszBasename);
5623	RTStrFree(pszBasedirectory);
5624	if (!pszFullname)
5625	return VERR_NO_STR_MEMORY;
5626	pExtent->pszFullname = pszFullname;
5627
5628	/* Create file for extent. Make it write only, no reading allowed. */
5629	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszBasename, pExtent->pszFullname,
5630	VDOpenFlagsToFileOpenFlags(pImage->uOpenFlags,
5631	true /* fCreate */)
5632	& ~RTFILE_O_READ);
5633	if (RT_FAILURE(rc))
5634	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pExtent->pszFullname);
5635
5636	/* Place descriptor file information. */
5637	pExtent->uDescriptorSector = 1;
5638	pExtent->cDescriptorSectors = VMDK_BYTE2SECTOR(pImage->cbDescAlloc);
5639	/* The descriptor is part of the (only) extent. */
5640	pExtent->pDescData = pImage->pDescData;
5641	pImage->pDescData = NULL;
5642
5643	uint64_t cSectorsPerGDE, cSectorsPerGD;
5644	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE;
5645	pExtent->cSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64(cbSize, _64K));
5646	pExtent->cSectorsPerGrain = VMDK_BYTE2SECTOR(_64K);
5647	pExtent->cGTEntries = 512;
5648	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
5649	pExtent->cSectorsPerGDE = cSectorsPerGDE;
5650	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
5651	cSectorsPerGD = (pExtent->cGDEntries + (512 / sizeof(uint32_t) - 1)) / (512 / sizeof(uint32_t));
5652
5653	/* The spec says version is 1 for all VMDKs, but the vast
5654	* majority of streamOptimized VMDKs actually contain
5655	* version 3 - so go with the majority. Both are accepted. */
5656	pExtent->uVersion = 3;
5657	pExtent->uCompression = VMDK_COMPRESSION_DEFLATE;
5658	pExtent->fFooter = true;
5659
5660	pExtent->enmAccess = VMDKACCESS_READONLY;
5661	pExtent->fUncleanShutdown = false;
5662	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
5663	pExtent->uSectorOffset = 0;
5664	pExtent->fMetaDirty = true;
5665
5666	/* Create grain directory, without preallocating it straight away. It will
5667	* be constructed on the fly when writing out the data and written when
5668	* closing the image. The end effect is that the full grain directory is
5669	* allocated, which is a requirement of the VMDK specs. */
5670	rc = vmdkCreateGrainDirectory(pImage, pExtent, VMDK_GD_AT_END,
5671	false /* fPreAlloc */);
5672	if (RT_FAILURE(rc))
5673	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new grain directory in '%s'"), pExtent->pszFullname);
5674
5675	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
5676	"streamOptimized");
5677	if (RT_FAILURE(rc))
5678	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
5679
5680	return rc;
5681	}
5682
5683	/**
5684	* Initializes the UUID fields in the DDB.
5685	*
5686	* @returns VBox status code.
5687	* @param pImage The VMDK image instance.
5688	*/
5689	static int vmdkCreateImageDdbUuidsInit(PVMDKIMAGE pImage)
5690	{
5691	int rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_IMAGE_UUID, &pImage->ImageUuid);
5692	if (RT_SUCCESS(rc))
5693	{
5694	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_PARENT_UUID, &pImage->ParentUuid);
5695	if (RT_SUCCESS(rc))
5696	{
5697	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_MODIFICATION_UUID,
5698	&pImage->ModificationUuid);
5699	if (RT_SUCCESS(rc))
5700	{
5701	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor, VMDK_DDB_PARENT_MODIFICATION_UUID,
5702	&pImage->ParentModificationUuid);
5703	if (RT_FAILURE(rc))
5704	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5705	N_("VMDK: error storing parent modification UUID in new descriptor in '%s'"), pImage->pszFilename);
5706	}
5707	else
5708	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5709	N_("VMDK: error storing modification UUID in new descriptor in '%s'"), pImage->pszFilename);
5710	}
5711	else
5712	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5713	N_("VMDK: error storing parent image UUID in new descriptor in '%s'"), pImage->pszFilename);
5714	}
5715	else
5716	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
5717	N_("VMDK: error storing image UUID in new descriptor in '%s'"), pImage->pszFilename);
5718
5719	return rc;
5720	}
5721
5722	/**
5723	* Internal: The actual code for creating any VMDK variant currently in
5724	* existence on hosted environments.
5725	*/
5726	static int vmdkCreateImage(PVMDKIMAGE pImage, uint64_t cbSize,
5727	unsigned uImageFlags, const char *pszComment,
5728	PCVDGEOMETRY pPCHSGeometry,
5729	PCVDGEOMETRY pLCHSGeometry, PCRTUUID pUuid,
5730	PVDINTERFACEPROGRESS pIfProgress,
5731	unsigned uPercentStart, unsigned uPercentSpan)
5732	{
5733	pImage->uImageFlags = uImageFlags;
5734
5735	pImage->pIfError = VDIfErrorGet(pImage->pVDIfsDisk);
5736	pImage->pIfIo = VDIfIoIntGet(pImage->pVDIfsImage);
5737	AssertPtrReturn(pImage->pIfIo, VERR_INVALID_PARAMETER);
5738
5739	int rc = vmdkCreateDescriptor(pImage, pImage->pDescData, pImage->cbDescAlloc,
5740	&pImage->Descriptor);
5741	if (RT_SUCCESS(rc))
5742	{
5743	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
5744	{
5745	/* Raw disk image (includes raw partition). */
5746	PVDISKRAW pRaw = NULL;
5747	rc = vmdkMakeRawDescriptor(pImage, &pRaw);
5748	if (RT_FAILURE(rc))
5749	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create raw descriptor for '%s'"),
5750	pImage->pszFilename);
5751	if (!cbSize)
5752	cbSize = pImage->cbSize;
5753
5754	rc = vmdkCreateRawImage(pImage, pRaw, cbSize);
5755	vmdkRawDescFree(pRaw);
5756	}
5757	else if (uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5758	{
5759	/* Stream optimized sparse image (monolithic). */
5760	rc = vmdkCreateStreamImage(pImage, cbSize);
5761	}
5762	else
5763	{
5764	/* Regular fixed or sparse image (monolithic or split). */
5765	rc = vmdkCreateRegularImage(pImage, cbSize, uImageFlags,
5766	pIfProgress, uPercentStart,
5767	uPercentSpan * 95 / 100);
5768	}
5769
5770	if (RT_SUCCESS(rc))
5771	{
5772	vdIfProgress(pIfProgress, uPercentStart + uPercentSpan * 98 / 100);
5773
5774	pImage->cbSize = cbSize;
5775
5776	for (unsigned i = 0; i < pImage->cExtents; i++)
5777	{
5778	PVMDKEXTENT pExtent = &pImage->pExtents[i];
5779
5780	rc = vmdkDescExtInsert(pImage, &pImage->Descriptor, pExtent->enmAccess,
5781	pExtent->cNominalSectors, pExtent->enmType,
5782	pExtent->pszBasename, pExtent->uSectorOffset);
5783	if (RT_FAILURE(rc))
5784	{
5785	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not insert the extent list into descriptor in '%s'"), pImage->pszFilename);
5786	break;
5787	}
5788	}
5789
5790	if (RT_SUCCESS(rc))
5791	vmdkDescExtRemoveDummy(pImage, &pImage->Descriptor);
5792
5793	pImage->LCHSGeometry = *pLCHSGeometry;
5794	pImage->PCHSGeometry = *pPCHSGeometry;
5795
5796	if (RT_SUCCESS(rc))
5797	{
5798	if ( pPCHSGeometry->cCylinders != 0
5799	&& pPCHSGeometry->cHeads != 0
5800	&& pPCHSGeometry->cSectors != 0)
5801	rc = vmdkDescSetPCHSGeometry(pImage, pPCHSGeometry);
5802	else if (uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
5803	{
5804	VDGEOMETRY RawDiskPCHSGeometry;
5805	RawDiskPCHSGeometry.cCylinders = (uint32_t)RT_MIN(pImage->cbSize / 512 / 16 / 63, 16383);
5806	RawDiskPCHSGeometry.cHeads = 16;
5807	RawDiskPCHSGeometry.cSectors = 63;
5808	rc = vmdkDescSetPCHSGeometry(pImage, &RawDiskPCHSGeometry);
5809	}
5810	}
5811
5812	if ( RT_SUCCESS(rc)
5813	&& pLCHSGeometry->cCylinders != 0
5814	&& pLCHSGeometry->cHeads != 0
5815	&& pLCHSGeometry->cSectors != 0)
5816	rc = vmdkDescSetLCHSGeometry(pImage, pLCHSGeometry);
5817
5818	pImage->ImageUuid = *pUuid;
5819	RTUuidClear(&pImage->ParentUuid);
5820	RTUuidClear(&pImage->ModificationUuid);
5821	RTUuidClear(&pImage->ParentModificationUuid);
5822
5823	if (RT_SUCCESS(rc))
5824	rc = vmdkCreateImageDdbUuidsInit(pImage);
5825
5826	if (RT_SUCCESS(rc))
5827	rc = vmdkAllocateGrainTableCache(pImage);
5828
5829	if (RT_SUCCESS(rc))
5830	{
5831	rc = vmdkSetImageComment(pImage, pszComment);
5832	if (RT_FAILURE(rc))
5833	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot set image comment in '%s'"), pImage->pszFilename);
5834	}
5835
5836	if (RT_SUCCESS(rc))
5837	{
5838	vdIfProgress(pIfProgress, uPercentStart + uPercentSpan * 99 / 100);
5839
5840	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
5841	{
5842	/* streamOptimized is a bit special, we cannot trigger the flush
5843	* until all data has been written. So we write the necessary
5844	* information explicitly. */
5845	pImage->pExtents[0].cDescriptorSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64( pImage->Descriptor.aLines[pImage->Descriptor.cLines]
5846	- pImage->Descriptor.aLines[0], 512));
5847	rc = vmdkWriteMetaSparseExtent(pImage, &pImage->pExtents[0], 0, NULL);
5848	if (RT_SUCCESS(rc))
5849	{
5850	rc = vmdkWriteDescriptor(pImage, NULL);
5851	if (RT_FAILURE(rc))
5852	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write VMDK descriptor in '%s'"), pImage->pszFilename);
5853	}
5854	else
5855	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write VMDK header in '%s'"), pImage->pszFilename);
5856	}
5857	else
5858	rc = vmdkFlushImage(pImage, NULL);
5859	}
5860	}
5861	}
5862	else
5863	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not create new descriptor in '%s'"), pImage->pszFilename);
5864
5865
5866	if (RT_SUCCESS(rc))
5867	{
5868	PVDREGIONDESC pRegion = &pImage->RegionList.aRegions[0];
5869	pImage->RegionList.fFlags = 0;
5870	pImage->RegionList.cRegions = 1;
5871
5872	pRegion->offRegion = 0; /* Disk start. */
5873	pRegion->cbBlock = 512;
5874	pRegion->enmDataForm = VDREGIONDATAFORM_RAW;
5875	pRegion->enmMetadataForm = VDREGIONMETADATAFORM_NONE;
5876	pRegion->cbData = 512;
5877	pRegion->cbMetadata = 0;
5878	pRegion->cRegionBlocksOrBytes = pImage->cbSize;
5879
5880	vdIfProgress(pIfProgress, uPercentStart + uPercentSpan);
5881	}
5882	else
5883	vmdkFreeImage(pImage, rc != VERR_ALREADY_EXISTS, false /fFlush/);
5884	return rc;
5885	}
5886
5887	/**
5888	* Internal: Update image comment.
5889	*/
5890	static int vmdkSetImageComment(PVMDKIMAGE pImage, const char *pszComment)
5891	{
5892	char *pszCommentEncoded = NULL;
5893	if (pszComment)
5894	{
5895	pszCommentEncoded = vmdkEncodeString(pszComment);
5896	if (!pszCommentEncoded)
5897	return VERR_NO_MEMORY;
5898	}
5899
5900	int rc = vmdkDescDDBSetStr(pImage, &pImage->Descriptor,
5901	"ddb.comment", pszCommentEncoded);
5902	if (pszCommentEncoded)
5903	RTStrFree(pszCommentEncoded);
5904	if (RT_FAILURE(rc))
5905	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing image comment in descriptor in '%s'"), pImage->pszFilename);
5906	return VINF_SUCCESS;
5907	}
5908
5909	/**
5910	* Internal. Clear the grain table buffer for real stream optimized writing.
5911	*/
5912	static void vmdkStreamClearGT(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
5913	{
5914	uint32_t cCacheLines = RT_ALIGN(pExtent->cGTEntries, VMDK_GT_CACHELINE_SIZE) / VMDK_GT_CACHELINE_SIZE;
5915	for (uint32_t i = 0; i < cCacheLines; i++)
5916	memset(&pImage->pGTCache->aGTCache[i].aGTData[0], '\0',
5917	VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t));
5918	}
5919
5920	/**
5921	* Internal. Flush the grain table buffer for real stream optimized writing.
5922	*/
5923	static int vmdkStreamFlushGT(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
5924	uint32_t uGDEntry)
5925	{
5926	int rc = VINF_SUCCESS;
5927	uint32_t cCacheLines = RT_ALIGN(pExtent->cGTEntries, VMDK_GT_CACHELINE_SIZE) / VMDK_GT_CACHELINE_SIZE;
5928
5929	/* VMware does not write out completely empty grain tables in the case
5930	* of streamOptimized images, which according to my interpretation of
5931	* the VMDK 1.1 spec is bending the rules. Since they do it and we can
5932	* handle it without problems do it the same way and save some bytes. */
5933	bool fAllZero = true;
5934	for (uint32_t i = 0; i < cCacheLines; i++)
5935	{
5936	/* Convert the grain table to little endian in place, as it will not
5937	* be used at all after this function has been called. */
5938	uint32_t *pGTTmp = &pImage->pGTCache->aGTCache[i].aGTData[0];
5939	for (uint32_t j = 0; j < VMDK_GT_CACHELINE_SIZE; j++, pGTTmp++)
5940	if (*pGTTmp)
5941	{
5942	fAllZero = false;
5943	break;
5944	}
5945	if (!fAllZero)
5946	break;
5947	}
5948	if (fAllZero)
5949	return VINF_SUCCESS;
5950
5951	uint64_t uFileOffset = pExtent->uAppendPosition;
5952	if (!uFileOffset)
5953	return VERR_INTERNAL_ERROR;
5954	/* Align to sector, as the previous write could have been any size. */
5955	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
5956
5957	/* Grain table marker. */
5958	uint8_t aMarker[512];
5959	PVMDKMARKER pMarker = (PVMDKMARKER)&aMarker[0];
5960	memset(pMarker, '\0', sizeof(aMarker));
5961	pMarker->uSector = RT_H2LE_U64(VMDK_BYTE2SECTOR((uint64_t)pExtent->cGTEntries * sizeof(uint32_t)));
5962	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_GT);
5963	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
5964	aMarker, sizeof(aMarker));
5965	AssertRC(rc);
5966	uFileOffset += 512;
5967
5968	if (!pExtent->pGD \|\| pExtent->pGD[uGDEntry])
5969	return VERR_INTERNAL_ERROR;
5970
5971	pExtent->pGD[uGDEntry] = VMDK_BYTE2SECTOR(uFileOffset);
5972
5973	for (uint32_t i = 0; i < cCacheLines; i++)
5974	{
5975	/* Convert the grain table to little endian in place, as it will not
5976	* be used at all after this function has been called. */
5977	uint32_t *pGTTmp = &pImage->pGTCache->aGTCache[i].aGTData[0];
5978	for (uint32_t j = 0; j < VMDK_GT_CACHELINE_SIZE; j++, pGTTmp++)
5979	pGTTmp = RT_H2LE_U32(pGTTmp);
5980
5981	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
5982	&pImage->pGTCache->aGTCache[i].aGTData[0],
5983	VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t));
5984	uFileOffset += VMDK_GT_CACHELINE_SIZE * sizeof(uint32_t);
5985	if (RT_FAILURE(rc))
5986	break;
5987	}
5988	Assert(!(uFileOffset % 512));
5989	pExtent->uAppendPosition = RT_ALIGN_64(uFileOffset, 512);
5990	return rc;
5991	}
5992
5993	/**
5994	* Internal. Free all allocated space for representing an image, and optionally
5995	* delete the image from disk.
5996	*/
5997	static int vmdkFreeImage(PVMDKIMAGE pImage, bool fDelete, bool fFlush)
5998	{
5999	int rc = VINF_SUCCESS;
6000
6001	/* Freeing a never allocated image (e.g. because the open failed) is
6002	* not signalled as an error. After all nothing bad happens. */
6003	if (pImage)
6004	{
6005	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
6006	{
6007	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6008	{
6009	/* Check if all extents are clean. */
6010	for (unsigned i = 0; i < pImage->cExtents; i++)
6011	{
6012	Assert(!pImage->pExtents[i].fUncleanShutdown);
6013	}
6014	}
6015	else
6016	{
6017	/* Mark all extents as clean. */
6018	for (unsigned i = 0; i < pImage->cExtents; i++)
6019	{
6020	if ( pImage->pExtents[i].enmType == VMDKETYPE_HOSTED_SPARSE
6021	&& pImage->pExtents[i].fUncleanShutdown)
6022	{
6023	pImage->pExtents[i].fUncleanShutdown = false;
6024	pImage->pExtents[i].fMetaDirty = true;
6025	}
6026
6027	/* From now on it's not safe to append any more data. */
6028	pImage->pExtents[i].uAppendPosition = 0;
6029	}
6030	}
6031	}
6032
6033	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6034	{
6035	/* No need to write any pending data if the file will be deleted
6036	* or if the new file wasn't successfully created. */
6037	if ( !fDelete && pImage->pExtents
6038	&& pImage->pExtents[0].cGTEntries
6039	&& pImage->pExtents[0].uAppendPosition)
6040	{
6041	PVMDKEXTENT pExtent = &pImage->pExtents[0];
6042	uint32_t uLastGDEntry = pExtent->uLastGrainAccess / pExtent->cGTEntries;
6043	rc = vmdkStreamFlushGT(pImage, pExtent, uLastGDEntry);
6044	AssertRC(rc);
6045	vmdkStreamClearGT(pImage, pExtent);
6046	for (uint32_t i = uLastGDEntry + 1; i < pExtent->cGDEntries; i++)
6047	{
6048	rc = vmdkStreamFlushGT(pImage, pExtent, i);
6049	AssertRC(rc);
6050	}
6051
6052	uint64_t uFileOffset = pExtent->uAppendPosition;
6053	if (!uFileOffset)
6054	return VERR_INTERNAL_ERROR;
6055	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
6056
6057	/* From now on it's not safe to append any more data. */
6058	pExtent->uAppendPosition = 0;
6059
6060	/* Grain directory marker. */
6061	uint8_t aMarker[512];
6062	PVMDKMARKER pMarker = (PVMDKMARKER)&aMarker[0];
6063	memset(pMarker, '\0', sizeof(aMarker));
6064	pMarker->uSector = VMDK_BYTE2SECTOR(RT_ALIGN_64(RT_H2LE_U64((uint64_t)pExtent->cGDEntries * sizeof(uint32_t)), 512));
6065	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_GD);
6066	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage, uFileOffset,
6067	aMarker, sizeof(aMarker));
6068	AssertRC(rc);
6069	uFileOffset += 512;
6070
6071	/* Write grain directory in little endian style. The array will
6072	* not be used after this, so convert in place. */
6073	uint32_t *pGDTmp = pExtent->pGD;
6074	for (uint32_t i = 0; i < pExtent->cGDEntries; i++, pGDTmp++)
6075	pGDTmp = RT_H2LE_U32(pGDTmp);
6076	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
6077	uFileOffset, pExtent->pGD,
6078	pExtent->cGDEntries * sizeof(uint32_t));
6079	AssertRC(rc);
6080
6081	pExtent->uSectorGD = VMDK_BYTE2SECTOR(uFileOffset);
6082	pExtent->uSectorRGD = VMDK_BYTE2SECTOR(uFileOffset);
6083	uFileOffset = RT_ALIGN_64( uFileOffset
6084	+ pExtent->cGDEntries * sizeof(uint32_t),
6085	512);
6086
6087	/* Footer marker. */
6088	memset(pMarker, '\0', sizeof(aMarker));
6089	pMarker->uSector = VMDK_BYTE2SECTOR(512);
6090	pMarker->uType = RT_H2LE_U32(VMDK_MARKER_FOOTER);
6091	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
6092	uFileOffset, aMarker, sizeof(aMarker));
6093	AssertRC(rc);
6094
6095	uFileOffset += 512;
6096	rc = vmdkWriteMetaSparseExtent(pImage, pExtent, uFileOffset, NULL);
6097	AssertRC(rc);
6098
6099	uFileOffset += 512;
6100	/* End-of-stream marker. */
6101	memset(pMarker, '\0', sizeof(aMarker));
6102	rc = vdIfIoIntFileWriteSync(pImage->pIfIo, pExtent->pFile->pStorage,
6103	uFileOffset, aMarker, sizeof(aMarker));
6104	AssertRC(rc);
6105	}
6106	}
6107	else if (!fDelete && fFlush)
6108	vmdkFlushImage(pImage, NULL);
6109
6110	if (pImage->pExtents != NULL)
6111	{
6112	for (unsigned i = 0 ; i < pImage->cExtents; i++)
6113	{
6114	int rc2 = vmdkFreeExtentData(pImage, &pImage->pExtents[i], fDelete);
6115	if (RT_SUCCESS(rc))
6116	rc = rc2; /* Propogate any error when closing the file. */
6117	}
6118	RTMemFree(pImage->pExtents);
6119	pImage->pExtents = NULL;
6120	}
6121	pImage->cExtents = 0;
6122	if (pImage->pFile != NULL)
6123	{
6124	int rc2 = vmdkFileClose(pImage, &pImage->pFile, fDelete);
6125	if (RT_SUCCESS(rc))
6126	rc = rc2; /* Propogate any error when closing the file. */
6127	}
6128	int rc2 = vmdkFileCheckAllClose(pImage);
6129	if (RT_SUCCESS(rc))
6130	rc = rc2; /* Propogate any error when closing the file. */
6131
6132	if (pImage->pGTCache)
6133	{
6134	RTMemFree(pImage->pGTCache);
6135	pImage->pGTCache = NULL;
6136	}
6137	if (pImage->pDescData)
6138	{
6139	RTMemFree(pImage->pDescData);
6140	pImage->pDescData = NULL;
6141	}
6142	}
6143
6144	LogFlowFunc(("returns %Rrc\n", rc));
6145	return rc;
6146	}
6147
6148	/**
6149	* Internal. Flush image data (and metadata) to disk.
6150	*/
6151	static int vmdkFlushImage(PVMDKIMAGE pImage, PVDIOCTX pIoCtx)
6152	{
6153	PVMDKEXTENT pExtent;
6154	int rc = VINF_SUCCESS;
6155
6156	/* Update descriptor if changed. */
6157	if (pImage->Descriptor.fDirty)
6158	rc = vmdkWriteDescriptor(pImage, pIoCtx);
6159
6160	if (RT_SUCCESS(rc))
6161	{
6162	for (unsigned i = 0; i < pImage->cExtents; i++)
6163	{
6164	pExtent = &pImage->pExtents[i];
6165	if (pExtent->pFile != NULL && pExtent->fMetaDirty)
6166	{
6167	switch (pExtent->enmType)
6168	{
6169	case VMDKETYPE_HOSTED_SPARSE:
6170	if (!pExtent->fFooter)
6171	rc = vmdkWriteMetaSparseExtent(pImage, pExtent, 0, pIoCtx);
6172	else
6173	{
6174	uint64_t uFileOffset = pExtent->uAppendPosition;
6175	/* Simply skip writing anything if the streamOptimized
6176	* image hasn't been just created. */
6177	if (!uFileOffset)
6178	break;
6179	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
6180	rc = vmdkWriteMetaSparseExtent(pImage, pExtent,
6181	uFileOffset, pIoCtx);
6182	}
6183	break;
6184	case VMDKETYPE_VMFS:
6185	case VMDKETYPE_FLAT:
6186	/* Nothing to do. */
6187	break;
6188	case VMDKETYPE_ZERO:
6189	default:
6190	AssertMsgFailed(("extent with type %d marked as dirty\n",
6191	pExtent->enmType));
6192	break;
6193	}
6194	}
6195
6196	if (RT_FAILURE(rc))
6197	break;
6198
6199	switch (pExtent->enmType)
6200	{
6201	case VMDKETYPE_HOSTED_SPARSE:
6202	case VMDKETYPE_VMFS:
6203	case VMDKETYPE_FLAT:
6204	/** @todo implement proper path absolute check. */
6205	if ( pExtent->pFile != NULL
6206	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
6207	&& !(pExtent->pszBasename[0] == RTPATH_SLASH))
6208	rc = vdIfIoIntFileFlush(pImage->pIfIo, pExtent->pFile->pStorage, pIoCtx,
6209	NULL, NULL);
6210	break;
6211	case VMDKETYPE_ZERO:
6212	/* No need to do anything for this extent. */
6213	break;
6214	default:
6215	AssertMsgFailed(("unknown extent type %d\n", pExtent->enmType));
6216	break;
6217	}
6218	}
6219	}
6220
6221	return rc;
6222	}
6223
6224	/**
6225	* Internal. Find extent corresponding to the sector number in the disk.
6226	*/
6227	static int vmdkFindExtent(PVMDKIMAGE pImage, uint64_t offSector,
6228	PVMDKEXTENT ppExtent, uint64_t puSectorInExtent)
6229	{
6230	PVMDKEXTENT pExtent = NULL;
6231	int rc = VINF_SUCCESS;
6232
6233	for (unsigned i = 0; i < pImage->cExtents; i++)
6234	{
6235	if (offSector < pImage->pExtents[i].cNominalSectors)
6236	{
6237	pExtent = &pImage->pExtents[i];
6238	*puSectorInExtent = offSector + pImage->pExtents[i].uSectorOffset;
6239	break;
6240	}
6241	offSector -= pImage->pExtents[i].cNominalSectors;
6242	}
6243
6244	if (pExtent)
6245	*ppExtent = pExtent;
6246	else
6247	rc = VERR_IO_SECTOR_NOT_FOUND;
6248
6249	return rc;
6250	}
6251
6252	/**
6253	* Internal. Hash function for placing the grain table hash entries.
6254	*/
6255	static uint32_t vmdkGTCacheHash(PVMDKGTCACHE pCache, uint64_t uSector,
6256	unsigned uExtent)
6257	{
6258	/** @todo this hash function is quite simple, maybe use a better one which
6259	* scrambles the bits better. */
6260	return (uSector + uExtent) % pCache->cEntries;
6261	}
6262
6263	/**
6264	* Internal. Get sector number in the extent file from the relative sector
6265	* number in the extent.
6266	*/
6267	static int vmdkGetSector(PVMDKIMAGE pImage, PVDIOCTX pIoCtx,
6268	PVMDKEXTENT pExtent, uint64_t uSector,
6269	uint64_t *puExtentSector)
6270	{
6271	PVMDKGTCACHE pCache = pImage->pGTCache;
6272	uint64_t uGDIndex, uGTSector, uGTBlock;
6273	uint32_t uGTHash, uGTBlockIndex;
6274	PVMDKGTCACHEENTRY pGTCacheEntry;
6275	uint32_t aGTDataTmp[VMDK_GT_CACHELINE_SIZE];
6276	int rc;
6277
6278	/* For newly created and readonly/sequentially opened streamOptimized
6279	* images this must be a no-op, as the grain directory is not there. */
6280	if ( ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
6281	&& pExtent->uAppendPosition)
6282	\|\| ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
6283	&& pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY
6284	&& pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
6285	{
6286	*puExtentSector = 0;
6287	return VINF_SUCCESS;
6288	}
6289
6290	uGDIndex = uSector / pExtent->cSectorsPerGDE;
6291	if (uGDIndex >= pExtent->cGDEntries)
6292	return VERR_OUT_OF_RANGE;
6293	uGTSector = pExtent->pGD[uGDIndex];
6294	if (!uGTSector)
6295	{
6296	/* There is no grain table referenced by this grain directory
6297	* entry. So there is absolutely no data in this area. */
6298	*puExtentSector = 0;
6299	return VINF_SUCCESS;
6300	}
6301
6302	uGTBlock = uSector / (pExtent->cSectorsPerGrain * VMDK_GT_CACHELINE_SIZE);
6303	uGTHash = vmdkGTCacheHash(pCache, uGTBlock, pExtent->uExtent);
6304	pGTCacheEntry = &pCache->aGTCache[uGTHash];
6305	if ( pGTCacheEntry->uExtent != pExtent->uExtent
6306	\|\| pGTCacheEntry->uGTBlock != uGTBlock)
6307	{
6308	/* Cache miss, fetch data from disk. */
6309	PVDMETAXFER pMetaXfer;
6310	rc = vdIfIoIntFileReadMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6311	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
6312	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx, &pMetaXfer, NULL, NULL);
6313	if (RT_FAILURE(rc))
6314	return rc;
6315	/* We can release the metadata transfer immediately. */
6316	vdIfIoIntMetaXferRelease(pImage->pIfIo, pMetaXfer);
6317	pGTCacheEntry->uExtent = pExtent->uExtent;
6318	pGTCacheEntry->uGTBlock = uGTBlock;
6319	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
6320	pGTCacheEntry->aGTData[i] = RT_LE2H_U32(aGTDataTmp[i]);
6321	}
6322	uGTBlockIndex = (uSector / pExtent->cSectorsPerGrain) % VMDK_GT_CACHELINE_SIZE;
6323	uint32_t uGrainSector = pGTCacheEntry->aGTData[uGTBlockIndex];
6324	if (uGrainSector)
6325	*puExtentSector = uGrainSector + uSector % pExtent->cSectorsPerGrain;
6326	else
6327	*puExtentSector = 0;
6328	return VINF_SUCCESS;
6329	}
6330
6331	/**
6332	* Internal. Writes the grain and also if necessary the grain tables.
6333	* Uses the grain table cache as a true grain table.
6334	*/
6335	static int vmdkStreamAllocGrain(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
6336	uint64_t uSector, PVDIOCTX pIoCtx,
6337	uint64_t cbWrite)
6338	{
6339	uint32_t uGrain;
6340	uint32_t uGDEntry, uLastGDEntry;
6341	uint32_t cbGrain = 0;
6342	uint32_t uCacheLine, uCacheEntry;
6343	const void *pData;
6344	int rc;
6345
6346	/* Very strict requirements: always write at least one full grain, with
6347	* proper alignment. Everything else would require reading of already
6348	* written data, which we don't support for obvious reasons. The only
6349	* exception is the last grain, and only if the image size specifies
6350	* that only some portion holds data. In any case the write must be
6351	* within the image limits, no "overshoot" allowed. */
6352	if ( cbWrite == 0
6353	\|\| ( cbWrite < VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain)
6354	&& pExtent->cNominalSectors - uSector >= pExtent->cSectorsPerGrain)
6355	\|\| uSector % pExtent->cSectorsPerGrain
6356	\|\| uSector + VMDK_BYTE2SECTOR(cbWrite) > pExtent->cNominalSectors)
6357	return VERR_INVALID_PARAMETER;
6358
6359	/* Clip write range to at most the rest of the grain. */
6360	cbWrite = RT_MIN(cbWrite, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain - uSector % pExtent->cSectorsPerGrain));
6361
6362	/* Do not allow to go back. */
6363	uGrain = uSector / pExtent->cSectorsPerGrain;
6364	uCacheLine = uGrain % pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE;
6365	uCacheEntry = uGrain % VMDK_GT_CACHELINE_SIZE;
6366	uGDEntry = uGrain / pExtent->cGTEntries;
6367	uLastGDEntry = pExtent->uLastGrainAccess / pExtent->cGTEntries;
6368	if (uGrain < pExtent->uLastGrainAccess)
6369	return VERR_VD_VMDK_INVALID_WRITE;
6370
6371	/* Zero byte write optimization. Since we don't tell VBoxHDD that we need
6372	* to allocate something, we also need to detect the situation ourself. */
6373	if ( !(pImage->uOpenFlags & VD_OPEN_FLAGS_HONOR_ZEROES)
6374	&& vdIfIoIntIoCtxIsZero(pImage->pIfIo, pIoCtx, cbWrite, true /* fAdvance */))
6375	return VINF_SUCCESS;
6376
6377	if (uGDEntry != uLastGDEntry)
6378	{
6379	rc = vmdkStreamFlushGT(pImage, pExtent, uLastGDEntry);
6380	if (RT_FAILURE(rc))
6381	return rc;
6382	vmdkStreamClearGT(pImage, pExtent);
6383	for (uint32_t i = uLastGDEntry + 1; i < uGDEntry; i++)
6384	{
6385	rc = vmdkStreamFlushGT(pImage, pExtent, i);
6386	if (RT_FAILURE(rc))
6387	return rc;
6388	}
6389	}
6390
6391	uint64_t uFileOffset;
6392	uFileOffset = pExtent->uAppendPosition;
6393	if (!uFileOffset)
6394	return VERR_INTERNAL_ERROR;
6395	/* Align to sector, as the previous write could have been any size. */
6396	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
6397
6398	/* Paranoia check: extent type, grain table buffer presence and
6399	* grain table buffer space. Also grain table entry must be clear. */
6400	if ( pExtent->enmType != VMDKETYPE_HOSTED_SPARSE
6401	\|\| !pImage->pGTCache
6402	\|\| pExtent->cGTEntries > VMDK_GT_CACHE_SIZE * VMDK_GT_CACHELINE_SIZE
6403	\|\| pImage->pGTCache->aGTCache[uCacheLine].aGTData[uCacheEntry])
6404	return VERR_INTERNAL_ERROR;
6405
6406	/* Update grain table entry. */
6407	pImage->pGTCache->aGTCache[uCacheLine].aGTData[uCacheEntry] = VMDK_BYTE2SECTOR(uFileOffset);
6408
6409	if (cbWrite != VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
6410	{
6411	vdIfIoIntIoCtxCopyFrom(pImage->pIfIo, pIoCtx, pExtent->pvGrain, cbWrite);
6412	memset((char *)pExtent->pvGrain + cbWrite, '\0',
6413	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain) - cbWrite);
6414	pData = pExtent->pvGrain;
6415	}
6416	else
6417	{
6418	RTSGSEG Segment;
6419	unsigned cSegments = 1;
6420	size_t cbSeg = 0;
6421
6422	cbSeg = vdIfIoIntIoCtxSegArrayCreate(pImage->pIfIo, pIoCtx, &Segment,
6423	&cSegments, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
6424	Assert(cbSeg == VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
6425	pData = Segment.pvSeg;
6426	}
6427	rc = vmdkFileDeflateSync(pImage, pExtent, uFileOffset, pData,
6428	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
6429	uSector, &cbGrain);
6430	if (RT_FAILURE(rc))
6431	{
6432	pExtent->uGrainSectorAbs = 0;
6433	AssertRC(rc);
6434	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write compressed data block in '%s'"), pExtent->pszFullname);
6435	}
6436	pExtent->uLastGrainAccess = uGrain;
6437	pExtent->uAppendPosition += cbGrain;
6438
6439	return rc;
6440	}
6441
6442	/**
6443	* Internal: Updates the grain table during grain allocation.
6444	*/
6445	static int vmdkAllocGrainGTUpdate(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, PVDIOCTX pIoCtx,
6446	PVMDKGRAINALLOCASYNC pGrainAlloc)
6447	{
6448	int rc = VINF_SUCCESS;
6449	PVMDKGTCACHE pCache = pImage->pGTCache;
6450	uint32_t aGTDataTmp[VMDK_GT_CACHELINE_SIZE];
6451	uint32_t uGTHash, uGTBlockIndex;
6452	uint64_t uGTSector, uRGTSector, uGTBlock;
6453	uint64_t uSector = pGrainAlloc->uSector;
6454	PVMDKGTCACHEENTRY pGTCacheEntry;
6455
6456	LogFlowFunc(("pImage=%#p pExtent=%#p pCache=%#p pIoCtx=%#p pGrainAlloc=%#p\n",
6457	pImage, pExtent, pCache, pIoCtx, pGrainAlloc));
6458
6459	uGTSector = pGrainAlloc->uGTSector;
6460	uRGTSector = pGrainAlloc->uRGTSector;
6461	LogFlow(("uGTSector=%llu uRGTSector=%llu\n", uGTSector, uRGTSector));
6462
6463	/* Update the grain table (and the cache). */
6464	uGTBlock = uSector / (pExtent->cSectorsPerGrain * VMDK_GT_CACHELINE_SIZE);
6465	uGTHash = vmdkGTCacheHash(pCache, uGTBlock, pExtent->uExtent);
6466	pGTCacheEntry = &pCache->aGTCache[uGTHash];
6467	if ( pGTCacheEntry->uExtent != pExtent->uExtent
6468	\|\| pGTCacheEntry->uGTBlock != uGTBlock)
6469	{
6470	/* Cache miss, fetch data from disk. */
6471	LogFlow(("Cache miss, fetch data from disk\n"));
6472	PVDMETAXFER pMetaXfer = NULL;
6473	rc = vdIfIoIntFileReadMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6474	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
6475	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
6476	&pMetaXfer, vmdkAllocGrainComplete, pGrainAlloc);
6477	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6478	{
6479	pGrainAlloc->cIoXfersPending++;
6480	pGrainAlloc->fGTUpdateNeeded = true;
6481	/* Leave early, we will be called again after the read completed. */
6482	LogFlowFunc(("Metadata read in progress, leaving\n"));
6483	return rc;
6484	}
6485	else if (RT_FAILURE(rc))
6486	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot read allocated grain table entry in '%s'"), pExtent->pszFullname);
6487	vdIfIoIntMetaXferRelease(pImage->pIfIo, pMetaXfer);
6488	pGTCacheEntry->uExtent = pExtent->uExtent;
6489	pGTCacheEntry->uGTBlock = uGTBlock;
6490	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
6491	pGTCacheEntry->aGTData[i] = RT_LE2H_U32(aGTDataTmp[i]);
6492	}
6493	else
6494	{
6495	/* Cache hit. Convert grain table block back to disk format, otherwise
6496	* the code below will write garbage for all but the updated entry. */
6497	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
6498	aGTDataTmp[i] = RT_H2LE_U32(pGTCacheEntry->aGTData[i]);
6499	}
6500	pGrainAlloc->fGTUpdateNeeded = false;
6501	uGTBlockIndex = (uSector / pExtent->cSectorsPerGrain) % VMDK_GT_CACHELINE_SIZE;
6502	aGTDataTmp[uGTBlockIndex] = RT_H2LE_U32(VMDK_BYTE2SECTOR(pGrainAlloc->uGrainOffset));
6503	pGTCacheEntry->aGTData[uGTBlockIndex] = VMDK_BYTE2SECTOR(pGrainAlloc->uGrainOffset);
6504	/* Update grain table on disk. */
6505	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6506	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
6507	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
6508	vmdkAllocGrainComplete, pGrainAlloc);
6509	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6510	pGrainAlloc->cIoXfersPending++;
6511	else if (RT_FAILURE(rc))
6512	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write updated grain table in '%s'"), pExtent->pszFullname);
6513	if (pExtent->pRGD)
6514	{
6515	/* Update backup grain table on disk. */
6516	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6517	VMDK_SECTOR2BYTE(uRGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
6518	aGTDataTmp, sizeof(aGTDataTmp), pIoCtx,
6519	vmdkAllocGrainComplete, pGrainAlloc);
6520	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6521	pGrainAlloc->cIoXfersPending++;
6522	else if (RT_FAILURE(rc))
6523	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write updated backup grain table in '%s'"), pExtent->pszFullname);
6524	}
6525
6526	LogFlowFunc(("leaving rc=%Rrc\n", rc));
6527	return rc;
6528	}
6529
6530	/**
6531	* Internal - complete the grain allocation by updating disk grain table if required.
6532	*/
6533	static DECLCALLBACK(int) vmdkAllocGrainComplete(void pBackendData, PVDIOCTX pIoCtx, void pvUser, int rcReq)
6534	{
6535	RT_NOREF1(rcReq);
6536	int rc = VINF_SUCCESS;
6537	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
6538	PVMDKGRAINALLOCASYNC pGrainAlloc = (PVMDKGRAINALLOCASYNC)pvUser;
6539
6540	LogFlowFunc(("pBackendData=%#p pIoCtx=%#p pvUser=%#p rcReq=%Rrc\n",
6541	pBackendData, pIoCtx, pvUser, rcReq));
6542
6543	pGrainAlloc->cIoXfersPending--;
6544	if (!pGrainAlloc->cIoXfersPending && pGrainAlloc->fGTUpdateNeeded)
6545	rc = vmdkAllocGrainGTUpdate(pImage, pGrainAlloc->pExtent, pIoCtx, pGrainAlloc);
6546
6547	if (!pGrainAlloc->cIoXfersPending)
6548	{
6549	/* Grain allocation completed. */
6550	RTMemFree(pGrainAlloc);
6551	}
6552
6553	LogFlowFunc(("Leaving rc=%Rrc\n", rc));
6554	return rc;
6555	}
6556
6557	/**
6558	* Internal. Allocates a new grain table (if necessary).
6559	*/
6560	static int vmdkAllocGrain(PVMDKIMAGE pImage, PVMDKEXTENT pExtent, PVDIOCTX pIoCtx,
6561	uint64_t uSector, uint64_t cbWrite)
6562	{
6563	PVMDKGTCACHE pCache = pImage->pGTCache; NOREF(pCache);
6564	uint64_t uGDIndex, uGTSector, uRGTSector;
6565	uint64_t uFileOffset;
6566	PVMDKGRAINALLOCASYNC pGrainAlloc = NULL;
6567	int rc;
6568
6569	LogFlowFunc(("pCache=%#p pExtent=%#p pIoCtx=%#p uSector=%llu cbWrite=%llu\n",
6570	pCache, pExtent, pIoCtx, uSector, cbWrite));
6571
6572	pGrainAlloc = (PVMDKGRAINALLOCASYNC)RTMemAllocZ(sizeof(VMDKGRAINALLOCASYNC));
6573	if (!pGrainAlloc)
6574	return VERR_NO_MEMORY;
6575
6576	pGrainAlloc->pExtent = pExtent;
6577	pGrainAlloc->uSector = uSector;
6578
6579	uGDIndex = uSector / pExtent->cSectorsPerGDE;
6580	if (uGDIndex >= pExtent->cGDEntries)
6581	{
6582	RTMemFree(pGrainAlloc);
6583	return VERR_OUT_OF_RANGE;
6584	}
6585	uGTSector = pExtent->pGD[uGDIndex];
6586	if (pExtent->pRGD)
6587	uRGTSector = pExtent->pRGD[uGDIndex];
6588	else
6589	uRGTSector = 0; /*< avoid compiler warning /
6590	if (!uGTSector)
6591	{
6592	LogFlow(("Allocating new grain table\n"));
6593
6594	/* There is no grain table referenced by this grain directory
6595	* entry. So there is absolutely no data in this area. Allocate
6596	* a new grain table and put the reference to it in the GDs. */
6597	uFileOffset = pExtent->uAppendPosition;
6598	if (!uFileOffset)
6599	{
6600	RTMemFree(pGrainAlloc);
6601	return VERR_INTERNAL_ERROR;
6602	}
6603	Assert(!(uFileOffset % 512));
6604
6605	uFileOffset = RT_ALIGN_64(uFileOffset, 512);
6606	uGTSector = VMDK_BYTE2SECTOR(uFileOffset);
6607
6608	/* Normally the grain table is preallocated for hosted sparse extents
6609	* that support more than 32 bit sector numbers. So this shouldn't
6610	* ever happen on a valid extent. */
6611	if (uGTSector > UINT32_MAX)
6612	{
6613	RTMemFree(pGrainAlloc);
6614	return VERR_VD_VMDK_INVALID_HEADER;
6615	}
6616
6617	/* Write grain table by writing the required number of grain table
6618	* cache chunks. Allocate memory dynamically here or we flood the
6619	* metadata cache with very small entries. */
6620	size_t cbGTDataTmp = pExtent->cGTEntries * sizeof(uint32_t);
6621	uint32_t paGTDataTmp = (uint32_t )RTMemTmpAllocZ(cbGTDataTmp);
6622
6623	if (!paGTDataTmp)
6624	{
6625	RTMemFree(pGrainAlloc);
6626	return VERR_NO_MEMORY;
6627	}
6628
6629	memset(paGTDataTmp, '\0', cbGTDataTmp);
6630	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6631	VMDK_SECTOR2BYTE(uGTSector),
6632	paGTDataTmp, cbGTDataTmp, pIoCtx,
6633	vmdkAllocGrainComplete, pGrainAlloc);
6634	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6635	pGrainAlloc->cIoXfersPending++;
6636	else if (RT_FAILURE(rc))
6637	{
6638	RTMemTmpFree(paGTDataTmp);
6639	RTMemFree(pGrainAlloc);
6640	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write grain table allocation in '%s'"), pExtent->pszFullname);
6641	}
6642	pExtent->uAppendPosition = RT_ALIGN_64( pExtent->uAppendPosition
6643	+ cbGTDataTmp, 512);
6644
6645	if (pExtent->pRGD)
6646	{
6647	AssertReturn(!uRGTSector, VERR_VD_VMDK_INVALID_HEADER);
6648	uFileOffset = pExtent->uAppendPosition;
6649	if (!uFileOffset)
6650	return VERR_INTERNAL_ERROR;
6651	Assert(!(uFileOffset % 512));
6652	uRGTSector = VMDK_BYTE2SECTOR(uFileOffset);
6653
6654	/* Normally the redundant grain table is preallocated for hosted
6655	* sparse extents that support more than 32 bit sector numbers. So
6656	* this shouldn't ever happen on a valid extent. */
6657	if (uRGTSector > UINT32_MAX)
6658	{
6659	RTMemTmpFree(paGTDataTmp);
6660	return VERR_VD_VMDK_INVALID_HEADER;
6661	}
6662
6663	/* Write grain table by writing the required number of grain table
6664	* cache chunks. Allocate memory dynamically here or we flood the
6665	* metadata cache with very small entries. */
6666	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6667	VMDK_SECTOR2BYTE(uRGTSector),
6668	paGTDataTmp, cbGTDataTmp, pIoCtx,
6669	vmdkAllocGrainComplete, pGrainAlloc);
6670	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6671	pGrainAlloc->cIoXfersPending++;
6672	else if (RT_FAILURE(rc))
6673	{
6674	RTMemTmpFree(paGTDataTmp);
6675	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write backup grain table allocation in '%s'"), pExtent->pszFullname);
6676	}
6677
6678	pExtent->uAppendPosition = pExtent->uAppendPosition + cbGTDataTmp;
6679	}
6680
6681	RTMemTmpFree(paGTDataTmp);
6682
6683	/* Update the grain directory on disk (doing it before writing the
6684	* grain table will result in a garbled extent if the operation is
6685	* aborted for some reason. Otherwise the worst that can happen is
6686	* some unused sectors in the extent. */
6687	uint32_t uGTSectorLE = RT_H2LE_U64(uGTSector);
6688	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6689	VMDK_SECTOR2BYTE(pExtent->uSectorGD) + uGDIndex * sizeof(uGTSectorLE),
6690	&uGTSectorLE, sizeof(uGTSectorLE), pIoCtx,
6691	vmdkAllocGrainComplete, pGrainAlloc);
6692	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6693	pGrainAlloc->cIoXfersPending++;
6694	else if (RT_FAILURE(rc))
6695	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write grain directory entry in '%s'"), pExtent->pszFullname);
6696	if (pExtent->pRGD)
6697	{
6698	uint32_t uRGTSectorLE = RT_H2LE_U64(uRGTSector);
6699	rc = vdIfIoIntFileWriteMeta(pImage->pIfIo, pExtent->pFile->pStorage,
6700	VMDK_SECTOR2BYTE(pExtent->uSectorRGD) + uGDIndex * sizeof(uGTSectorLE),
6701	&uRGTSectorLE, sizeof(uRGTSectorLE), pIoCtx,
6702	vmdkAllocGrainComplete, pGrainAlloc);
6703	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6704	pGrainAlloc->cIoXfersPending++;
6705	else if (RT_FAILURE(rc))
6706	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write backup grain directory entry in '%s'"), pExtent->pszFullname);
6707	}
6708
6709	/* As the final step update the in-memory copy of the GDs. */
6710	pExtent->pGD[uGDIndex] = uGTSector;
6711	if (pExtent->pRGD)
6712	pExtent->pRGD[uGDIndex] = uRGTSector;
6713	}
6714
6715	LogFlow(("uGTSector=%llu uRGTSector=%llu\n", uGTSector, uRGTSector));
6716	pGrainAlloc->uGTSector = uGTSector;
6717	pGrainAlloc->uRGTSector = uRGTSector;
6718
6719	uFileOffset = pExtent->uAppendPosition;
6720	if (!uFileOffset)
6721	return VERR_INTERNAL_ERROR;
6722	Assert(!(uFileOffset % 512));
6723
6724	pGrainAlloc->uGrainOffset = uFileOffset;
6725
6726	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
6727	{
6728	AssertMsgReturn(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
6729	("Accesses to stream optimized images must be synchronous\n"),
6730	VERR_INVALID_STATE);
6731
6732	if (cbWrite != VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
6733	return vdIfError(pImage->pIfError, VERR_INTERNAL_ERROR, RT_SRC_POS, N_("VMDK: not enough data for a compressed data block in '%s'"), pExtent->pszFullname);
6734
6735	/* Invalidate cache, just in case some code incorrectly allows mixing
6736	* of reads and writes. Normally shouldn't be needed. */
6737	pExtent->uGrainSectorAbs = 0;
6738
6739	/* Write compressed data block and the markers. */
6740	uint32_t cbGrain = 0;
6741	size_t cbSeg = 0;
6742	RTSGSEG Segment;
6743	unsigned cSegments = 1;
6744
6745	cbSeg = vdIfIoIntIoCtxSegArrayCreate(pImage->pIfIo, pIoCtx, &Segment,
6746	&cSegments, cbWrite);
6747	Assert(cbSeg == cbWrite);
6748
6749	rc = vmdkFileDeflateSync(pImage, pExtent, uFileOffset,
6750	Segment.pvSeg, cbWrite, uSector, &cbGrain);
6751	if (RT_FAILURE(rc))
6752	{
6753	AssertRC(rc);
6754	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write allocated compressed data block in '%s'"), pExtent->pszFullname);
6755	}
6756	pExtent->uLastGrainAccess = uSector / pExtent->cSectorsPerGrain;
6757	pExtent->uAppendPosition += cbGrain;
6758	}
6759	else
6760	{
6761	/* Write the data. Always a full grain, or we're in big trouble. */
6762	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
6763	uFileOffset, pIoCtx, cbWrite,
6764	vmdkAllocGrainComplete, pGrainAlloc);
6765	if (rc == VERR_VD_ASYNC_IO_IN_PROGRESS)
6766	pGrainAlloc->cIoXfersPending++;
6767	else if (RT_FAILURE(rc))
6768	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: cannot write allocated data block in '%s'"), pExtent->pszFullname);
6769
6770	pExtent->uAppendPosition += cbWrite;
6771	}
6772
6773	rc = vmdkAllocGrainGTUpdate(pImage, pExtent, pIoCtx, pGrainAlloc);
6774
6775	if (!pGrainAlloc->cIoXfersPending)
6776	{
6777	/* Grain allocation completed. */
6778	RTMemFree(pGrainAlloc);
6779	}
6780
6781	LogFlowFunc(("leaving rc=%Rrc\n", rc));
6782
6783	return rc;
6784	}
6785
6786	/**
6787	* Internal. Reads the contents by sequentially going over the compressed
6788	* grains (hoping that they are in sequence).
6789	*/
6790	static int vmdkStreamReadSequential(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
6791	uint64_t uSector, PVDIOCTX pIoCtx,
6792	uint64_t cbRead)
6793	{
6794	int rc;
6795
6796	LogFlowFunc(("pImage=%#p pExtent=%#p uSector=%llu pIoCtx=%#p cbRead=%llu\n",
6797	pImage, pExtent, uSector, pIoCtx, cbRead));
6798
6799	AssertMsgReturn(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
6800	("Async I/O not supported for sequential stream optimized images\n"),
6801	VERR_INVALID_STATE);
6802
6803	/* Do not allow to go back. */
6804	uint32_t uGrain = uSector / pExtent->cSectorsPerGrain;
6805	if (uGrain < pExtent->uLastGrainAccess)
6806	return VERR_VD_VMDK_INVALID_STATE;
6807	pExtent->uLastGrainAccess = uGrain;
6808
6809	/* After a previous error do not attempt to recover, as it would need
6810	* seeking (in the general case backwards which is forbidden). */
6811	if (!pExtent->uGrainSectorAbs)
6812	return VERR_VD_VMDK_INVALID_STATE;
6813
6814	/* Check if we need to read something from the image or if what we have
6815	* in the buffer is good to fulfill the request. */
6816	if (!pExtent->cbGrainStreamRead \|\| uGrain > pExtent->uGrain)
6817	{
6818	uint32_t uGrainSectorAbs = pExtent->uGrainSectorAbs
6819	+ VMDK_BYTE2SECTOR(pExtent->cbGrainStreamRead);
6820
6821	/* Get the marker from the next data block - and skip everything which
6822	* is not a compressed grain. If it's a compressed grain which is for
6823	* the requested sector (or after), read it. */
6824	VMDKMARKER Marker;
6825	do
6826	{
6827	RT_ZERO(Marker);
6828	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
6829	VMDK_SECTOR2BYTE(uGrainSectorAbs),
6830	&Marker, RT_UOFFSETOF(VMDKMARKER, uType));
6831	if (RT_FAILURE(rc))
6832	return rc;
6833	Marker.uSector = RT_LE2H_U64(Marker.uSector);
6834	Marker.cbSize = RT_LE2H_U32(Marker.cbSize);
6835
6836	if (Marker.cbSize == 0)
6837	{
6838	/* A marker for something else than a compressed grain. */
6839	rc = vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
6840	VMDK_SECTOR2BYTE(uGrainSectorAbs)
6841	+ RT_UOFFSETOF(VMDKMARKER, uType),
6842	&Marker.uType, sizeof(Marker.uType));
6843	if (RT_FAILURE(rc))
6844	return rc;
6845	Marker.uType = RT_LE2H_U32(Marker.uType);
6846	switch (Marker.uType)
6847	{
6848	case VMDK_MARKER_EOS:
6849	uGrainSectorAbs++;
6850	/* Read (or mostly skip) to the end of file. Uses the
6851	* Marker (LBA sector) as it is unused anyway. This
6852	* makes sure that really everything is read in the
6853	* success case. If this read fails it means the image
6854	* is truncated, but this is harmless so ignore. */
6855	vdIfIoIntFileReadSync(pImage->pIfIo, pExtent->pFile->pStorage,
6856	VMDK_SECTOR2BYTE(uGrainSectorAbs)
6857	+ 511,
6858	&Marker.uSector, 1);
6859	break;
6860	case VMDK_MARKER_GT:
6861	uGrainSectorAbs += 1 + VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
6862	break;
6863	case VMDK_MARKER_GD:
6864	uGrainSectorAbs += 1 + VMDK_BYTE2SECTOR(RT_ALIGN(pExtent->cGDEntries * sizeof(uint32_t), 512));
6865	break;
6866	case VMDK_MARKER_FOOTER:
6867	uGrainSectorAbs += 2;
6868	break;
6869	case VMDK_MARKER_UNSPECIFIED:
6870	/* Skip over the contents of the unspecified marker
6871	* type 4 which exists in some vSphere created files. */
6872	/** @todo figure out what the payload means. */
6873	uGrainSectorAbs += 1;
6874	break;
6875	default:
6876	AssertMsgFailed(("VMDK: corrupted marker, type=%#x\n", Marker.uType));
6877	pExtent->uGrainSectorAbs = 0;
6878	return VERR_VD_VMDK_INVALID_STATE;
6879	}
6880	pExtent->cbGrainStreamRead = 0;
6881	}
6882	else
6883	{
6884	/* A compressed grain marker. If it is at/after what we're
6885	* interested in read and decompress data. */
6886	if (uSector > Marker.uSector + pExtent->cSectorsPerGrain)
6887	{
6888	uGrainSectorAbs += VMDK_BYTE2SECTOR(RT_ALIGN(Marker.cbSize + RT_UOFFSETOF(VMDKMARKER, uType), 512));
6889	continue;
6890	}
6891	uint64_t uLBA = 0;
6892	uint32_t cbGrainStreamRead = 0;
6893	rc = vmdkFileInflateSync(pImage, pExtent,
6894	VMDK_SECTOR2BYTE(uGrainSectorAbs),
6895	pExtent->pvGrain,
6896	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
6897	&Marker, &uLBA, &cbGrainStreamRead);
6898	if (RT_FAILURE(rc))
6899	{
6900	pExtent->uGrainSectorAbs = 0;
6901	return rc;
6902	}
6903	if ( pExtent->uGrain
6904	&& uLBA / pExtent->cSectorsPerGrain <= pExtent->uGrain)
6905	{
6906	pExtent->uGrainSectorAbs = 0;
6907	return VERR_VD_VMDK_INVALID_STATE;
6908	}
6909	pExtent->uGrain = uLBA / pExtent->cSectorsPerGrain;
6910	pExtent->cbGrainStreamRead = cbGrainStreamRead;
6911	break;
6912	}
6913	} while (Marker.uType != VMDK_MARKER_EOS);
6914
6915	pExtent->uGrainSectorAbs = uGrainSectorAbs;
6916
6917	if (!pExtent->cbGrainStreamRead && Marker.uType == VMDK_MARKER_EOS)
6918	{
6919	pExtent->uGrain = UINT32_MAX;
6920	/* Must set a non-zero value for pExtent->cbGrainStreamRead or
6921	* the next read would try to get more data, and we're at EOF. */
6922	pExtent->cbGrainStreamRead = 1;
6923	}
6924	}
6925
6926	if (pExtent->uGrain > uSector / pExtent->cSectorsPerGrain)
6927	{
6928	/* The next data block we have is not for this area, so just return
6929	* that there is no data. */
6930	LogFlowFunc(("returns VERR_VD_BLOCK_FREE\n"));
6931	return VERR_VD_BLOCK_FREE;
6932	}
6933
6934	uint32_t uSectorInGrain = uSector % pExtent->cSectorsPerGrain;
6935	vdIfIoIntIoCtxCopyTo(pImage->pIfIo, pIoCtx,
6936	(uint8_t *)pExtent->pvGrain + VMDK_SECTOR2BYTE(uSectorInGrain),
6937	cbRead);
6938	LogFlowFunc(("returns VINF_SUCCESS\n"));
6939	return VINF_SUCCESS;
6940	}
6941
6942	/**
6943	* Replaces a fragment of a string with the specified string.
6944	*
6945	* @returns Pointer to the allocated UTF-8 string.
6946	* @param pszWhere UTF-8 string to search in.
6947	* @param pszWhat UTF-8 string to search for.
6948	* @param pszByWhat UTF-8 string to replace the found string with.
6949	*
6950	* @note r=bird: This is only used by vmdkRenameWorker(). The first use is
6951	* for updating the base name in the descriptor, the second is for
6952	* generating new filenames for extents. This code borked when
6953	* RTPathAbs started correcting the driver letter case on windows,
6954	* when strstr failed because the pExtent->pszFullname was not
6955	* subjected to RTPathAbs but while pExtent->pszFullname was. I fixed
6956	* this by apply RTPathAbs to the places it wasn't applied.
6957	*
6958	* However, this highlights some undocumented ASSUMPTIONS as well as
6959	* terrible short commings of the approach.
6960	*
6961	* Given the right filename, it may also screw up the descriptor. Take
6962	* the descriptor text 'RW 2048 SPARSE "Test0.vmdk"' for instance,
6963	* we'll be asked to replace "Test0" with something, no problem. No,
6964	* imagine 'RW 2048 SPARSE "SPARSE.vmdk"', 'RW 2048 SPARSE "RW.vmdk"'
6965	* or 'RW 2048 SPARSE "2048.vmdk"', and the strstr approach falls on
6966	* its bum. The descriptor string must be parsed and reconstructed,
6967	* the lazy strstr approach doesn't cut it.
6968	*
6969	* I'm also curious as to what would be the correct escaping of '"' in
6970	* the file name and how that is supposed to be handled, because it
6971	* needs to be or such names must be rejected in several places (maybe
6972	* they are, I didn't check).
6973	*
6974	* When this function is used to replace the start of a path, I think
6975	* the assumption from the prep/setup code is that we kind of knows
6976	* what we're working on (I could be wrong). However, using strstr
6977	* instead of strncmp/RTStrNICmp makes no sense and isn't future proof.
6978	* Especially on unix systems, weird stuff could happen if someone
6979	* unwittingly tinkers with the prep/setup code. What should really be
6980	* done here is using a new RTPathStartEx function that (via flags)
6981	* allows matching partial final component and returns the length of
6982	* what it matched up (in case it skipped slashes and '.' components).
6983	*
6984	*/
6985	static char vmdkStrReplace(const char pszWhere, const char *pszWhat,
6986	const char *pszByWhat)
6987	{
6988	AssertPtr(pszWhere);
6989	AssertPtr(pszWhat);
6990	AssertPtr(pszByWhat);
6991	const char *pszFoundStr = strstr(pszWhere, pszWhat);
6992	if (!pszFoundStr)
6993	{
6994	LogFlowFunc(("Failed to find '%s' in '%s'!\n", pszWhat, pszWhere));
6995	return NULL;
6996	}
6997	size_t cbFinal = strlen(pszWhere) + 1 + strlen(pszByWhat) - strlen(pszWhat);
6998	char *pszNewStr = RTStrAlloc(cbFinal);
6999	if (pszNewStr)
7000	{
7001	char *pszTmp = pszNewStr;
7002	memcpy(pszTmp, pszWhere, pszFoundStr - pszWhere);
7003	pszTmp += pszFoundStr - pszWhere;
7004	memcpy(pszTmp, pszByWhat, strlen(pszByWhat));
7005	pszTmp += strlen(pszByWhat);
7006	strcpy(pszTmp, pszFoundStr + strlen(pszWhat));
7007	}
7008	return pszNewStr;
7009	}
7010
7011
7012	/** @copydoc VDIMAGEBACKEND::pfnProbe */
7013	static DECLCALLBACK(int) vmdkProbe(const char *pszFilename, PVDINTERFACE pVDIfsDisk,
7014	PVDINTERFACE pVDIfsImage, VDTYPE enmDesiredType, VDTYPE *penmType)
7015	{
7016	RT_NOREF(enmDesiredType);
7017	LogFlowFunc(("pszFilename=\"%s\" pVDIfsDisk=%#p pVDIfsImage=%#p penmType=%#p\n",
7018	pszFilename, pVDIfsDisk, pVDIfsImage, penmType));
7019	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
7020	AssertReturn(*pszFilename != '\0', VERR_INVALID_PARAMETER);
7021
7022	int rc = VINF_SUCCESS;
7023	PVMDKIMAGE pImage = (PVMDKIMAGE)RTMemAllocZ(RT_UOFFSETOF(VMDKIMAGE, RegionList.aRegions[1]));
7024	if (RT_LIKELY(pImage))
7025	{
7026	pImage->pszFilename = pszFilename;
7027	pImage->pFile = NULL;
7028	pImage->pExtents = NULL;
7029	pImage->pFiles = NULL;
7030	pImage->pGTCache = NULL;
7031	pImage->pDescData = NULL;
7032	pImage->pVDIfsDisk = pVDIfsDisk;
7033	pImage->pVDIfsImage = pVDIfsImage;
7034	/** @todo speed up this test open (VD_OPEN_FLAGS_INFO) by skipping as
7035	* much as possible in vmdkOpenImage. */
7036	rc = vmdkOpenImage(pImage, VD_OPEN_FLAGS_INFO \| VD_OPEN_FLAGS_READONLY);
7037	vmdkFreeImage(pImage, false, false /fFlush/);
7038	RTMemFree(pImage);
7039
7040	if (RT_SUCCESS(rc))
7041	*penmType = VDTYPE_HDD;
7042	}
7043	else
7044	rc = VERR_NO_MEMORY;
7045
7046	LogFlowFunc(("returns %Rrc\n", rc));
7047	return rc;
7048	}
7049
7050	/** @copydoc VDIMAGEBACKEND::pfnOpen */
7051	static DECLCALLBACK(int) vmdkOpen(const char *pszFilename, unsigned uOpenFlags,
7052	PVDINTERFACE pVDIfsDisk, PVDINTERFACE pVDIfsImage,
7053	VDTYPE enmType, void **ppBackendData)
7054	{
7055	RT_NOREF1(enmType); /*< @todo r=klaus make use of the type info. /
7056
7057	LogFlowFunc(("pszFilename=\"%s\" uOpenFlags=%#x pVDIfsDisk=%#p pVDIfsImage=%#p enmType=%u ppBackendData=%#p\n",
7058	pszFilename, uOpenFlags, pVDIfsDisk, pVDIfsImage, enmType, ppBackendData));
7059	int rc;
7060
7061	/* Check open flags. All valid flags are supported. */
7062	AssertReturn(!(uOpenFlags & ~VD_OPEN_FLAGS_MASK), VERR_INVALID_PARAMETER);
7063	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
7064	AssertReturn(*pszFilename != '\0', VERR_INVALID_PARAMETER);
7065
7066
7067	PVMDKIMAGE pImage = (PVMDKIMAGE)RTMemAllocZ(RT_UOFFSETOF(VMDKIMAGE, RegionList.aRegions[1]));
7068	if (RT_LIKELY(pImage))
7069	{
7070	pImage->pszFilename = pszFilename;
7071	pImage->pFile = NULL;
7072	pImage->pExtents = NULL;
7073	pImage->pFiles = NULL;
7074	pImage->pGTCache = NULL;
7075	pImage->pDescData = NULL;
7076	pImage->pVDIfsDisk = pVDIfsDisk;
7077	pImage->pVDIfsImage = pVDIfsImage;
7078
7079	rc = vmdkOpenImage(pImage, uOpenFlags);
7080	if (RT_SUCCESS(rc))
7081	*ppBackendData = pImage;
7082	else
7083	RTMemFree(pImage);
7084	}
7085	else
7086	rc = VERR_NO_MEMORY;
7087
7088	LogFlowFunc(("returns %Rrc (pBackendData=%#p)\n", rc, *ppBackendData));
7089	return rc;
7090	}
7091
7092	/** @copydoc VDIMAGEBACKEND::pfnCreate */
7093	static DECLCALLBACK(int) vmdkCreate(const char *pszFilename, uint64_t cbSize,
7094	unsigned uImageFlags, const char *pszComment,
7095	PCVDGEOMETRY pPCHSGeometry, PCVDGEOMETRY pLCHSGeometry,
7096	PCRTUUID pUuid, unsigned uOpenFlags,
7097	unsigned uPercentStart, unsigned uPercentSpan,
7098	PVDINTERFACE pVDIfsDisk, PVDINTERFACE pVDIfsImage,
7099	PVDINTERFACE pVDIfsOperation, VDTYPE enmType,
7100	void **ppBackendData)
7101	{
7102	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",
7103	pszFilename, cbSize, uImageFlags, pszComment, pPCHSGeometry, pLCHSGeometry, pUuid, uOpenFlags, uPercentStart, uPercentSpan, pVDIfsDisk, pVDIfsImage, pVDIfsOperation, enmType, ppBackendData));
7104	int rc;
7105
7106	/* Check the VD container type and image flags. */
7107	if ( enmType != VDTYPE_HDD
7108	\|\| (uImageFlags & ~VD_VMDK_IMAGE_FLAGS_MASK) != 0)
7109	return VERR_VD_INVALID_TYPE;
7110
7111	/* Check size. Maximum 256TB-64K for sparse images, otherwise unlimited. */
7112	if ( !(uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
7113	&& ( !cbSize
7114	\|\| (!(uImageFlags & VD_IMAGE_FLAGS_FIXED) && cbSize >= _1T * 256 - _64K)))
7115	return VERR_VD_INVALID_SIZE;
7116
7117	/* Check image flags for invalid combinations. */
7118	if ( (uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7119	&& (uImageFlags & ~(VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED \| VD_IMAGE_FLAGS_DIFF)))
7120	return VERR_INVALID_PARAMETER;
7121
7122	/* Check open flags. All valid flags are supported. */
7123	AssertReturn(!(uOpenFlags & ~VD_OPEN_FLAGS_MASK), VERR_INVALID_PARAMETER);
7124	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
7125	AssertReturn(*pszFilename != '\0', VERR_INVALID_PARAMETER);
7126	AssertPtrReturn(pPCHSGeometry, VERR_INVALID_POINTER);
7127	AssertPtrReturn(pLCHSGeometry, VERR_INVALID_POINTER);
7128	AssertReturn(!( uImageFlags & VD_VMDK_IMAGE_FLAGS_ESX
7129	&& !(uImageFlags & VD_IMAGE_FLAGS_FIXED)),
7130	VERR_INVALID_PARAMETER);
7131
7132	PVMDKIMAGE pImage = (PVMDKIMAGE)RTMemAllocZ(RT_UOFFSETOF(VMDKIMAGE, RegionList.aRegions[1]));
7133	if (RT_LIKELY(pImage))
7134	{
7135	PVDINTERFACEPROGRESS pIfProgress = VDIfProgressGet(pVDIfsOperation);
7136
7137	pImage->pszFilename = pszFilename;
7138	pImage->pFile = NULL;
7139	pImage->pExtents = NULL;
7140	pImage->pFiles = NULL;
7141	pImage->pGTCache = NULL;
7142	pImage->pDescData = NULL;
7143	pImage->pVDIfsDisk = pVDIfsDisk;
7144	pImage->pVDIfsImage = pVDIfsImage;
7145	/* Descriptors for split images can be pretty large, especially if the
7146	* filename is long. So prepare for the worst, and allocate quite some
7147	* memory for the descriptor in this case. */
7148	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
7149	pImage->cbDescAlloc = VMDK_SECTOR2BYTE(200);
7150	else
7151	pImage->cbDescAlloc = VMDK_SECTOR2BYTE(20);
7152	pImage->pDescData = (char *)RTMemAllocZ(pImage->cbDescAlloc);
7153	if (RT_LIKELY(pImage->pDescData))
7154	{
7155	rc = vmdkCreateImage(pImage, cbSize, uImageFlags, pszComment,
7156	pPCHSGeometry, pLCHSGeometry, pUuid,
7157	pIfProgress, uPercentStart, uPercentSpan);
7158	if (RT_SUCCESS(rc))
7159	{
7160	/* So far the image is opened in read/write mode. Make sure the
7161	* image is opened in read-only mode if the caller requested that. */
7162	if (uOpenFlags & VD_OPEN_FLAGS_READONLY)
7163	{
7164	vmdkFreeImage(pImage, false, true /fFlush/);
7165	rc = vmdkOpenImage(pImage, uOpenFlags);
7166	}
7167
7168	if (RT_SUCCESS(rc))
7169	*ppBackendData = pImage;
7170	}
7171
7172	if (RT_FAILURE(rc))
7173	RTMemFree(pImage->pDescData);
7174	}
7175	else
7176	rc = VERR_NO_MEMORY;
7177
7178	if (RT_FAILURE(rc))
7179	RTMemFree(pImage);
7180	}
7181	else
7182	rc = VERR_NO_MEMORY;
7183
7184	LogFlowFunc(("returns %Rrc (pBackendData=%#p)\n", rc, *ppBackendData));
7185	return rc;
7186	}
7187
7188	/**
7189	* Prepares the state for renaming a VMDK image, setting up the state and allocating
7190	* memory.
7191	*
7192	* @returns VBox status code.
7193	* @param pImage VMDK image instance.
7194	* @param pRenameState The state to initialize.
7195	* @param pszFilename The new filename.
7196	*/
7197	static int vmdkRenameStatePrepare(PVMDKIMAGE pImage, PVMDKRENAMESTATE pRenameState, const char *pszFilename)
7198	{
7199	AssertReturn(RTPathFilename(pszFilename) != NULL, VERR_INVALID_PARAMETER);
7200
7201	int rc = VINF_SUCCESS;
7202
7203	memset(&pRenameState->DescriptorCopy, 0, sizeof(pRenameState->DescriptorCopy));
7204
7205	/*
7206	* Allocate an array to store both old and new names of renamed files
7207	* in case we have to roll back the changes. Arrays are initialized
7208	* with zeros. We actually save stuff when and if we change it.
7209	*/
7210	pRenameState->cExtents = pImage->cExtents;
7211	pRenameState->apszOldName = (char *)RTMemTmpAllocZ((pRenameState->cExtents + 1) sizeof(char *));
7212	pRenameState->apszNewName = (char *)RTMemTmpAllocZ((pRenameState->cExtents + 1) sizeof(char *));
7213	pRenameState->apszNewLines = (char *)RTMemTmpAllocZ(pRenameState->cExtents sizeof(char *));
7214	if ( pRenameState->apszOldName
7215	&& pRenameState->apszNewName
7216	&& pRenameState->apszNewLines)
7217	{
7218	/* Save the descriptor size and position. */
7219	if (pImage->pDescData)
7220	{
7221	/* Separate descriptor file. */
7222	pRenameState->fEmbeddedDesc = false;
7223	}
7224	else
7225	{
7226	/* Embedded descriptor file. */
7227	pRenameState->ExtentCopy = pImage->pExtents[0];
7228	pRenameState->fEmbeddedDesc = true;
7229	}
7230
7231	/* Save the descriptor content. */
7232	pRenameState->DescriptorCopy.cLines = pImage->Descriptor.cLines;
7233	for (unsigned i = 0; i < pRenameState->DescriptorCopy.cLines; i++)
7234	{
7235	pRenameState->DescriptorCopy.aLines[i] = RTStrDup(pImage->Descriptor.aLines[i]);
7236	if (!pRenameState->DescriptorCopy.aLines[i])
7237	{
7238	rc = VERR_NO_MEMORY;
7239	break;
7240	}
7241	}
7242
7243	if (RT_SUCCESS(rc))
7244	{
7245	/* Prepare both old and new base names used for string replacement. */
7246	pRenameState->pszNewBaseName = RTStrDup(RTPathFilename(pszFilename));
7247	AssertReturn(pRenameState->pszNewBaseName, VERR_NO_STR_MEMORY);
7248	RTPathStripSuffix(pRenameState->pszNewBaseName);
7249
7250	pRenameState->pszOldBaseName = RTStrDup(RTPathFilename(pImage->pszFilename));
7251	AssertReturn(pRenameState->pszOldBaseName, VERR_NO_STR_MEMORY);
7252	RTPathStripSuffix(pRenameState->pszOldBaseName);
7253
7254	/* Prepare both old and new full names used for string replacement.
7255	Note! Must abspath the stuff here, so the strstr weirdness later in
7256	the renaming process get a match against abspath'ed extent paths.
7257	See RTPathAbsDup call in vmdkDescriptorReadSparse(). */
7258	pRenameState->pszNewFullName = RTPathAbsDup(pszFilename);
7259	AssertReturn(pRenameState->pszNewFullName, VERR_NO_STR_MEMORY);
7260	RTPathStripSuffix(pRenameState->pszNewFullName);
7261
7262	pRenameState->pszOldFullName = RTPathAbsDup(pImage->pszFilename);
7263	AssertReturn(pRenameState->pszOldFullName, VERR_NO_STR_MEMORY);
7264	RTPathStripSuffix(pRenameState->pszOldFullName);
7265
7266	/* Save the old name for easy access to the old descriptor file. */
7267	pRenameState->pszOldDescName = RTStrDup(pImage->pszFilename);
7268	AssertReturn(pRenameState->pszOldDescName, VERR_NO_STR_MEMORY);
7269
7270	/* Save old image name. */
7271	pRenameState->pszOldImageName = pImage->pszFilename;
7272	}
7273	}
7274	else
7275	rc = VERR_NO_TMP_MEMORY;
7276
7277	return rc;
7278	}
7279
7280	/**
7281	* Destroys the given rename state, freeing all allocated memory.
7282	*
7283	* @returns nothing.
7284	* @param pRenameState The rename state to destroy.
7285	*/
7286	static void vmdkRenameStateDestroy(PVMDKRENAMESTATE pRenameState)
7287	{
7288	for (unsigned i = 0; i < pRenameState->DescriptorCopy.cLines; i++)
7289	if (pRenameState->DescriptorCopy.aLines[i])
7290	RTStrFree(pRenameState->DescriptorCopy.aLines[i]);
7291	if (pRenameState->apszOldName)
7292	{
7293	for (unsigned i = 0; i <= pRenameState->cExtents; i++)
7294	if (pRenameState->apszOldName[i])
7295	RTStrFree(pRenameState->apszOldName[i]);
7296	RTMemTmpFree(pRenameState->apszOldName);
7297	}
7298	if (pRenameState->apszNewName)
7299	{
7300	for (unsigned i = 0; i <= pRenameState->cExtents; i++)
7301	if (pRenameState->apszNewName[i])
7302	RTStrFree(pRenameState->apszNewName[i]);
7303	RTMemTmpFree(pRenameState->apszNewName);
7304	}
7305	if (pRenameState->apszNewLines)
7306	{
7307	for (unsigned i = 0; i < pRenameState->cExtents; i++)
7308	if (pRenameState->apszNewLines[i])
7309	RTStrFree(pRenameState->apszNewLines[i]);
7310	RTMemTmpFree(pRenameState->apszNewLines);
7311	}
7312	if (pRenameState->pszOldDescName)
7313	RTStrFree(pRenameState->pszOldDescName);
7314	if (pRenameState->pszOldBaseName)
7315	RTStrFree(pRenameState->pszOldBaseName);
7316	if (pRenameState->pszNewBaseName)
7317	RTStrFree(pRenameState->pszNewBaseName);
7318	if (pRenameState->pszOldFullName)
7319	RTStrFree(pRenameState->pszOldFullName);
7320	if (pRenameState->pszNewFullName)
7321	RTStrFree(pRenameState->pszNewFullName);
7322	}
7323
7324	/**
7325	* Rolls back the rename operation to the original state.
7326	*
7327	* @returns VBox status code.
7328	* @param pImage VMDK image instance.
7329	* @param pRenameState The rename state.
7330	*/
7331	static int vmdkRenameRollback(PVMDKIMAGE pImage, PVMDKRENAMESTATE pRenameState)
7332	{
7333	int rc = VINF_SUCCESS;
7334
7335	if (!pRenameState->fImageFreed)
7336	{
7337	/*
7338	* Some extents may have been closed, close the rest. We will
7339	* re-open the whole thing later.
7340	*/
7341	vmdkFreeImage(pImage, false, true /fFlush/);
7342	}
7343
7344	/* Rename files back. */
7345	for (unsigned i = 0; i <= pRenameState->cExtents; i++)
7346	{
7347	if (pRenameState->apszOldName[i])
7348	{
7349	rc = vdIfIoIntFileMove(pImage->pIfIo, pRenameState->apszNewName[i], pRenameState->apszOldName[i], 0);
7350	AssertRC(rc);
7351	}
7352	}
7353	/* Restore the old descriptor. */
7354	PVMDKFILE pFile;
7355	rc = vmdkFileOpen(pImage, &pFile, NULL, pRenameState->pszOldDescName,
7356	VDOpenFlagsToFileOpenFlags(VD_OPEN_FLAGS_NORMAL,
7357	false /* fCreate */));
7358	AssertRC(rc);
7359	if (pRenameState->fEmbeddedDesc)
7360	{
7361	pRenameState->ExtentCopy.pFile = pFile;
7362	pImage->pExtents = &pRenameState->ExtentCopy;
7363	}
7364	else
7365	{
7366	/* Shouldn't be null for separate descriptor.
7367	* There will be no access to the actual content.
7368	*/
7369	pImage->pDescData = pRenameState->pszOldDescName;
7370	pImage->pFile = pFile;
7371	}
7372	pImage->Descriptor = pRenameState->DescriptorCopy;
7373	vmdkWriteDescriptor(pImage, NULL);
7374	vmdkFileClose(pImage, &pFile, false);
7375	/* Get rid of the stuff we implanted. */
7376	pImage->pExtents = NULL;
7377	pImage->pFile = NULL;
7378	pImage->pDescData = NULL;
7379	/* Re-open the image back. */
7380	pImage->pszFilename = pRenameState->pszOldImageName;
7381	rc = vmdkOpenImage(pImage, pImage->uOpenFlags);
7382
7383	return rc;
7384	}
7385
7386	/**
7387	* Rename worker doing the real work.
7388	*
7389	* @returns VBox status code.
7390	* @param pImage VMDK image instance.
7391	* @param pRenameState The rename state.
7392	* @param pszFilename The new filename.
7393	*/
7394	static int vmdkRenameWorker(PVMDKIMAGE pImage, PVMDKRENAMESTATE pRenameState, const char *pszFilename)
7395	{
7396	int rc = VINF_SUCCESS;
7397	unsigned i, line;
7398
7399	/* Update the descriptor with modified extent names. */
7400	for (i = 0, line = pImage->Descriptor.uFirstExtent;
7401	i < pRenameState->cExtents;
7402	i++, line = pImage->Descriptor.aNextLines[line])
7403	{
7404	/* Update the descriptor. */
7405	pRenameState->apszNewLines[i] = vmdkStrReplace(pImage->Descriptor.aLines[line],
7406	pRenameState->pszOldBaseName,
7407	pRenameState->pszNewBaseName);
7408	if (!pRenameState->apszNewLines[i])
7409	{
7410	rc = VERR_NO_MEMORY;
7411	break;
7412	}
7413	pImage->Descriptor.aLines[line] = pRenameState->apszNewLines[i];
7414	}
7415
7416	if (RT_SUCCESS(rc))
7417	{
7418	/* Make sure the descriptor gets written back. */
7419	pImage->Descriptor.fDirty = true;
7420	/* Flush the descriptor now, in case it is embedded. */
7421	vmdkFlushImage(pImage, NULL);
7422
7423	/* Close and rename/move extents. */
7424	for (i = 0; i < pRenameState->cExtents; i++)
7425	{
7426	PVMDKEXTENT pExtent = &pImage->pExtents[i];
7427	/* Compose new name for the extent. */
7428	pRenameState->apszNewName[i] = vmdkStrReplace(pExtent->pszFullname,
7429	pRenameState->pszOldFullName,
7430	pRenameState->pszNewFullName);
7431	if (!pRenameState->apszNewName[i])
7432	{
7433	rc = VERR_NO_MEMORY;
7434	break;
7435	}
7436	/* Close the extent file. */
7437	rc = vmdkFileClose(pImage, &pExtent->pFile, false);
7438	if (RT_FAILURE(rc))
7439	break;;
7440
7441	/* Rename the extent file. */
7442	rc = vdIfIoIntFileMove(pImage->pIfIo, pExtent->pszFullname, pRenameState->apszNewName[i], 0);
7443	if (RT_FAILURE(rc))
7444	break;
7445	/* Remember the old name. */
7446	pRenameState->apszOldName[i] = RTStrDup(pExtent->pszFullname);
7447	}
7448
7449	if (RT_SUCCESS(rc))
7450	{
7451	/* Release all old stuff. */
7452	rc = vmdkFreeImage(pImage, false, true /fFlush/);
7453	if (RT_SUCCESS(rc))
7454	{
7455	pRenameState->fImageFreed = true;
7456
7457	/* Last elements of new/old name arrays are intended for
7458	* storing descriptor's names.
7459	*/
7460	pRenameState->apszNewName[pRenameState->cExtents] = RTStrDup(pszFilename);
7461	/* Rename the descriptor file if it's separate. */
7462	if (!pRenameState->fEmbeddedDesc)
7463	{
7464	rc = vdIfIoIntFileMove(pImage->pIfIo, pImage->pszFilename, pRenameState->apszNewName[pRenameState->cExtents], 0);
7465	if (RT_SUCCESS(rc))
7466	{
7467	/* Save old name only if we may need to change it back. */
7468	pRenameState->apszOldName[pRenameState->cExtents] = RTStrDup(pszFilename);
7469	}
7470	}
7471
7472	/* Update pImage with the new information. */
7473	pImage->pszFilename = pszFilename;
7474
7475	/* Open the new image. */
7476	rc = vmdkOpenImage(pImage, pImage->uOpenFlags);
7477	}
7478	}
7479	}
7480
7481	return rc;
7482	}
7483
7484	/** @copydoc VDIMAGEBACKEND::pfnRename */
7485	static DECLCALLBACK(int) vmdkRename(void pBackendData, const char pszFilename)
7486	{
7487	LogFlowFunc(("pBackendData=%#p pszFilename=%#p\n", pBackendData, pszFilename));
7488
7489	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7490	VMDKRENAMESTATE RenameState;
7491
7492	memset(&RenameState, 0, sizeof(RenameState));
7493
7494	/* Check arguments. */
7495	AssertPtrReturn(pImage, VERR_INVALID_POINTER);
7496	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
7497	AssertReturn(*pszFilename != '\0', VERR_INVALID_PARAMETER);
7498	AssertReturn(!(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK), VERR_INVALID_PARAMETER);
7499
7500	int rc = vmdkRenameStatePrepare(pImage, &RenameState, pszFilename);
7501	if (RT_SUCCESS(rc))
7502	{
7503	/* --- Up to this point we have not done any damage yet. --- */
7504
7505	rc = vmdkRenameWorker(pImage, &RenameState, pszFilename);
7506	/* Roll back all changes in case of failure. */
7507	if (RT_FAILURE(rc))
7508	{
7509	int rrc = vmdkRenameRollback(pImage, &RenameState);
7510	AssertRC(rrc);
7511	}
7512	}
7513
7514	vmdkRenameStateDestroy(&RenameState);
7515	LogFlowFunc(("returns %Rrc\n", rc));
7516	return rc;
7517	}
7518
7519	/** @copydoc VDIMAGEBACKEND::pfnClose */
7520	static DECLCALLBACK(int) vmdkClose(void *pBackendData, bool fDelete)
7521	{
7522	LogFlowFunc(("pBackendData=%#p fDelete=%d\n", pBackendData, fDelete));
7523	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7524
7525	int rc = vmdkFreeImage(pImage, fDelete, true /fFlush/);
7526	RTMemFree(pImage);
7527
7528	LogFlowFunc(("returns %Rrc\n", rc));
7529	return rc;
7530	}
7531
7532	/** @copydoc VDIMAGEBACKEND::pfnRead */
7533	static DECLCALLBACK(int) vmdkRead(void *pBackendData, uint64_t uOffset, size_t cbToRead,
7534	PVDIOCTX pIoCtx, size_t *pcbActuallyRead)
7535	{
7536	LogFlowFunc(("pBackendData=%#p uOffset=%llu pIoCtx=%#p cbToRead=%zu pcbActuallyRead=%#p\n",
7537	pBackendData, uOffset, pIoCtx, cbToRead, pcbActuallyRead));
7538	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7539
7540	AssertPtr(pImage);
7541	Assert(uOffset % 512 == 0);
7542	Assert(cbToRead % 512 == 0);
7543	AssertPtrReturn(pIoCtx, VERR_INVALID_POINTER);
7544	AssertReturn(cbToRead, VERR_INVALID_PARAMETER);
7545	AssertReturn(uOffset + cbToRead <= pImage->cbSize, VERR_INVALID_PARAMETER);
7546
7547	/* Find the extent and check access permissions as defined in the extent descriptor. */
7548	PVMDKEXTENT pExtent;
7549	uint64_t uSectorExtentRel;
7550	int rc = vmdkFindExtent(pImage, VMDK_BYTE2SECTOR(uOffset),
7551	&pExtent, &uSectorExtentRel);
7552	if ( RT_SUCCESS(rc)
7553	&& pExtent->enmAccess != VMDKACCESS_NOACCESS)
7554	{
7555	/* Clip read range to remain in this extent. */
7556	cbToRead = RT_MIN(cbToRead, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
7557
7558	/* Handle the read according to the current extent type. */
7559	switch (pExtent->enmType)
7560	{
7561	case VMDKETYPE_HOSTED_SPARSE:
7562	{
7563	uint64_t uSectorExtentAbs;
7564
7565	rc = vmdkGetSector(pImage, pIoCtx, pExtent, uSectorExtentRel, &uSectorExtentAbs);
7566	if (RT_FAILURE(rc))
7567	break;
7568	/* Clip read range to at most the rest of the grain. */
7569	cbToRead = RT_MIN(cbToRead, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain - uSectorExtentRel % pExtent->cSectorsPerGrain));
7570	Assert(!(cbToRead % 512));
7571	if (uSectorExtentAbs == 0)
7572	{
7573	if ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7574	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
7575	\|\| !(pImage->uOpenFlags & VD_OPEN_FLAGS_SEQUENTIAL))
7576	rc = VERR_VD_BLOCK_FREE;
7577	else
7578	rc = vmdkStreamReadSequential(pImage, pExtent,
7579	uSectorExtentRel,
7580	pIoCtx, cbToRead);
7581	}
7582	else
7583	{
7584	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7585	{
7586	AssertMsg(vdIfIoIntIoCtxIsSynchronous(pImage->pIfIo, pIoCtx),
7587	("Async I/O is not supported for stream optimized VMDK's\n"));
7588
7589	uint32_t uSectorInGrain = uSectorExtentRel % pExtent->cSectorsPerGrain;
7590	uSectorExtentAbs -= uSectorInGrain;
7591	if (pExtent->uGrainSectorAbs != uSectorExtentAbs)
7592	{
7593	uint64_t uLBA = 0; /* gcc maybe uninitialized */
7594	rc = vmdkFileInflateSync(pImage, pExtent,
7595	VMDK_SECTOR2BYTE(uSectorExtentAbs),
7596	pExtent->pvGrain,
7597	VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain),
7598	NULL, &uLBA, NULL);
7599	if (RT_FAILURE(rc))
7600	{
7601	pExtent->uGrainSectorAbs = 0;
7602	break;
7603	}
7604	pExtent->uGrainSectorAbs = uSectorExtentAbs;
7605	pExtent->uGrain = uSectorExtentRel / pExtent->cSectorsPerGrain;
7606	Assert(uLBA == uSectorExtentRel);
7607	}
7608	vdIfIoIntIoCtxCopyTo(pImage->pIfIo, pIoCtx,
7609	(uint8_t *)pExtent->pvGrain
7610	+ VMDK_SECTOR2BYTE(uSectorInGrain),
7611	cbToRead);
7612	}
7613	else
7614	rc = vdIfIoIntFileReadUser(pImage->pIfIo, pExtent->pFile->pStorage,
7615	VMDK_SECTOR2BYTE(uSectorExtentAbs),
7616	pIoCtx, cbToRead);
7617	}
7618	break;
7619	}
7620	case VMDKETYPE_VMFS:
7621	case VMDKETYPE_FLAT:
7622	rc = vdIfIoIntFileReadUser(pImage->pIfIo, pExtent->pFile->pStorage,
7623	VMDK_SECTOR2BYTE(uSectorExtentRel),
7624	pIoCtx, cbToRead);
7625	break;
7626	case VMDKETYPE_ZERO:
7627	{
7628	size_t cbSet;
7629
7630	cbSet = vdIfIoIntIoCtxSet(pImage->pIfIo, pIoCtx, 0, cbToRead);
7631	Assert(cbSet == cbToRead);
7632	break;
7633	}
7634	}
7635	if (pcbActuallyRead)
7636	*pcbActuallyRead = cbToRead;
7637	}
7638	else if (RT_SUCCESS(rc))
7639	rc = VERR_VD_VMDK_INVALID_STATE;
7640
7641	LogFlowFunc(("returns %Rrc\n", rc));
7642	return rc;
7643	}
7644
7645	/** @copydoc VDIMAGEBACKEND::pfnWrite */
7646	static DECLCALLBACK(int) vmdkWrite(void *pBackendData, uint64_t uOffset, size_t cbToWrite,
7647	PVDIOCTX pIoCtx, size_t pcbWriteProcess, size_t pcbPreRead,
7648	size_t *pcbPostRead, unsigned fWrite)
7649	{
7650	LogFlowFunc(("pBackendData=%#p uOffset=%llu pIoCtx=%#p cbToWrite=%zu pcbWriteProcess=%#p pcbPreRead=%#p pcbPostRead=%#p\n",
7651	pBackendData, uOffset, pIoCtx, cbToWrite, pcbWriteProcess, pcbPreRead, pcbPostRead));
7652	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7653	int rc;
7654
7655	AssertPtr(pImage);
7656	Assert(uOffset % 512 == 0);
7657	Assert(cbToWrite % 512 == 0);
7658	AssertPtrReturn(pIoCtx, VERR_INVALID_POINTER);
7659	AssertReturn(cbToWrite, VERR_INVALID_PARAMETER);
7660
7661	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7662	{
7663	PVMDKEXTENT pExtent;
7664	uint64_t uSectorExtentRel;
7665	uint64_t uSectorExtentAbs;
7666
7667	/* No size check here, will do that later when the extent is located.
7668	* There are sparse images out there which according to the spec are
7669	* invalid, because the total size is not a multiple of the grain size.
7670	* Also for sparse images which are stitched together in odd ways (not at
7671	* grain boundaries, and with the nominal size not being a multiple of the
7672	* grain size), this would prevent writing to the last grain. */
7673
7674	rc = vmdkFindExtent(pImage, VMDK_BYTE2SECTOR(uOffset),
7675	&pExtent, &uSectorExtentRel);
7676	if (RT_SUCCESS(rc))
7677	{
7678	if ( pExtent->enmAccess != VMDKACCESS_READWRITE
7679	&& ( !(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7680	&& !pImage->pExtents[0].uAppendPosition
7681	&& pExtent->enmAccess != VMDKACCESS_READONLY))
7682	rc = VERR_VD_VMDK_INVALID_STATE;
7683	else
7684	{
7685	/* Handle the write according to the current extent type. */
7686	switch (pExtent->enmType)
7687	{
7688	case VMDKETYPE_HOSTED_SPARSE:
7689	rc = vmdkGetSector(pImage, pIoCtx, pExtent, uSectorExtentRel, &uSectorExtentAbs);
7690	if (RT_SUCCESS(rc))
7691	{
7692	if ( pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED
7693	&& uSectorExtentRel < (uint64_t)pExtent->uLastGrainAccess * pExtent->cSectorsPerGrain)
7694	rc = VERR_VD_VMDK_INVALID_WRITE;
7695	else
7696	{
7697	/* Clip write range to at most the rest of the grain. */
7698	cbToWrite = RT_MIN(cbToWrite,
7699	VMDK_SECTOR2BYTE( pExtent->cSectorsPerGrain
7700	- uSectorExtentRel % pExtent->cSectorsPerGrain));
7701	if (uSectorExtentAbs == 0)
7702	{
7703	if (!(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7704	{
7705	if (cbToWrite == VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
7706	{
7707	/* Full block write to a previously unallocated block.
7708	* Check if the caller wants to avoid the automatic alloc. */
7709	if (!(fWrite & VD_WRITE_NO_ALLOC))
7710	{
7711	/* Allocate GT and find out where to store the grain. */
7712	rc = vmdkAllocGrain(pImage, pExtent, pIoCtx,
7713	uSectorExtentRel, cbToWrite);
7714	}
7715	else
7716	rc = VERR_VD_BLOCK_FREE;
7717	*pcbPreRead = 0;
7718	*pcbPostRead = 0;
7719	}
7720	else
7721	{
7722	/* Clip write range to remain in this extent. */
7723	cbToWrite = RT_MIN(cbToWrite,
7724	VMDK_SECTOR2BYTE( pExtent->uSectorOffset
7725	+ pExtent->cNominalSectors - uSectorExtentRel));
7726	*pcbPreRead = VMDK_SECTOR2BYTE(uSectorExtentRel % pExtent->cSectorsPerGrain);
7727	pcbPostRead = VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain) - cbToWrite - pcbPreRead;
7728	rc = VERR_VD_BLOCK_FREE;
7729	}
7730	}
7731	else
7732	rc = vmdkStreamAllocGrain(pImage, pExtent, uSectorExtentRel,
7733	pIoCtx, cbToWrite);
7734	}
7735	else
7736	{
7737	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7738	{
7739	/* A partial write to a streamOptimized image is simply
7740	* invalid. It requires rewriting already compressed data
7741	* which is somewhere between expensive and impossible. */
7742	rc = VERR_VD_VMDK_INVALID_STATE;
7743	pExtent->uGrainSectorAbs = 0;
7744	AssertRC(rc);
7745	}
7746	else
7747	{
7748	Assert(!(pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED));
7749	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
7750	VMDK_SECTOR2BYTE(uSectorExtentAbs),
7751	pIoCtx, cbToWrite, NULL, NULL);
7752	}
7753	}
7754	}
7755	}
7756	break;
7757	case VMDKETYPE_VMFS:
7758	case VMDKETYPE_FLAT:
7759	/* Clip write range to remain in this extent. */
7760	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
7761	rc = vdIfIoIntFileWriteUser(pImage->pIfIo, pExtent->pFile->pStorage,
7762	VMDK_SECTOR2BYTE(uSectorExtentRel),
7763	pIoCtx, cbToWrite, NULL, NULL);
7764	break;
7765	case VMDKETYPE_ZERO:
7766	/* Clip write range to remain in this extent. */
7767	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
7768	break;
7769	}
7770	}
7771
7772	if (pcbWriteProcess)
7773	*pcbWriteProcess = cbToWrite;
7774	}
7775	}
7776	else
7777	rc = VERR_VD_IMAGE_READ_ONLY;
7778
7779	LogFlowFunc(("returns %Rrc\n", rc));
7780	return rc;
7781	}
7782
7783	/** @copydoc VDIMAGEBACKEND::pfnFlush */
7784	static DECLCALLBACK(int) vmdkFlush(void *pBackendData, PVDIOCTX pIoCtx)
7785	{
7786	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7787
7788	return vmdkFlushImage(pImage, pIoCtx);
7789	}
7790
7791	/** @copydoc VDIMAGEBACKEND::pfnGetVersion */
7792	static DECLCALLBACK(unsigned) vmdkGetVersion(void *pBackendData)
7793	{
7794	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
7795	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7796
7797	AssertPtrReturn(pImage, 0);
7798
7799	return VMDK_IMAGE_VERSION;
7800	}
7801
7802	/** @copydoc VDIMAGEBACKEND::pfnGetFileSize */
7803	static DECLCALLBACK(uint64_t) vmdkGetFileSize(void *pBackendData)
7804	{
7805	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
7806	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7807	uint64_t cb = 0;
7808
7809	AssertPtrReturn(pImage, 0);
7810
7811	if (pImage->pFile != NULL)
7812	{
7813	uint64_t cbFile;
7814	int rc = vdIfIoIntFileGetSize(pImage->pIfIo, pImage->pFile->pStorage, &cbFile);
7815	if (RT_SUCCESS(rc))
7816	cb += cbFile;
7817	}
7818	for (unsigned i = 0; i < pImage->cExtents; i++)
7819	{
7820	if (pImage->pExtents[i].pFile != NULL)
7821	{
7822	uint64_t cbFile;
7823	int rc = vdIfIoIntFileGetSize(pImage->pIfIo, pImage->pExtents[i].pFile->pStorage, &cbFile);
7824	if (RT_SUCCESS(rc))
7825	cb += cbFile;
7826	}
7827	}
7828
7829	LogFlowFunc(("returns %lld\n", cb));
7830	return cb;
7831	}
7832
7833	/** @copydoc VDIMAGEBACKEND::pfnGetPCHSGeometry */
7834	static DECLCALLBACK(int) vmdkGetPCHSGeometry(void *pBackendData, PVDGEOMETRY pPCHSGeometry)
7835	{
7836	LogFlowFunc(("pBackendData=%#p pPCHSGeometry=%#p\n", pBackendData, pPCHSGeometry));
7837	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7838	int rc = VINF_SUCCESS;
7839
7840	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7841
7842	if (pImage->PCHSGeometry.cCylinders)
7843	*pPCHSGeometry = pImage->PCHSGeometry;
7844	else
7845	rc = VERR_VD_GEOMETRY_NOT_SET;
7846
7847	LogFlowFunc(("returns %Rrc (PCHS=%u/%u/%u)\n", rc, pPCHSGeometry->cCylinders, pPCHSGeometry->cHeads, pPCHSGeometry->cSectors));
7848	return rc;
7849	}
7850
7851	/** @copydoc VDIMAGEBACKEND::pfnSetPCHSGeometry */
7852	static DECLCALLBACK(int) vmdkSetPCHSGeometry(void *pBackendData, PCVDGEOMETRY pPCHSGeometry)
7853	{
7854	LogFlowFunc(("pBackendData=%#p pPCHSGeometry=%#p PCHS=%u/%u/%u\n",
7855	pBackendData, pPCHSGeometry, pPCHSGeometry->cCylinders, pPCHSGeometry->cHeads, pPCHSGeometry->cSectors));
7856	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7857	int rc = VINF_SUCCESS;
7858
7859	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7860
7861	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7862	{
7863	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7864	{
7865	rc = vmdkDescSetPCHSGeometry(pImage, pPCHSGeometry);
7866	if (RT_SUCCESS(rc))
7867	pImage->PCHSGeometry = *pPCHSGeometry;
7868	}
7869	else
7870	rc = VERR_NOT_SUPPORTED;
7871	}
7872	else
7873	rc = VERR_VD_IMAGE_READ_ONLY;
7874
7875	LogFlowFunc(("returns %Rrc\n", rc));
7876	return rc;
7877	}
7878
7879	/** @copydoc VDIMAGEBACKEND::pfnGetLCHSGeometry */
7880	static DECLCALLBACK(int) vmdkGetLCHSGeometry(void *pBackendData, PVDGEOMETRY pLCHSGeometry)
7881	{
7882	LogFlowFunc(("pBackendData=%#p pLCHSGeometry=%#p\n", pBackendData, pLCHSGeometry));
7883	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7884	int rc = VINF_SUCCESS;
7885
7886	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7887
7888	if (pImage->LCHSGeometry.cCylinders)
7889	*pLCHSGeometry = pImage->LCHSGeometry;
7890	else
7891	rc = VERR_VD_GEOMETRY_NOT_SET;
7892
7893	LogFlowFunc(("returns %Rrc (LCHS=%u/%u/%u)\n", rc, pLCHSGeometry->cCylinders, pLCHSGeometry->cHeads, pLCHSGeometry->cSectors));
7894	return rc;
7895	}
7896
7897	/** @copydoc VDIMAGEBACKEND::pfnSetLCHSGeometry */
7898	static DECLCALLBACK(int) vmdkSetLCHSGeometry(void *pBackendData, PCVDGEOMETRY pLCHSGeometry)
7899	{
7900	LogFlowFunc(("pBackendData=%#p pLCHSGeometry=%#p LCHS=%u/%u/%u\n",
7901	pBackendData, pLCHSGeometry, pLCHSGeometry->cCylinders, pLCHSGeometry->cHeads, pLCHSGeometry->cSectors));
7902	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7903	int rc = VINF_SUCCESS;
7904
7905	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
7906
7907	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
7908	{
7909	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
7910	{
7911	rc = vmdkDescSetLCHSGeometry(pImage, pLCHSGeometry);
7912	if (RT_SUCCESS(rc))
7913	pImage->LCHSGeometry = *pLCHSGeometry;
7914	}
7915	else
7916	rc = VERR_NOT_SUPPORTED;
7917	}
7918	else
7919	rc = VERR_VD_IMAGE_READ_ONLY;
7920
7921	LogFlowFunc(("returns %Rrc\n", rc));
7922	return rc;
7923	}
7924
7925	/** @copydoc VDIMAGEBACKEND::pfnQueryRegions */
7926	static DECLCALLBACK(int) vmdkQueryRegions(void pBackendData, PCVDREGIONLIST ppRegionList)
7927	{
7928	LogFlowFunc(("pBackendData=%#p ppRegionList=%#p\n", pBackendData, ppRegionList));
7929	PVMDKIMAGE pThis = (PVMDKIMAGE)pBackendData;
7930
7931	AssertPtrReturn(pThis, VERR_VD_NOT_OPENED);
7932
7933	*ppRegionList = &pThis->RegionList;
7934	LogFlowFunc(("returns %Rrc\n", VINF_SUCCESS));
7935	return VINF_SUCCESS;
7936	}
7937
7938	/** @copydoc VDIMAGEBACKEND::pfnRegionListRelease */
7939	static DECLCALLBACK(void) vmdkRegionListRelease(void *pBackendData, PCVDREGIONLIST pRegionList)
7940	{
7941	RT_NOREF1(pRegionList);
7942	LogFlowFunc(("pBackendData=%#p pRegionList=%#p\n", pBackendData, pRegionList));
7943	PVMDKIMAGE pThis = (PVMDKIMAGE)pBackendData;
7944	AssertPtr(pThis); RT_NOREF(pThis);
7945
7946	/* Nothing to do here. */
7947	}
7948
7949	/** @copydoc VDIMAGEBACKEND::pfnGetImageFlags */
7950	static DECLCALLBACK(unsigned) vmdkGetImageFlags(void *pBackendData)
7951	{
7952	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
7953	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7954
7955	AssertPtrReturn(pImage, 0);
7956
7957	LogFlowFunc(("returns %#x\n", pImage->uImageFlags));
7958	return pImage->uImageFlags;
7959	}
7960
7961	/** @copydoc VDIMAGEBACKEND::pfnGetOpenFlags */
7962	static DECLCALLBACK(unsigned) vmdkGetOpenFlags(void *pBackendData)
7963	{
7964	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
7965	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7966
7967	AssertPtrReturn(pImage, 0);
7968
7969	LogFlowFunc(("returns %#x\n", pImage->uOpenFlags));
7970	return pImage->uOpenFlags;
7971	}
7972
7973	/** @copydoc VDIMAGEBACKEND::pfnSetOpenFlags */
7974	static DECLCALLBACK(int) vmdkSetOpenFlags(void *pBackendData, unsigned uOpenFlags)
7975	{
7976	LogFlowFunc(("pBackendData=%#p uOpenFlags=%#x\n", pBackendData, uOpenFlags));
7977	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
7978	int rc;
7979
7980	/* Image must be opened and the new flags must be valid. */
7981	if (!pImage \|\| (uOpenFlags & ~( VD_OPEN_FLAGS_READONLY \| VD_OPEN_FLAGS_INFO
7982	\| VD_OPEN_FLAGS_ASYNC_IO \| VD_OPEN_FLAGS_SHAREABLE
7983	\| VD_OPEN_FLAGS_SEQUENTIAL \| VD_OPEN_FLAGS_SKIP_CONSISTENCY_CHECKS)))
7984	rc = VERR_INVALID_PARAMETER;
7985	else
7986	{
7987	/* StreamOptimized images need special treatment: reopen is prohibited. */
7988	if (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED)
7989	{
7990	if (pImage->uOpenFlags == uOpenFlags)
7991	rc = VINF_SUCCESS;
7992	else
7993	rc = VERR_INVALID_PARAMETER;
7994	}
7995	else
7996	{
7997	/* Implement this operation via reopening the image. */
7998	vmdkFreeImage(pImage, false, true /fFlush/);
7999	rc = vmdkOpenImage(pImage, uOpenFlags);
8000	}
8001	}
8002
8003	LogFlowFunc(("returns %Rrc\n", rc));
8004	return rc;
8005	}
8006
8007	/** @copydoc VDIMAGEBACKEND::pfnGetComment */
8008	static DECLCALLBACK(int) vmdkGetComment(void pBackendData, char pszComment, size_t cbComment)
8009	{
8010	LogFlowFunc(("pBackendData=%#p pszComment=%#p cbComment=%zu\n", pBackendData, pszComment, cbComment));
8011	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8012
8013	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8014
8015	char *pszCommentEncoded = NULL;
8016	int rc = vmdkDescDDBGetStr(pImage, &pImage->Descriptor,
8017	"ddb.comment", &pszCommentEncoded);
8018	if (rc == VERR_VD_VMDK_VALUE_NOT_FOUND)
8019	{
8020	pszCommentEncoded = NULL;
8021	rc = VINF_SUCCESS;
8022	}
8023
8024	if (RT_SUCCESS(rc))
8025	{
8026	if (pszComment && pszCommentEncoded)
8027	rc = vmdkDecodeString(pszCommentEncoded, pszComment, cbComment);
8028	else if (pszComment)
8029	*pszComment = '\0';
8030
8031	if (pszCommentEncoded)
8032	RTMemTmpFree(pszCommentEncoded);
8033	}
8034
8035	LogFlowFunc(("returns %Rrc comment='%s'\n", rc, pszComment));
8036	return rc;
8037	}
8038
8039	/** @copydoc VDIMAGEBACKEND::pfnSetComment */
8040	static DECLCALLBACK(int) vmdkSetComment(void pBackendData, const char pszComment)
8041	{
8042	LogFlowFunc(("pBackendData=%#p pszComment=\"%s\"\n", pBackendData, pszComment));
8043	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8044	int rc;
8045
8046	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8047
8048	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8049	{
8050	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8051	rc = vmdkSetImageComment(pImage, pszComment);
8052	else
8053	rc = VERR_NOT_SUPPORTED;
8054	}
8055	else
8056	rc = VERR_VD_IMAGE_READ_ONLY;
8057
8058	LogFlowFunc(("returns %Rrc\n", rc));
8059	return rc;
8060	}
8061
8062	/** @copydoc VDIMAGEBACKEND::pfnGetUuid */
8063	static DECLCALLBACK(int) vmdkGetUuid(void *pBackendData, PRTUUID pUuid)
8064	{
8065	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
8066	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8067
8068	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8069
8070	*pUuid = pImage->ImageUuid;
8071
8072	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
8073	return VINF_SUCCESS;
8074	}
8075
8076	/** @copydoc VDIMAGEBACKEND::pfnSetUuid */
8077	static DECLCALLBACK(int) vmdkSetUuid(void *pBackendData, PCRTUUID pUuid)
8078	{
8079	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
8080	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8081	int rc = VINF_SUCCESS;
8082
8083	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8084
8085	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8086	{
8087	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8088	{
8089	pImage->ImageUuid = *pUuid;
8090	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
8091	VMDK_DDB_IMAGE_UUID, pUuid);
8092	if (RT_FAILURE(rc))
8093	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
8094	N_("VMDK: error storing image UUID in descriptor in '%s'"), pImage->pszFilename);
8095	}
8096	else
8097	rc = VERR_NOT_SUPPORTED;
8098	}
8099	else
8100	rc = VERR_VD_IMAGE_READ_ONLY;
8101
8102	LogFlowFunc(("returns %Rrc\n", rc));
8103	return rc;
8104	}
8105
8106	/** @copydoc VDIMAGEBACKEND::pfnGetModificationUuid */
8107	static DECLCALLBACK(int) vmdkGetModificationUuid(void *pBackendData, PRTUUID pUuid)
8108	{
8109	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
8110	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8111
8112	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8113
8114	*pUuid = pImage->ModificationUuid;
8115
8116	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
8117	return VINF_SUCCESS;
8118	}
8119
8120	/** @copydoc VDIMAGEBACKEND::pfnSetModificationUuid */
8121	static DECLCALLBACK(int) vmdkSetModificationUuid(void *pBackendData, PCRTUUID pUuid)
8122	{
8123	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
8124	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8125	int rc = VINF_SUCCESS;
8126
8127	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8128
8129	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8130	{
8131	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8132	{
8133	/* Only touch the modification uuid if it changed. */
8134	if (RTUuidCompare(&pImage->ModificationUuid, pUuid))
8135	{
8136	pImage->ModificationUuid = *pUuid;
8137	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
8138	VMDK_DDB_MODIFICATION_UUID, pUuid);
8139	if (RT_FAILURE(rc))
8140	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing modification UUID in descriptor in '%s'"), pImage->pszFilename);
8141	}
8142	}
8143	else
8144	rc = VERR_NOT_SUPPORTED;
8145	}
8146	else
8147	rc = VERR_VD_IMAGE_READ_ONLY;
8148
8149	LogFlowFunc(("returns %Rrc\n", rc));
8150	return rc;
8151	}
8152
8153	/** @copydoc VDIMAGEBACKEND::pfnGetParentUuid */
8154	static DECLCALLBACK(int) vmdkGetParentUuid(void *pBackendData, PRTUUID pUuid)
8155	{
8156	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
8157	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8158
8159	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8160
8161	*pUuid = pImage->ParentUuid;
8162
8163	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
8164	return VINF_SUCCESS;
8165	}
8166
8167	/** @copydoc VDIMAGEBACKEND::pfnSetParentUuid */
8168	static DECLCALLBACK(int) vmdkSetParentUuid(void *pBackendData, PCRTUUID pUuid)
8169	{
8170	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
8171	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8172	int rc = VINF_SUCCESS;
8173
8174	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8175
8176	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8177	{
8178	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8179	{
8180	pImage->ParentUuid = *pUuid;
8181	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
8182	VMDK_DDB_PARENT_UUID, pUuid);
8183	if (RT_FAILURE(rc))
8184	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS,
8185	N_("VMDK: error storing parent image UUID in descriptor in '%s'"), pImage->pszFilename);
8186	}
8187	else
8188	rc = VERR_NOT_SUPPORTED;
8189	}
8190	else
8191	rc = VERR_VD_IMAGE_READ_ONLY;
8192
8193	LogFlowFunc(("returns %Rrc\n", rc));
8194	return rc;
8195	}
8196
8197	/** @copydoc VDIMAGEBACKEND::pfnGetParentModificationUuid */
8198	static DECLCALLBACK(int) vmdkGetParentModificationUuid(void *pBackendData, PRTUUID pUuid)
8199	{
8200	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
8201	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8202
8203	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8204
8205	*pUuid = pImage->ParentModificationUuid;
8206
8207	LogFlowFunc(("returns %Rrc (%RTuuid)\n", VINF_SUCCESS, pUuid));
8208	return VINF_SUCCESS;
8209	}
8210
8211	/** @copydoc VDIMAGEBACKEND::pfnSetParentModificationUuid */
8212	static DECLCALLBACK(int) vmdkSetParentModificationUuid(void *pBackendData, PCRTUUID pUuid)
8213	{
8214	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
8215	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8216	int rc = VINF_SUCCESS;
8217
8218	AssertPtrReturn(pImage, VERR_VD_NOT_OPENED);
8219
8220	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
8221	{
8222	if (!(pImage->uOpenFlags & VD_VMDK_IMAGE_FLAGS_STREAM_OPTIMIZED))
8223	{
8224	pImage->ParentModificationUuid = *pUuid;
8225	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
8226	VMDK_DDB_PARENT_MODIFICATION_UUID, pUuid);
8227	if (RT_FAILURE(rc))
8228	rc = vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: error storing parent image UUID in descriptor in '%s'"), pImage->pszFilename);
8229	}
8230	else
8231	rc = VERR_NOT_SUPPORTED;
8232	}
8233	else
8234	rc = VERR_VD_IMAGE_READ_ONLY;
8235
8236	LogFlowFunc(("returns %Rrc\n", rc));
8237	return rc;
8238	}
8239
8240	/** @copydoc VDIMAGEBACKEND::pfnDump */
8241	static DECLCALLBACK(void) vmdkDump(void *pBackendData)
8242	{
8243	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8244
8245	AssertPtrReturnVoid(pImage);
8246	vdIfErrorMessage(pImage->pIfError, "Header: Geometry PCHS=%u/%u/%u LCHS=%u/%u/%u cbSector=%llu\n",
8247	pImage->PCHSGeometry.cCylinders, pImage->PCHSGeometry.cHeads, pImage->PCHSGeometry.cSectors,
8248	pImage->LCHSGeometry.cCylinders, pImage->LCHSGeometry.cHeads, pImage->LCHSGeometry.cSectors,
8249	VMDK_BYTE2SECTOR(pImage->cbSize));
8250	vdIfErrorMessage(pImage->pIfError, "Header: uuidCreation={%RTuuid}\n", &pImage->ImageUuid);
8251	vdIfErrorMessage(pImage->pIfError, "Header: uuidModification={%RTuuid}\n", &pImage->ModificationUuid);
8252	vdIfErrorMessage(pImage->pIfError, "Header: uuidParent={%RTuuid}\n", &pImage->ParentUuid);
8253	vdIfErrorMessage(pImage->pIfError, "Header: uuidParentModification={%RTuuid}\n", &pImage->ParentModificationUuid);
8254	}
8255
8256	static int vmdkRepaceExtentSize(PVMDKIMAGE pImage, unsigned line, uint64_t cSectorsOld,
8257	uint64_t cSectorsNew)
8258	{
8259	char * szOldExtentSectors = (char *)RTMemAlloc(UINT64_MAX_BUFF_SIZE);
8260	if (!szOldExtentSectors)
8261	return VERR_NO_MEMORY;
8262
8263	int cbWritten = RTStrPrintf2(szOldExtentSectors, UINT64_MAX_BUFF_SIZE, "%llu", cSectorsOld);
8264	if (cbWritten <= 0 \|\| cbWritten > UINT64_MAX_BUFF_SIZE)
8265	{
8266	RTMemFree(szOldExtentSectors);
8267	szOldExtentSectors = NULL;
8268
8269	return VERR_BUFFER_OVERFLOW;
8270	}
8271
8272	char * szNewExtentSectors = (char *)RTMemAlloc(UINT64_MAX_BUFF_SIZE);
8273	if (!szNewExtentSectors)
8274	return VERR_NO_MEMORY;
8275
8276	cbWritten = RTStrPrintf2(szNewExtentSectors, UINT64_MAX_BUFF_SIZE, "%llu", cSectorsNew);
8277	if (cbWritten <= 0 \|\| cbWritten > UINT64_MAX_BUFF_SIZE)
8278	{
8279	RTMemFree(szOldExtentSectors);
8280	szOldExtentSectors = NULL;
8281
8282	RTMemFree(szNewExtentSectors);
8283	szNewExtentSectors = NULL;
8284
8285	return VERR_BUFFER_OVERFLOW;
8286	}
8287
8288	char * szNewExtentLine = vmdkStrReplace(pImage->Descriptor.aLines[line],
8289	szOldExtentSectors,
8290	szNewExtentSectors);
8291
8292	RTMemFree(szOldExtentSectors);
8293	szOldExtentSectors = NULL;
8294
8295	RTMemFree(szNewExtentSectors);
8296	szNewExtentSectors = NULL;
8297
8298	if (!szNewExtentLine)
8299	return VERR_INVALID_PARAMETER;
8300
8301	pImage->Descriptor.aLines[line] = szNewExtentLine;
8302
8303	return VINF_SUCCESS;
8304	}
8305
8306	/** @copydoc VDIMAGEBACKEND::pfnResize */
8307	static DECLCALLBACK(int) vmdkResize(void *pBackendData, uint64_t cbSize,
8308	PCVDGEOMETRY pPCHSGeometry, PCVDGEOMETRY pLCHSGeometry,
8309	unsigned uPercentStart, unsigned uPercentSpan,
8310	PVDINTERFACE pVDIfsDisk, PVDINTERFACE pVDIfsImage,
8311	PVDINTERFACE pVDIfsOperation)
8312	{
8313	RT_NOREF5(uPercentStart, uPercentSpan, pVDIfsDisk, pVDIfsImage, pVDIfsOperation);
8314
8315	// Establish variables and objects needed
8316	int rc = VINF_SUCCESS;
8317	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
8318	unsigned uImageFlags = pImage->uImageFlags;
8319	PVMDKEXTENT pExtent = &pImage->pExtents[0];
8320
8321	uint64_t cSectorsNew = cbSize / VMDK_SECTOR_SIZE; /** < New number of sectors in the image after the resize */
8322	if (cbSize % VMDK_SECTOR_SIZE)
8323	cSectorsNew++;
8324
8325	uint64_t cSectorsOld = pImage->cbSize / VMDK_SECTOR_SIZE; /** < Number of sectors before the resize. Only for FLAT images. */
8326	if (pImage->cbSize % VMDK_SECTOR_SIZE)
8327	cSectorsOld++;
8328	unsigned cExtents = pImage->cExtents;
8329
8330	/* Check size is within min/max bounds. */
8331	if ( !(uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
8332	&& ( !cbSize
8333	\|\| (!(uImageFlags & VD_IMAGE_FLAGS_FIXED) && cbSize >= _1T * 256 - _64K)) )
8334	return VERR_VD_INVALID_SIZE;
8335
8336	/*
8337	* Making the image smaller is not supported at the moment.
8338	*/
8339	/** @todo implement making the image smaller, it is the responsibility of
8340	* the user to know what he's doing. */
8341	if (cbSize < pImage->cbSize)
8342	rc = VERR_VD_SHRINK_NOT_SUPPORTED;
8343	else if (cbSize > pImage->cbSize)
8344	{
8345	/**
8346	* monolithicFlat. FIXED flag and not split up into 2 GB parts.
8347	*/
8348	if ((uImageFlags & VD_IMAGE_FLAGS_FIXED) && !(uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G))
8349	{
8350	/** Required space in bytes for the extent after the resize. */
8351	uint64_t cbSectorSpaceNew = cSectorsNew * VMDK_SECTOR_SIZE;
8352	pExtent = &pImage->pExtents[0];
8353
8354	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage, cbSectorSpaceNew,
8355	0 /* fFlags */, NULL,
8356	uPercentStart, uPercentSpan);
8357	if (RT_FAILURE(rc))
8358	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
8359
8360	rc = vmdkRepaceExtentSize(pImage, pImage->Descriptor.uFirstExtent, cSectorsOld, cSectorsNew);
8361	if (RT_FAILURE(rc))
8362	return rc;
8363	}
8364
8365	/**
8366	* twoGbMaxExtentFlat. FIXED flag and SPLIT into 2 GB parts.
8367	*/
8368	if ((uImageFlags & VD_IMAGE_FLAGS_FIXED) && (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G))
8369	{
8370	/* Check to see how much space remains in last extent */
8371	bool fSpaceAvailible = false;
8372	uint64_t cLastExtentRemSectors = cSectorsOld % VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
8373	if (cLastExtentRemSectors)
8374	fSpaceAvailible = true;
8375
8376	uint64_t cSectorsNeeded = cSectorsNew - cSectorsOld;
8377	if (fSpaceAvailible && cSectorsNeeded + cLastExtentRemSectors <= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE))
8378	{
8379	pExtent = &pImage->pExtents[cExtents - 1];
8380	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage,
8381	VMDK_SECTOR2BYTE(cSectorsNeeded + cLastExtentRemSectors),
8382	0 /* fFlags */, NULL, uPercentStart, uPercentSpan);
8383	if (RT_FAILURE(rc))
8384	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
8385
8386	rc = vmdkRepaceExtentSize(pImage, pImage->Descriptor.uFirstExtent + cExtents - 1,
8387	pExtent->cNominalSectors, cSectorsNeeded + cLastExtentRemSectors);
8388	if (RT_FAILURE(rc))
8389	return rc;
8390	}
8391	else
8392	{
8393	if (fSpaceAvailible)
8394	{
8395	pExtent = &pImage->pExtents[cExtents - 1];
8396	rc = vdIfIoIntFileSetAllocationSize(pImage->pIfIo, pExtent->pFile->pStorage, VMDK_2G_SPLIT_SIZE,
8397	0 /* fFlags */, NULL,
8398	uPercentStart, uPercentSpan);
8399	if (RT_FAILURE(rc))
8400	return vdIfError(pImage->pIfError, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
8401
8402	cSectorsNeeded = cSectorsNeeded - VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE) + cLastExtentRemSectors;
8403
8404	rc = vmdkRepaceExtentSize(pImage, pImage->Descriptor.uFirstExtent + cExtents - 1,
8405	pExtent->cNominalSectors, VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE));
8406	if (RT_FAILURE(rc))
8407	return rc;
8408	}
8409
8410	unsigned cNewExtents = VMDK_SECTOR2BYTE(cSectorsNeeded) / VMDK_2G_SPLIT_SIZE;
8411	if (cNewExtents % VMDK_2G_SPLIT_SIZE \|\| cNewExtents < VMDK_2G_SPLIT_SIZE)
8412	cNewExtents++;
8413
8414	for (unsigned i = cExtents;
8415	i < cExtents + cNewExtents && cSectorsNeeded >= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
8416	i++)
8417	{
8418	rc = vmdkAddFileBackedExtent(pImage, VMDK_2G_SPLIT_SIZE);
8419	if (RT_FAILURE(rc))
8420	return rc;
8421
8422	pExtent = &pImage->pExtents[i];
8423
8424	pExtent->cSectors = VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
8425	cSectorsNeeded -= VMDK_BYTE2SECTOR(VMDK_2G_SPLIT_SIZE);
8426	}
8427
8428	if (cSectorsNeeded)
8429	{
8430	rc = vmdkAddFileBackedExtent(pImage, VMDK_SECTOR2BYTE(cSectorsNeeded));
8431	if (RT_FAILURE(rc))
8432	return rc;
8433	}
8434	}
8435	}
8436
8437	/* Successful resize. Update metadata */
8438	if (RT_SUCCESS(rc))
8439	{
8440	/* Update size and new block count. */
8441	pImage->cbSize = cbSize;
8442	/** @todo r=jack: update cExtents if needed */
8443	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
8444
8445	/* Update geometry. */
8446	pImage->PCHSGeometry = *pPCHSGeometry;
8447	pImage->LCHSGeometry = *pLCHSGeometry;
8448	}
8449
8450	/* Update header information in base image file. */
8451	rc = vmdkWriteDescriptor(pImage, NULL);
8452
8453	if (RT_FAILURE(rc))
8454	return rc;
8455
8456	rc = vmdkFlushImage(pImage, NULL);
8457
8458	if (RT_FAILURE(rc))
8459	return rc;
8460	}
8461	/* Same size doesn't change the image at all. */
8462
8463	LogFlowFunc(("returns %Rrc\n", rc));
8464	return rc;
8465	}
8466
8467
8468	const VDIMAGEBACKEND g_VmdkBackend =
8469	{
8470	/* u32Version */
8471	VD_IMGBACKEND_VERSION,
8472	/* pszBackendName */
8473	"VMDK",
8474	/* uBackendCaps */
8475	VD_CAP_UUID \| VD_CAP_CREATE_FIXED \| VD_CAP_CREATE_DYNAMIC
8476	\| VD_CAP_CREATE_SPLIT_2G \| VD_CAP_DIFF \| VD_CAP_FILE \| VD_CAP_ASYNC
8477	\| VD_CAP_VFS \| VD_CAP_PREFERRED,
8478	/* paFileExtensions */
8479	s_aVmdkFileExtensions,
8480	/* paConfigInfo */
8481	s_aVmdkConfigInfo,
8482	/* pfnProbe */
8483	vmdkProbe,
8484	/* pfnOpen */
8485	vmdkOpen,
8486	/* pfnCreate */
8487	vmdkCreate,
8488	/* pfnRename */
8489	vmdkRename,
8490	/* pfnClose */
8491	vmdkClose,
8492	/* pfnRead */
8493	vmdkRead,
8494	/* pfnWrite */
8495	vmdkWrite,
8496	/* pfnFlush */
8497	vmdkFlush,
8498	/* pfnDiscard */
8499	NULL,
8500	/* pfnGetVersion */
8501	vmdkGetVersion,
8502	/* pfnGetFileSize */
8503	vmdkGetFileSize,
8504	/* pfnGetPCHSGeometry */
8505	vmdkGetPCHSGeometry,
8506	/* pfnSetPCHSGeometry */
8507	vmdkSetPCHSGeometry,
8508	/* pfnGetLCHSGeometry */
8509	vmdkGetLCHSGeometry,
8510	/* pfnSetLCHSGeometry */
8511	vmdkSetLCHSGeometry,
8512	/* pfnQueryRegions */
8513	vmdkQueryRegions,
8514	/* pfnRegionListRelease */
8515	vmdkRegionListRelease,
8516	/* pfnGetImageFlags */
8517	vmdkGetImageFlags,
8518	/* pfnGetOpenFlags */
8519	vmdkGetOpenFlags,
8520	/* pfnSetOpenFlags */
8521	vmdkSetOpenFlags,
8522	/* pfnGetComment */
8523	vmdkGetComment,
8524	/* pfnSetComment */
8525	vmdkSetComment,
8526	/* pfnGetUuid */
8527	vmdkGetUuid,
8528	/* pfnSetUuid */
8529	vmdkSetUuid,
8530	/* pfnGetModificationUuid */
8531	vmdkGetModificationUuid,
8532	/* pfnSetModificationUuid */
8533	vmdkSetModificationUuid,
8534	/* pfnGetParentUuid */
8535	vmdkGetParentUuid,
8536	/* pfnSetParentUuid */
8537	vmdkSetParentUuid,
8538	/* pfnGetParentModificationUuid */
8539	vmdkGetParentModificationUuid,
8540	/* pfnSetParentModificationUuid */
8541	vmdkSetParentModificationUuid,
8542	/* pfnDump */
8543	vmdkDump,
8544	/* pfnGetTimestamp */
8545	NULL,
8546	/* pfnGetParentTimestamp */
8547	NULL,
8548	/* pfnSetParentTimestamp */
8549	NULL,
8550	/* pfnGetParentFilename */
8551	NULL,
8552	/* pfnSetParentFilename */
8553	NULL,
8554	/* pfnComposeLocation */
8555	genericFileComposeLocation,
8556	/* pfnComposeName */
8557	genericFileComposeName,
8558	/* pfnCompact */
8559	NULL,
8560	/* pfnResize */
8561	vmdkResize,
8562	/* pfnRepair */
8563	NULL,
8564	/* pfnTraverseMetadata */
8565	NULL,
8566	/* u32VersionEnd */
8567	VD_IMGBACKEND_VERSION
8568	};

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

source: vbox/trunk/src/VBox/Storage/VMDK.cpp@ 97839

Download in other formats: