VmdkHDDCore.cpp@ 12367

Last change on this file since 12367 was 12367, checked in by vboxsync, 16 years ago
Storage/VMDK: disable paranoid size check because there are vmdks out there which are accepted by vmware despite they're too short.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 188.9 KB

Line
1	/** $Id: VmdkHDDCore.cpp 12367 2008-09-10 16:21:34Z vboxsync $ */
2	/** @file
3	* VMDK Disk image, Core Code.
4	*/
5
6	/*
7	* Copyright (C) 2006-2007 Sun Microsystems, Inc.
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*
17	* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa
18	* Clara, CA 95054 USA or visit http://www.sun.com if you need
19	* additional information or have any questions.
20	*/
21
22	/*******************************************************************************
23	* Header Files *
24	*******************************************************************************/
25	#define LOG_GROUP LOG_GROUP_VD_VMDK
26	#include "VBoxHDD-newInternal.h"
27	#include <VBox/err.h>
28
29	#include <VBox/log.h>
30	#include <iprt/assert.h>
31	#include <iprt/alloc.h>
32	#include <iprt/uuid.h>
33	#include <iprt/file.h>
34	#include <iprt/path.h>
35	#include <iprt/string.h>
36	#include <iprt/rand.h>
37
38
39	/*******************************************************************************
40	* Constants And Macros, Structures and Typedefs *
41	*******************************************************************************/
42
43	/** Maximum encoded string size (including NUL) we allow for VMDK images.
44	* Deliberately not set high to avoid running out of descriptor space. */
45	#define VMDK_ENCODED_COMMENT_MAX 1024
46
47	/** VMDK descriptor DDB entry for PCHS cylinders. */
48	#define VMDK_DDB_GEO_PCHS_CYLINDERS "ddb.geometry.cylinders"
49
50	/** VMDK descriptor DDB entry for PCHS heads. */
51	#define VMDK_DDB_GEO_PCHS_HEADS "ddb.geometry.heads"
52
53	/** VMDK descriptor DDB entry for PCHS sectors. */
54	#define VMDK_DDB_GEO_PCHS_SECTORS "ddb.geometry.sectors"
55
56	/** VMDK descriptor DDB entry for LCHS cylinders. */
57	#define VMDK_DDB_GEO_LCHS_CYLINDERS "ddb.geometry.biosCylinders"
58
59	/** VMDK descriptor DDB entry for LCHS heads. */
60	#define VMDK_DDB_GEO_LCHS_HEADS "ddb.geometry.biosHeads"
61
62	/** VMDK descriptor DDB entry for LCHS sectors. */
63	#define VMDK_DDB_GEO_LCHS_SECTORS "ddb.geometry.biosSectors"
64
65	/** VMDK descriptor DDB entry for image UUID. */
66	#define VMDK_DDB_IMAGE_UUID "ddb.uuid.image"
67
68	/** VMDK descriptor DDB entry for image modification UUID. */
69	#define VMDK_DDB_MODIFICATION_UUID "ddb.uuid.modification"
70
71	/** VMDK descriptor DDB entry for parent image UUID. */
72	#define VMDK_DDB_PARENT_UUID "ddb.uuid.parent"
73
74	/** VMDK descriptor DDB entry for parent image modification UUID. */
75	#define VMDK_DDB_PARENT_MODIFICATION_UUID "ddb.uuid.parentmodification"
76
77	/**
78	* Magic number for hosted images created by VMware Workstation 4, VMware
79	* Workstation 5, VMware Server or VMware Player. Not necessarily sparse.
80	*/
81	#define VMDK_SPARSE_MAGICNUMBER 0x564d444b /* 'V' 'M' 'D' 'K' */
82
83	/**
84	* VMDK hosted binary extent header. The "Sparse" is a total misnomer, as
85	* this header is also used for monolithic flat images.
86	*/
87	#pragma pack(1)
88	typedef struct SparseExtentHeader
89	{
90	uint32_t magicNumber;
91	uint32_t version;
92	uint32_t flags;
93	uint64_t capacity;
94	uint64_t grainSize;
95	uint64_t descriptorOffset;
96	uint64_t descriptorSize;
97	uint32_t numGTEsPerGT;
98	uint64_t rgdOffset;
99	uint64_t gdOffset;
100	uint64_t overHead;
101	bool uncleanShutdown;
102	char singleEndLineChar;
103	char nonEndLineChar;
104	char doubleEndLineChar1;
105	char doubleEndLineChar2;
106	uint8_t pad[435];
107	} SparseExtentHeader;
108	#pragma pack()
109
110	/** VMDK capacity for a single chunk when 2G splitting is turned on. Should be
111	* divisible by the default grain size (64K) */
112	#define VMDK_2G_SPLIT_SIZE (2047 * 1024 * 1024)
113
114
115	#ifdef VBOX_WITH_VMDK_ESX
116
117	/** @todo the ESX code is not tested, not used, and lacks error messages. */
118
119	/**
120	* Magic number for images created by VMware GSX Server 3 or ESX Server 3.
121	*/
122	#define VMDK_ESX_SPARSE_MAGICNUMBER 0x44574f43 /* 'C' 'O' 'W' 'D' */
123
124	#pragma pack(1)
125	typedef struct COWDisk_Header
126	{
127	uint32_t magicNumber;
128	uint32_t version;
129	uint32_t flags;
130	uint32_t numSectors;
131	uint32_t grainSize;
132	uint32_t gdOffset;
133	uint32_t numGDEntries;
134	uint32_t freeSector;
135	/* The spec incompletely documents quite a few further fields, but states
136	* that they are unused by the current format. Replace them by padding. */
137	char reserved1[1604];
138	uint32_t savedGeneration;
139	char reserved2[8];
140	uint32_t uncleanShutdown;
141	char padding[396];
142	} COWDisk_Header;
143	#pragma pack()
144	#endif /* VBOX_WITH_VMDK_ESX */
145
146
147	/** Convert sector number/size to byte offset/size. */
148	#define VMDK_SECTOR2BYTE(u) ((u) << 9)
149
150	/** Convert byte offset/size to sector number/size. */
151	#define VMDK_BYTE2SECTOR(u) ((u) >> 9)
152
153	/**
154	* VMDK extent type.
155	*/
156	typedef enum VMDKETYPE
157	{
158	/** Hosted sparse extent. */
159	VMDKETYPE_HOSTED_SPARSE = 1,
160	/** Flat extent. */
161	VMDKETYPE_FLAT,
162	/** Zero extent. */
163	VMDKETYPE_ZERO
164	#ifdef VBOX_WITH_VMDK_ESX
165	,
166	/** ESX sparse extent. */
167	VMDKETYPE_ESX_SPARSE
168	#endif /* VBOX_WITH_VMDK_ESX */
169	} VMDKETYPE, *PVMDKETYPE;
170
171	/**
172	* VMDK access type for a extent.
173	*/
174	typedef enum VMDKACCESS
175	{
176	/** No access allowed. */
177	VMDKACCESS_NOACCESS = 0,
178	/** Read-only access. */
179	VMDKACCESS_READONLY,
180	/** Read-write access. */
181	VMDKACCESS_READWRITE
182	} VMDKACCESS, *PVMDKACCESS;
183
184	/** Forward declaration for PVMDKIMAGE. */
185	typedef struct VMDKIMAGE *PVMDKIMAGE;
186
187	/**
188	* Extents files entry. Used for opening a particular file only once.
189	*/
190	typedef struct VMDKFILE
191	{
192	/** Pointer to filename. Local copy. */
193	const char *pszFilename;
194	/** File open flags for consistency checking. */
195	unsigned fOpen;
196	/** File handle. */
197	RTFILE File;
198	/** Handle for asnychronous access if requested.*/
199	void *pStorage;
200	/** Flag whether to use File or pStorage. */
201	bool fAsyncIO;
202	/** Reference counter. */
203	unsigned uReferences;
204	/** Flag whether the file should be deleted on last close. */
205	bool fDelete;
206	/** Pointer to the image we belong to. */
207	PVMDKIMAGE pImage;
208	/** Pointer to next file descriptor. */
209	struct VMDKFILE *pNext;
210	/** Pointer to the previous file descriptor. */
211	struct VMDKFILE *pPrev;
212	} VMDKFILE, *PVMDKFILE;
213
214	/**
215	* VMDK extent data structure.
216	*/
217	typedef struct VMDKEXTENT
218	{
219	/** File handle. */
220	PVMDKFILE pFile;
221	/** Base name of the image extent. */
222	const char *pszBasename;
223	/** Full name of the image extent. */
224	const char *pszFullname;
225	/** Number of sectors in this extent. */
226	uint64_t cSectors;
227	/** Number of sectors per block (grain in VMDK speak). */
228	uint64_t cSectorsPerGrain;
229	/** Starting sector number of descriptor. */
230	uint64_t uDescriptorSector;
231	/** Size of descriptor in sectors. */
232	uint64_t cDescriptorSectors;
233	/** Starting sector number of grain directory. */
234	uint64_t uSectorGD;
235	/** Starting sector number of redundant grain directory. */
236	uint64_t uSectorRGD;
237	/** Total number of metadata sectors. */
238	uint64_t cOverheadSectors;
239	/** Nominal size (i.e. as described by the descriptor) of this extent. */
240	uint64_t cNominalSectors;
241	/** Sector offset (i.e. as described by the descriptor) of this extent. */
242	uint64_t uSectorOffset;
243	/** Number of entries in a grain table. */
244	uint32_t cGTEntries;
245	/** Number of sectors reachable via a grain directory entry. */
246	uint32_t cSectorsPerGDE;
247	/** Number of entries in the grain directory. */
248	uint32_t cGDEntries;
249	/** Pointer to the next free sector. Legacy information. Do not use. */
250	uint32_t uFreeSector;
251	/** Number of this extent in the list of images. */
252	uint32_t uExtent;
253	/** Pointer to the descriptor (NULL if no descriptor in this extent). */
254	char *pDescData;
255	/** Pointer to the grain directory. */
256	uint32_t *pGD;
257	/** Pointer to the redundant grain directory. */
258	uint32_t *pRGD;
259	/** Type of this extent. */
260	VMDKETYPE enmType;
261	/** Access to this extent. */
262	VMDKACCESS enmAccess;
263	/** Flag whether this extent is marked as unclean. */
264	bool fUncleanShutdown;
265	/** Flag whether the metadata in the extent header needs to be updated. */
266	bool fMetaDirty;
267	/** Reference to the image in which this extent is used. Do not use this
268	* on a regular basis to avoid passing pImage references to functions
269	* explicitly. */
270	struct VMDKIMAGE *pImage;
271	} VMDKEXTENT, *PVMDKEXTENT;
272
273	/**
274	* Grain table cache size. Allocated per image.
275	*/
276	#define VMDK_GT_CACHE_SIZE 256
277
278	/**
279	* Grain table block size. Smaller than an actual grain table block to allow
280	* more grain table blocks to be cached without having to allocate excessive
281	* amounts of memory for the cache.
282	*/
283	#define VMDK_GT_CACHELINE_SIZE 128
284
285
286	/**
287	* Maximum number of lines in a descriptor file. Not worth the effort of
288	* making it variable. Descriptor files are generally very short (~20 lines).
289	*/
290	#define VMDK_DESCRIPTOR_LINES_MAX 100U
291
292	/**
293	* Parsed descriptor information. Allows easy access and update of the
294	* descriptor (whether separate file or not). Free form text files suck.
295	*/
296	typedef struct VMDKDESCRIPTOR
297	{
298	/** Line number of first entry of the disk descriptor. */
299	unsigned uFirstDesc;
300	/** Line number of first entry in the extent description. */
301	unsigned uFirstExtent;
302	/** Line number of first disk database entry. */
303	unsigned uFirstDDB;
304	/** Total number of lines. */
305	unsigned cLines;
306	/** Total amount of memory available for the descriptor. */
307	size_t cbDescAlloc;
308	/** Set if descriptor has been changed and not yet written to disk. */
309	bool fDirty;
310	/** Array of pointers to the data in the descriptor. */
311	char *aLines[VMDK_DESCRIPTOR_LINES_MAX];
312	/** Array of line indices pointing to the next non-comment line. */
313	unsigned aNextLines[VMDK_DESCRIPTOR_LINES_MAX];
314	} VMDKDESCRIPTOR, *PVMDKDESCRIPTOR;
315
316
317	/**
318	* Cache entry for translating extent/sector to a sector number in that
319	* extent.
320	*/
321	typedef struct VMDKGTCACHEENTRY
322	{
323	/** Extent number for which this entry is valid. */
324	uint32_t uExtent;
325	/** GT data block number. */
326	uint64_t uGTBlock;
327	/** Data part of the cache entry. */
328	uint32_t aGTData[VMDK_GT_CACHELINE_SIZE];
329	} VMDKGTCACHEENTRY, *PVMDKGTCACHEENTRY;
330
331	/**
332	* Cache data structure for blocks of grain table entries. For now this is a
333	* fixed size direct mapping cache, but this should be adapted to the size of
334	* the sparse image and maybe converted to a set-associative cache. The
335	* implementation below implements a write-through cache with write allocate.
336	*/
337	typedef struct VMDKGTCACHE
338	{
339	/** Cache entries. */
340	VMDKGTCACHEENTRY aGTCache[VMDK_GT_CACHE_SIZE];
341	/** Number of cache entries (currently unused). */
342	unsigned cEntries;
343	} VMDKGTCACHE, *PVMDKGTCACHE;
344
345	/**
346	* Complete VMDK image data structure. Mainly a collection of extents and a few
347	* extra global data fields.
348	*/
349	typedef struct VMDKIMAGE
350	{
351	/** Pointer to the image extents. */
352	PVMDKEXTENT pExtents;
353	/** Number of image extents. */
354	unsigned cExtents;
355	/** Pointer to the files list, for opening a file referenced multiple
356	* times only once (happens mainly with raw partition access). */
357	PVMDKFILE pFiles;
358
359	/** Base image name. */
360	const char *pszFilename;
361	/** Descriptor file if applicable. */
362	PVMDKFILE pFile;
363
364	/** Pointer to the per-disk VD interface list. */
365	PVDINTERFACE pVDIfsDisk;
366
367	/** Error interface. */
368	PVDINTERFACE pInterfaceError;
369	/** Error interface callbacks. */
370	PVDINTERFACEERROR pInterfaceErrorCallbacks;
371
372	/** Async I/O interface. */
373	PVDINTERFACE pInterfaceAsyncIO;
374	/** Async I/O interface callbacks. */
375	PVDINTERFACEASYNCIO pInterfaceAsyncIOCallbacks;
376	/**
377	* Pointer to an array of task handles for task submission.
378	* This is an optimization because the task number to submit is not known
379	* and allocating/freeing an array in the read/write functions every time
380	* is too expensive.
381	*/
382	void **apTask;
383	/** Entries available in the task handle array. */
384	unsigned cTask;
385
386	/** Open flags passed by VBoxHD layer. */
387	unsigned uOpenFlags;
388	/** Image type. */
389	VDIMAGETYPE enmImageType;
390	/** Image flags defined during creation or determined during open. */
391	unsigned uImageFlags;
392	/** Total size of the image. */
393	uint64_t cbSize;
394	/** Physical geometry of this image. */
395	PDMMEDIAGEOMETRY PCHSGeometry;
396	/** Logical geometry of this image. */
397	PDMMEDIAGEOMETRY LCHSGeometry;
398	/** Image UUID. */
399	RTUUID ImageUuid;
400	/** Image modification UUID. */
401	RTUUID ModificationUuid;
402	/** Parent image UUID. */
403	RTUUID ParentUuid;
404	/** Parent image modification UUID. */
405	RTUUID ParentModificationUuid;
406
407	/** Pointer to grain table cache, if this image contains sparse extents. */
408	PVMDKGTCACHE pGTCache;
409	/** Pointer to the descriptor (NULL if no separate descriptor file). */
410	char *pDescData;
411	/** Allocation size of the descriptor file. */
412	size_t cbDescAlloc;
413	/** Parsed descriptor file content. */
414	VMDKDESCRIPTOR Descriptor;
415	} VMDKIMAGE;
416
417	/*******************************************************************************
418	* Static Variables *
419	*******************************************************************************/
420
421	/** NULL-terminated array of supported file extensions. */
422	static const char *const s_apszVmdkFileExtensions[] =
423	{
424	"vmdk",
425	NULL
426	};
427
428	/*******************************************************************************
429	* Internal Functions *
430	*******************************************************************************/
431
432	static void vmdkFreeGrainDirectory(PVMDKEXTENT pExtent);
433
434	static void vmdkFreeExtentData(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
435	bool fDelete);
436
437	static int vmdkCreateExtents(PVMDKIMAGE pImage, unsigned cExtents);
438	static int vmdkFlushImage(PVMDKIMAGE pImage);
439	static int vmdkSetImageComment(PVMDKIMAGE pImage, const char *pszComment);
440	static void vmdkFreeImage(PVMDKIMAGE pImage, bool fDelete);
441
442
443	/**
444	* Internal: signal an error to the frontend.
445	*/
446	DECLINLINE(int) vmdkError(PVMDKIMAGE pImage, int rc, RT_SRC_POS_DECL,
447	const char *pszFormat, ...)
448	{
449	va_list va;
450	va_start(va, pszFormat);
451	if (pImage->pInterfaceError && pImage->pInterfaceErrorCallbacks)
452	pImage->pInterfaceErrorCallbacks->pfnError(pImage->pInterfaceError->pvUser, rc, RT_SRC_POS_ARGS,
453	pszFormat, va);
454	va_end(va);
455	return rc;
456	}
457
458	/**
459	* Internal: open a file (using a file descriptor cache to ensure each file
460	* is only opened once - anything else can cause locking problems).
461	*/
462	static int vmdkFileOpen(PVMDKIMAGE pImage, PVMDKFILE *ppVmdkFile,
463	const char *pszFilename, unsigned fOpen, bool fAsyncIO)
464	{
465	int rc = VINF_SUCCESS;
466	PVMDKFILE pVmdkFile;
467
468	for (pVmdkFile = pImage->pFiles;
469	pVmdkFile != NULL;
470	pVmdkFile = pVmdkFile->pNext)
471	{
472	if (!strcmp(pszFilename, pVmdkFile->pszFilename))
473	{
474	Assert(fOpen == pVmdkFile->fOpen);
475	pVmdkFile->uReferences++;
476
477	*ppVmdkFile = pVmdkFile;
478
479	return rc;
480	}
481	}
482
483	/* If we get here, there's no matching entry in the cache. */
484	pVmdkFile = (PVMDKFILE)RTMemAllocZ(sizeof(VMDKFILE));
485	if (!VALID_PTR(pVmdkFile))
486	{
487	*ppVmdkFile = NULL;
488	return VERR_NO_MEMORY;
489	}
490
491	pVmdkFile->pszFilename = RTStrDup(pszFilename);
492	if (!VALID_PTR(pVmdkFile->pszFilename))
493	{
494	RTMemFree(pVmdkFile);
495	*ppVmdkFile = NULL;
496	return VERR_NO_MEMORY;
497	}
498	pVmdkFile->fOpen = fOpen;
499	if ((pImage->uOpenFlags & VD_OPEN_FLAGS_ASYNC_IO) && (fAsyncIO))
500	{
501	rc = pImage->pInterfaceAsyncIOCallbacks->pfnOpen(pImage->pInterfaceAsyncIO->pvUser,
502	pszFilename,
503	pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY
504	? true
505	: false,
506	&pVmdkFile->pStorage);
507	pVmdkFile->fAsyncIO = true;
508	}
509	else
510	{
511	rc = RTFileOpen(&pVmdkFile->File, pszFilename, fOpen);
512	pVmdkFile->fAsyncIO = false;
513	}
514	if (RT_SUCCESS(rc))
515	{
516	pVmdkFile->uReferences = 1;
517	pVmdkFile->pImage = pImage;
518	pVmdkFile->pNext = pImage->pFiles;
519	if (pImage->pFiles)
520	pImage->pFiles->pPrev = pVmdkFile;
521	pImage->pFiles = pVmdkFile;
522	*ppVmdkFile = pVmdkFile;
523	}
524	else
525	{
526	RTStrFree((char )(void )pVmdkFile->pszFilename);
527	RTMemFree(pVmdkFile);
528	*ppVmdkFile = NULL;
529	}
530
531	return rc;
532	}
533
534	/**
535	* Internal: close a file, updating the file descriptor cache.
536	*/
537	static int vmdkFileClose(PVMDKIMAGE pImage, PVMDKFILE *ppVmdkFile, bool fDelete)
538	{
539	int rc = VINF_SUCCESS;
540	PVMDKFILE pVmdkFile = *ppVmdkFile;
541
542	Assert(VALID_PTR(pVmdkFile));
543
544	pVmdkFile->fDelete \|= fDelete;
545	Assert(pVmdkFile->uReferences);
546	pVmdkFile->uReferences--;
547	if (pVmdkFile->uReferences == 0)
548	{
549	PVMDKFILE pPrev;
550	PVMDKFILE pNext;
551
552	/* Unchain the element from the list. */
553	pPrev = pVmdkFile->pPrev;
554	pNext = pVmdkFile->pNext;
555
556	if (pNext)
557	pNext->pPrev = pPrev;
558	if (pPrev)
559	pPrev->pNext = pNext;
560	else
561	pImage->pFiles = pNext;
562
563	if (pVmdkFile->fAsyncIO)
564	{
565	rc = pImage->pInterfaceAsyncIOCallbacks->pfnClose(pImage->pInterfaceAsyncIO->pvUser,
566	pVmdkFile->pStorage);
567	}
568	else
569	{
570	rc = RTFileClose(pVmdkFile->File);
571	}
572	if (RT_SUCCESS(rc) && pVmdkFile->fDelete)
573	rc = RTFileDelete(pVmdkFile->pszFilename);
574	RTStrFree((char )(void )pVmdkFile->pszFilename);
575	RTMemFree(pVmdkFile);
576	}
577
578	*ppVmdkFile = NULL;
579	return rc;
580	}
581
582	/**
583	* Internal: read from a file distinguishing between async and normal operation
584	*/
585	DECLINLINE(int) vmdkFileReadAt(PVMDKFILE pVmdkFile,
586	uint64_t uOffset, void *pvBuf,
587	size_t cbToRead, size_t *pcbRead)
588	{
589	PVMDKIMAGE pImage = pVmdkFile->pImage;
590
591	if (pVmdkFile->fAsyncIO)
592	return pImage->pInterfaceAsyncIOCallbacks->pfnRead(pImage->pInterfaceAsyncIO->pvUser,
593	pVmdkFile->pStorage, uOffset,
594	cbToRead, pvBuf, pcbRead);
595	else
596	return RTFileReadAt(pVmdkFile->File, uOffset, pvBuf, cbToRead, pcbRead);
597	}
598
599	/**
600	* Internal: write to a file distinguishing between async and normal operation
601	*/
602	DECLINLINE(int) vmdkFileWriteAt(PVMDKFILE pVmdkFile,
603	uint64_t uOffset, const void *pvBuf,
604	size_t cbToWrite, size_t *pcbWritten)
605	{
606	PVMDKIMAGE pImage = pVmdkFile->pImage;
607
608	if (pVmdkFile->fAsyncIO)
609	return pImage->pInterfaceAsyncIOCallbacks->pfnWrite(pImage->pInterfaceAsyncIO->pvUser,
610	pVmdkFile->pStorage, uOffset,
611	cbToWrite, pvBuf, pcbWritten);
612	else
613	return RTFileWriteAt(pVmdkFile->File, uOffset, pvBuf, cbToWrite, pcbWritten);
614	}
615
616	/**
617	* Internal: get the size of a file distinguishing beween async and normal operation
618	*/
619	DECLINLINE(int) vmdkFileGetSize(PVMDKFILE pVmdkFile, uint64_t *pcbSize)
620	{
621	if (pVmdkFile->fAsyncIO)
622	{
623	AssertMsgFailed(("TODO\n"));
624	return 0;
625	}
626	else
627	return RTFileGetSize(pVmdkFile->File, pcbSize);
628	}
629
630	/**
631	* Internal: set the size of a file distinguishing beween async and normal operation
632	*/
633	DECLINLINE(int) vmdkFileSetSize(PVMDKFILE pVmdkFile, uint64_t cbSize)
634	{
635	if (pVmdkFile->fAsyncIO)
636	{
637	AssertMsgFailed(("TODO\n"));
638	return VERR_NOT_SUPPORTED;
639	}
640	else
641	return RTFileSetSize(pVmdkFile->File, cbSize);
642	}
643
644	/**
645	* Internal: flush a file distinguishing between async and normal operation
646	*/
647	DECLINLINE(int) vmdkFileFlush(PVMDKFILE pVmdkFile)
648	{
649	PVMDKIMAGE pImage = pVmdkFile->pImage;
650
651	if (pVmdkFile->fAsyncIO)
652	return pImage->pInterfaceAsyncIOCallbacks->pfnFlush(pImage->pInterfaceAsyncIO->pvUser,
653	pVmdkFile->pStorage);
654	else
655	return RTFileFlush(pVmdkFile->File);
656	}
657
658	/**
659	* Internal: check if all files are closed, prevent leaking resources.
660	*/
661	static int vmdkFileCheckAllClose(PVMDKIMAGE pImage)
662	{
663	int rc = VINF_SUCCESS, rc2;
664	PVMDKFILE pVmdkFile;
665
666	Assert(pImage->pFiles == NULL);
667	for (pVmdkFile = pImage->pFiles;
668	pVmdkFile != NULL;
669	pVmdkFile = pVmdkFile->pNext)
670	{
671	LogRel(("VMDK: leaking reference to file \"%s\"\n",
672	pVmdkFile->pszFilename));
673	pImage->pFiles = pVmdkFile->pNext;
674
675	if (pImage->uOpenFlags & VD_OPEN_FLAGS_ASYNC_IO)
676	rc2 = pImage->pInterfaceAsyncIOCallbacks->pfnClose(pImage->pInterfaceAsyncIO->pvUser,
677	pVmdkFile->pStorage);
678	else
679	rc2 = RTFileClose(pVmdkFile->File);
680
681	if (RT_SUCCESS(rc) && pVmdkFile->fDelete)
682	rc2 = RTFileDelete(pVmdkFile->pszFilename);
683	RTStrFree((char )(void )pVmdkFile->pszFilename);
684	RTMemFree(pVmdkFile);
685	if (RT_SUCCESS(rc))
686	rc = rc2;
687	}
688	return rc;
689	}
690
691	/**
692	* Internal: truncate a string (at a UTF8 code point boundary) and encode the
693	* critical non-ASCII characters.
694	*/
695	static char vmdkEncodeString(const char psz)
696	{
697	char szEnc[VMDK_ENCODED_COMMENT_MAX + 3];
698	char *pszDst = szEnc;
699
700	Assert(VALID_PTR(psz));
701
702	for (; *psz; psz = RTStrNextCp(psz))
703	{
704	char *pszDstPrev = pszDst;
705	RTUNICP Cp = RTStrGetCp(psz);
706	if (Cp == '\\')
707	{
708	pszDst = RTStrPutCp(pszDst, Cp);
709	pszDst = RTStrPutCp(pszDst, Cp);
710	}
711	else if (Cp == '\n')
712	{
713	pszDst = RTStrPutCp(pszDst, '\\');
714	pszDst = RTStrPutCp(pszDst, 'n');
715	}
716	else if (Cp == '\r')
717	{
718	pszDst = RTStrPutCp(pszDst, '\\');
719	pszDst = RTStrPutCp(pszDst, 'r');
720	}
721	else
722	pszDst = RTStrPutCp(pszDst, Cp);
723	if (pszDst - szEnc >= VMDK_ENCODED_COMMENT_MAX - 1)
724	{
725	pszDst = pszDstPrev;
726	break;
727	}
728	}
729	*pszDst = '\0';
730	return RTStrDup(szEnc);
731	}
732
733	/**
734	* Internal: decode a string and store it into the specified string.
735	*/
736	static int vmdkDecodeString(const char pszEncoded, char psz, size_t cb)
737	{
738	int rc = VINF_SUCCESS;
739	char szBuf[4];
740
741	if (!cb)
742	return VERR_BUFFER_OVERFLOW;
743
744	Assert(VALID_PTR(psz));
745
746	for (; *pszEncoded; pszEncoded = RTStrNextCp(pszEncoded))
747	{
748	char *pszDst = szBuf;
749	RTUNICP Cp = RTStrGetCp(pszEncoded);
750	if (Cp == '\\')
751	{
752	pszEncoded = RTStrNextCp(pszEncoded);
753	RTUNICP CpQ = RTStrGetCp(pszEncoded);
754	if (CpQ == 'n')
755	RTStrPutCp(pszDst, '\n');
756	else if (CpQ == 'r')
757	RTStrPutCp(pszDst, '\r');
758	else if (CpQ == '\0')
759	{
760	rc = VERR_VDI_INVALID_HEADER;
761	break;
762	}
763	else
764	RTStrPutCp(pszDst, CpQ);
765	}
766	else
767	pszDst = RTStrPutCp(pszDst, Cp);
768
769	/* Need to leave space for terminating NUL. */
770	if ((size_t)(pszDst - szBuf) + 1 >= cb)
771	{
772	rc = VERR_BUFFER_OVERFLOW;
773	break;
774	}
775	memcpy(psz, szBuf, pszDst - szBuf);
776	psz += pszDst - szBuf;
777	}
778	*psz = '\0';
779	return rc;
780	}
781
782	static int vmdkReadGrainDirectory(PVMDKEXTENT pExtent)
783	{
784	int rc = VINF_SUCCESS;
785	unsigned i;
786	uint32_t pGD = NULL, pRGD = NULL, pGDTmp, pRGDTmp;
787	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
788
789	if (pExtent->enmType != VMDKETYPE_HOSTED_SPARSE)
790	goto out;
791
792	pGD = (uint32_t *)RTMemAllocZ(cbGD);
793	if (!pGD)
794	{
795	rc = VERR_NO_MEMORY;
796	goto out;
797	}
798	pExtent->pGD = pGD;
799	rc = vmdkFileReadAt(pExtent->pFile, VMDK_SECTOR2BYTE(pExtent->uSectorGD),
800	pGD, cbGD, NULL);
801	AssertRC(rc);
802	if (RT_FAILURE(rc))
803	{
804	rc = vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: could not read grain directory in '%s'"), pExtent->pszFullname);
805	goto out;
806	}
807	for (i = 0, pGDTmp = pGD; i < pExtent->cGDEntries; i++, pGDTmp++)
808	pGDTmp = RT_LE2H_U32(pGDTmp);
809
810	if (pExtent->uSectorRGD)
811	{
812	pRGD = (uint32_t *)RTMemAllocZ(cbGD);
813	if (!pRGD)
814	{
815	rc = VERR_NO_MEMORY;
816	goto out;
817	}
818	pExtent->pRGD = pRGD;
819	rc = vmdkFileReadAt(pExtent->pFile, VMDK_SECTOR2BYTE(pExtent->uSectorRGD),
820	pRGD, cbGD, NULL);
821	AssertRC(rc);
822	if (RT_FAILURE(rc))
823	{
824	rc = vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: could not read redundant grain directory in '%s'"), pExtent->pszFullname);
825	goto out;
826	}
827	for (i = 0, pRGDTmp = pRGD; i < pExtent->cGDEntries; i++, pRGDTmp++)
828	pRGDTmp = RT_LE2H_U32(pRGDTmp);
829
830	/* Check grain table and redundant grain table for consistency. */
831	size_t cbGT = pExtent->cGTEntries;
832	uint32_t pTmpGT1 = (uint32_t )RTMemTmpAlloc(cbGT);
833	if (!pTmpGT1)
834	{
835	rc = VERR_NO_MEMORY;
836	goto out;
837	}
838	uint32_t pTmpGT2 = (uint32_t )RTMemTmpAlloc(cbGT);
839	if (!pTmpGT2)
840	{
841	RTMemTmpFree(pTmpGT1);
842	rc = VERR_NO_MEMORY;
843	goto out;
844	}
845
846	for (i = 0, pGDTmp = pGD, pRGDTmp = pRGD;
847	i < pExtent->cGDEntries;
848	i++, pGDTmp++, pRGDTmp++)
849	{
850	/* If no grain table is allocated skip the entry. */
851	if (pGDTmp == 0 && pRGDTmp == 0)
852	continue;
853
854	if (pGDTmp == 0 \|\| pRGDTmp == 0 \|\| pGDTmp == pRGDTmp)
855	{
856	/* Just one grain directory entry refers to a not yet allocated
857	* grain table or both grain directory copies refer to the same
858	* grain table. Not allowed. */
859	RTMemTmpFree(pTmpGT1);
860	RTMemTmpFree(pTmpGT2);
861	rc = vmdkError(pExtent->pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: inconsistent references to grain directory in '%s'"), pExtent->pszFullname);
862	goto out;
863	}
864	rc = vmdkFileReadAt(pExtent->pFile, VMDK_SECTOR2BYTE(*pGDTmp),
865	pTmpGT1, cbGT, NULL);
866	if (RT_FAILURE(rc))
867	{
868	rc = vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: error reading grain table in '%s'"), pExtent->pszFullname);
869	RTMemTmpFree(pTmpGT1);
870	RTMemTmpFree(pTmpGT2);
871	goto out;
872	}
873	rc = vmdkFileReadAt(pExtent->pFile, VMDK_SECTOR2BYTE(*pRGDTmp),
874	pTmpGT2, cbGT, NULL);
875	if (RT_FAILURE(rc))
876	{
877	rc = vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: error reading backup grain table in '%s'"), pExtent->pszFullname);
878	RTMemTmpFree(pTmpGT1);
879	RTMemTmpFree(pTmpGT2);
880	goto out;
881	}
882	if (memcmp(pTmpGT1, pTmpGT2, cbGT))
883	{
884	RTMemTmpFree(pTmpGT1);
885	RTMemTmpFree(pTmpGT2);
886	rc = vmdkError(pExtent->pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: inconsistency between grain table and backup grain table in '%s'"), pExtent->pszFullname);
887	goto out;
888	}
889	}
890
891	/** @todo figure out what to do for unclean VMDKs. */
892	}
893
894	out:
895	if (RT_FAILURE(rc))
896	vmdkFreeGrainDirectory(pExtent);
897	return rc;
898	}
899
900	static int vmdkCreateGrainDirectory(PVMDKEXTENT pExtent, uint64_t uStartSector,
901	bool fPreAlloc)
902	{
903	int rc = VINF_SUCCESS;
904	unsigned i;
905	uint32_t pGD = NULL, pRGD = NULL;
906	size_t cbGD = pExtent->cGDEntries * sizeof(uint32_t);
907	size_t cbGDRounded = RT_ALIGN_64(pExtent->cGDEntries * sizeof(uint32_t), 512);
908	size_t cbGTRounded;
909	uint64_t cbOverhead;
910
911	if (fPreAlloc)
912	cbGTRounded = RT_ALIGN_64(pExtent->cGDEntries * pExtent->cGTEntries * sizeof(uint32_t), 512);
913	else
914	cbGTRounded = 0;
915
916	pGD = (uint32_t *)RTMemAllocZ(cbGD);
917	if (!pGD)
918	{
919	rc = VERR_NO_MEMORY;
920	goto out;
921	}
922	pExtent->pGD = pGD;
923	pRGD = (uint32_t *)RTMemAllocZ(cbGD);
924	if (!pRGD)
925	{
926	rc = VERR_NO_MEMORY;
927	goto out;
928	}
929	pExtent->pRGD = pRGD;
930
931	cbOverhead = RT_ALIGN_64(VMDK_SECTOR2BYTE(uStartSector) + 2 * (cbGDRounded + cbGTRounded), VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain));
932	rc = vmdkFileSetSize(pExtent->pFile, cbOverhead);
933	if (RT_FAILURE(rc))
934	goto out;
935	pExtent->uSectorRGD = uStartSector;
936	pExtent->uSectorGD = uStartSector + VMDK_BYTE2SECTOR(cbGDRounded + cbGTRounded);
937
938	if (fPreAlloc)
939	{
940	uint32_t uGTSectorLE;
941	uint64_t uOffsetSectors;
942
943	uOffsetSectors = pExtent->uSectorRGD + VMDK_BYTE2SECTOR(cbGDRounded);
944	for (i = 0; i < pExtent->cGDEntries; i++)
945	{
946	pRGD[i] = uOffsetSectors;
947	uGTSectorLE = RT_H2LE_U64(uOffsetSectors);
948	/* Write the redundant grain directory entry to disk. */
949	rc = vmdkFileWriteAt(pExtent->pFile,
950	VMDK_SECTOR2BYTE(pExtent->uSectorRGD) + i * sizeof(uGTSectorLE),
951	&uGTSectorLE, sizeof(uGTSectorLE), NULL);
952	if (RT_FAILURE(rc))
953	return vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: cannot write new redundant grain directory entry in '%s'"), pExtent->pszFullname);
954	uOffsetSectors += VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
955	}
956
957	uOffsetSectors = pExtent->uSectorGD + VMDK_BYTE2SECTOR(cbGDRounded);
958	for (i = 0; i < pExtent->cGDEntries; i++)
959	{
960	pGD[i] = uOffsetSectors;
961	uGTSectorLE = RT_H2LE_U64(uOffsetSectors);
962	/* Write the grain directory entry to disk. */
963	rc = vmdkFileWriteAt(pExtent->pFile,
964	VMDK_SECTOR2BYTE(pExtent->uSectorGD) + i * sizeof(uGTSectorLE),
965	&uGTSectorLE, sizeof(uGTSectorLE), NULL);
966	if (RT_FAILURE(rc))
967	return vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: cannot write new grain directory entry in '%s'"), pExtent->pszFullname);
968	uOffsetSectors += VMDK_BYTE2SECTOR(pExtent->cGTEntries * sizeof(uint32_t));
969	}
970	}
971	pExtent->cOverheadSectors = VMDK_BYTE2SECTOR(cbOverhead);
972
973	out:
974	if (RT_FAILURE(rc))
975	vmdkFreeGrainDirectory(pExtent);
976	return rc;
977	}
978
979	static void vmdkFreeGrainDirectory(PVMDKEXTENT pExtent)
980	{
981	if (pExtent->pGD)
982	{
983	RTMemFree(pExtent->pGD);
984	pExtent->pGD = NULL;
985	}
986	if (pExtent->pRGD)
987	{
988	RTMemFree(pExtent->pRGD);
989	pExtent->pRGD = NULL;
990	}
991	}
992
993	static int vmdkStringUnquote(PVMDKIMAGE pImage, const char *pszStr,
994	char ppszUnquoted, char ppszNext)
995	{
996	char *pszQ;
997	char *pszUnquoted;
998
999	/* Skip over whitespace. */
1000	while (pszStr == ' ' \|\| pszStr == '\t')
1001	pszStr++;
1002	if (*pszStr++ != '"')
1003	return vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrectly quoted value in descriptor in '%s'"), pImage->pszFilename);
1004
1005	pszQ = (char *)strchr(pszStr, '"');
1006	if (pszQ == NULL)
1007	return vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrectly quoted value in descriptor in '%s'"), pImage->pszFilename);
1008	pszUnquoted = (char *)RTMemTmpAlloc(pszQ - pszStr + 1);
1009	if (!pszUnquoted)
1010	return VERR_NO_MEMORY;
1011	memcpy(pszUnquoted, pszStr, pszQ - pszStr);
1012	pszUnquoted[pszQ - pszStr] = '\0';
1013	*ppszUnquoted = pszUnquoted;
1014	if (ppszNext)
1015	*ppszNext = pszQ + 1;
1016	return VINF_SUCCESS;
1017	}
1018
1019	static int vmdkDescInitStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1020	const char *pszLine)
1021	{
1022	char *pEnd = pDescriptor->aLines[pDescriptor->cLines];
1023	ssize_t cbDiff = strlen(pszLine) + 1;
1024
1025	if ( pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1
1026	&& pEnd - pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff)
1027	return vmdkError(pImage, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1028
1029	memcpy(pEnd, pszLine, cbDiff);
1030	pDescriptor->cLines++;
1031	pDescriptor->aLines[pDescriptor->cLines] = pEnd + cbDiff;
1032	pDescriptor->fDirty = true;
1033
1034	return VINF_SUCCESS;
1035	}
1036
1037	static bool vmdkDescGetStr(PVMDKDESCRIPTOR pDescriptor, unsigned uStart,
1038	const char pszKey, const char *ppszValue)
1039	{
1040	size_t cbKey = strlen(pszKey);
1041	const char *pszValue;
1042
1043	while (uStart != 0)
1044	{
1045	if (!strncmp(pDescriptor->aLines[uStart], pszKey, cbKey))
1046	{
1047	/* Key matches, check for a '=' (preceded by whitespace). */
1048	pszValue = pDescriptor->aLines[uStart] + cbKey;
1049	while (pszValue == ' ' \|\| pszValue == '\t')
1050	pszValue++;
1051	if (*pszValue == '=')
1052	{
1053	*ppszValue = pszValue + 1;
1054	break;
1055	}
1056	}
1057	uStart = pDescriptor->aNextLines[uStart];
1058	}
1059	return !!uStart;
1060	}
1061
1062	static int vmdkDescSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1063	unsigned uStart,
1064	const char pszKey, const char pszValue)
1065	{
1066	char *pszTmp;
1067	size_t cbKey = strlen(pszKey);
1068	unsigned uLast = 0;
1069
1070	while (uStart != 0)
1071	{
1072	if (!strncmp(pDescriptor->aLines[uStart], pszKey, cbKey))
1073	{
1074	/* Key matches, check for a '=' (preceded by whitespace). */
1075	pszTmp = pDescriptor->aLines[uStart] + cbKey;
1076	while (pszTmp == ' ' \|\| pszTmp == '\t')
1077	pszTmp++;
1078	if (*pszTmp == '=')
1079	{
1080	pszTmp++;
1081	while (pszTmp == ' ' \|\| pszTmp == '\t')
1082	pszTmp++;
1083	break;
1084	}
1085	}
1086	if (!pDescriptor->aNextLines[uStart])
1087	uLast = uStart;
1088	uStart = pDescriptor->aNextLines[uStart];
1089	}
1090	if (uStart)
1091	{
1092	if (pszValue)
1093	{
1094	/* Key already exists, replace existing value. */
1095	size_t cbOldVal = strlen(pszTmp);
1096	size_t cbNewVal = strlen(pszValue);
1097	ssize_t cbDiff = cbNewVal - cbOldVal;
1098	/* Check for buffer overflow. */
1099	if ( pDescriptor->aLines[pDescriptor->cLines]
1100	- pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff)
1101	return vmdkError(pImage, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1102
1103	memmove(pszTmp + cbNewVal, pszTmp + cbOldVal,
1104	pDescriptor->aLines[pDescriptor->cLines] - pszTmp - cbOldVal);
1105	memcpy(pszTmp, pszValue, cbNewVal + 1);
1106	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1107	pDescriptor->aLines[i] += cbDiff;
1108	}
1109	else
1110	{
1111	memmove(pDescriptor->aLines[uStart], pDescriptor->aLines[uStart+1],
1112	pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[uStart+1] + 1);
1113	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1114	{
1115	pDescriptor->aLines[i-1] = pDescriptor->aLines[i];
1116	if (pDescriptor->aNextLines[i])
1117	pDescriptor->aNextLines[i-1] = pDescriptor->aNextLines[i] - 1;
1118	else
1119	pDescriptor->aNextLines[i-1] = 0;
1120	}
1121	pDescriptor->cLines--;
1122	/* Adjust starting line numbers of following descriptor sections. */
1123	if (uStart < pDescriptor->uFirstExtent)
1124	pDescriptor->uFirstExtent--;
1125	if (uStart < pDescriptor->uFirstDDB)
1126	pDescriptor->uFirstDDB--;
1127	}
1128	}
1129	else
1130	{
1131	/* Key doesn't exist, append after the last entry in this category. */
1132	if (!pszValue)
1133	{
1134	/* Key doesn't exist, and it should be removed. Simply a no-op. */
1135	return VINF_SUCCESS;
1136	}
1137	size_t cbKey = strlen(pszKey);
1138	size_t cbValue = strlen(pszValue);
1139	ssize_t cbDiff = cbKey + 1 + cbValue + 1;
1140	/* Check for buffer overflow. */
1141	if ( (pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1)
1142	\|\| ( pDescriptor->aLines[pDescriptor->cLines]
1143	- pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff))
1144	return vmdkError(pImage, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1145	for (unsigned i = pDescriptor->cLines + 1; i > uLast + 1; i--)
1146	{
1147	pDescriptor->aLines[i] = pDescriptor->aLines[i - 1];
1148	if (pDescriptor->aNextLines[i - 1])
1149	pDescriptor->aNextLines[i] = pDescriptor->aNextLines[i - 1] + 1;
1150	else
1151	pDescriptor->aNextLines[i] = 0;
1152	}
1153	uStart = uLast + 1;
1154	pDescriptor->aNextLines[uLast] = uStart;
1155	pDescriptor->aNextLines[uStart] = 0;
1156	pDescriptor->cLines++;
1157	pszTmp = pDescriptor->aLines[uStart];
1158	memmove(pszTmp + cbDiff, pszTmp,
1159	pDescriptor->aLines[pDescriptor->cLines] - pszTmp);
1160	memcpy(pDescriptor->aLines[uStart], pszKey, cbKey);
1161	pDescriptor->aLines[uStart][cbKey] = '=';
1162	memcpy(pDescriptor->aLines[uStart] + cbKey + 1, pszValue, cbValue + 1);
1163	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1164	pDescriptor->aLines[i] += cbDiff;
1165
1166	/* Adjust starting line numbers of following descriptor sections. */
1167	if (uStart <= pDescriptor->uFirstExtent)
1168	pDescriptor->uFirstExtent++;
1169	if (uStart <= pDescriptor->uFirstDDB)
1170	pDescriptor->uFirstDDB++;
1171	}
1172	pDescriptor->fDirty = true;
1173	return VINF_SUCCESS;
1174	}
1175
1176	static int vmdkDescBaseGetU32(PVMDKDESCRIPTOR pDescriptor, const char *pszKey,
1177	uint32_t *puValue)
1178	{
1179	const char *pszValue;
1180
1181	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDesc, pszKey,
1182	&pszValue))
1183	return VERR_VDI_VALUE_NOT_FOUND;
1184	return RTStrToUInt32Ex(pszValue, NULL, 10, puValue);
1185	}
1186
1187	static int vmdkDescBaseGetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1188	const char pszKey, const char *ppszValue)
1189	{
1190	const char *pszValue;
1191	char *pszValueUnquoted;
1192
1193	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDesc, pszKey,
1194	&pszValue))
1195	return VERR_VDI_VALUE_NOT_FOUND;
1196	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1197	if (RT_FAILURE(rc))
1198	return rc;
1199	*ppszValue = pszValueUnquoted;
1200	return rc;
1201	}
1202
1203	static int vmdkDescBaseSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1204	const char pszKey, const char pszValue)
1205	{
1206	char *pszValueQuoted;
1207
1208	int rc = RTStrAPrintf(&pszValueQuoted, "\"%s\"", pszValue);
1209	if (RT_FAILURE(rc))
1210	return rc;
1211	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc, pszKey,
1212	pszValueQuoted);
1213	RTStrFree(pszValueQuoted);
1214	return rc;
1215	}
1216
1217	static void vmdkDescExtRemoveDummy(PVMDKIMAGE pImage,
1218	PVMDKDESCRIPTOR pDescriptor)
1219	{
1220	unsigned uEntry = pDescriptor->uFirstExtent;
1221	ssize_t cbDiff;
1222
1223	if (!uEntry)
1224	return;
1225
1226	cbDiff = strlen(pDescriptor->aLines[uEntry]) + 1;
1227	/* Move everything including \0 in the entry marking the end of buffer. */
1228	memmove(pDescriptor->aLines[uEntry], pDescriptor->aLines[uEntry + 1],
1229	pDescriptor->aLines[pDescriptor->cLines] - pDescriptor->aLines[uEntry + 1] + 1);
1230	for (unsigned i = uEntry + 1; i <= pDescriptor->cLines; i++)
1231	{
1232	pDescriptor->aLines[i - 1] = pDescriptor->aLines[i] - cbDiff;
1233	if (pDescriptor->aNextLines[i])
1234	pDescriptor->aNextLines[i - 1] = pDescriptor->aNextLines[i] - 1;
1235	else
1236	pDescriptor->aNextLines[i - 1] = 0;
1237	}
1238	pDescriptor->cLines--;
1239	if (pDescriptor->uFirstDDB)
1240	pDescriptor->uFirstDDB--;
1241
1242	return;
1243	}
1244
1245	static int vmdkDescExtInsert(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1246	VMDKACCESS enmAccess, uint64_t cNominalSectors,
1247	VMDKETYPE enmType, const char *pszBasename,
1248	uint64_t uSectorOffset)
1249	{
1250	static const char *apszAccess[] = { "NOACCESS", "RDONLY", "RW" };
1251	static const char *apszType[] = { "", "SPARSE", "FLAT", "ZERO" };
1252	char *pszTmp;
1253	unsigned uStart = pDescriptor->uFirstExtent, uLast = 0;
1254	char szExt[1024];
1255	ssize_t cbDiff;
1256
1257	/* Find last entry in extent description. */
1258	while (uStart)
1259	{
1260	if (!pDescriptor->aNextLines[uStart])
1261	uLast = uStart;
1262	uStart = pDescriptor->aNextLines[uStart];
1263	}
1264
1265	if (enmType == VMDKETYPE_ZERO)
1266	{
1267	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s ", apszAccess[enmAccess],
1268	cNominalSectors, apszType[enmType]);
1269	}
1270	else
1271	{
1272	if (!uSectorOffset)
1273	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s \"%s\"",
1274	apszAccess[enmAccess], cNominalSectors,
1275	apszType[enmType], pszBasename);
1276	else
1277	RTStrPrintf(szExt, sizeof(szExt), "%s %llu %s \"%s\" %llu",
1278	apszAccess[enmAccess], cNominalSectors,
1279	apszType[enmType], pszBasename, uSectorOffset);
1280	}
1281	cbDiff = strlen(szExt) + 1;
1282
1283	/* Check for buffer overflow. */
1284	if ( (pDescriptor->cLines >= VMDK_DESCRIPTOR_LINES_MAX - 1)
1285	\|\| ( pDescriptor->aLines[pDescriptor->cLines]
1286	- pDescriptor->aLines[0] > (ptrdiff_t)pDescriptor->cbDescAlloc - cbDiff))
1287	return vmdkError(pImage, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1288
1289	for (unsigned i = pDescriptor->cLines + 1; i > uLast + 1; i--)
1290	{
1291	pDescriptor->aLines[i] = pDescriptor->aLines[i - 1];
1292	if (pDescriptor->aNextLines[i - 1])
1293	pDescriptor->aNextLines[i] = pDescriptor->aNextLines[i - 1] + 1;
1294	else
1295	pDescriptor->aNextLines[i] = 0;
1296	}
1297	uStart = uLast + 1;
1298	pDescriptor->aNextLines[uLast] = uStart;
1299	pDescriptor->aNextLines[uStart] = 0;
1300	pDescriptor->cLines++;
1301	pszTmp = pDescriptor->aLines[uStart];
1302	memmove(pszTmp + cbDiff, pszTmp,
1303	pDescriptor->aLines[pDescriptor->cLines] - pszTmp);
1304	memcpy(pDescriptor->aLines[uStart], szExt, cbDiff);
1305	for (unsigned i = uStart + 1; i <= pDescriptor->cLines; i++)
1306	pDescriptor->aLines[i] += cbDiff;
1307
1308	/* Adjust starting line numbers of following descriptor sections. */
1309	if (uStart <= pDescriptor->uFirstDDB)
1310	pDescriptor->uFirstDDB++;
1311
1312	pDescriptor->fDirty = true;
1313	return VINF_SUCCESS;
1314	}
1315
1316	static int vmdkDescDDBGetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1317	const char pszKey, const char *ppszValue)
1318	{
1319	const char *pszValue;
1320	char *pszValueUnquoted;
1321
1322	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1323	&pszValue))
1324	return VERR_VDI_VALUE_NOT_FOUND;
1325	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1326	if (RT_FAILURE(rc))
1327	return rc;
1328	*ppszValue = pszValueUnquoted;
1329	return rc;
1330	}
1331
1332	static int vmdkDescDDBGetU32(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1333	const char pszKey, uint32_t puValue)
1334	{
1335	const char *pszValue;
1336	char *pszValueUnquoted;
1337
1338	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1339	&pszValue))
1340	return VERR_VDI_VALUE_NOT_FOUND;
1341	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1342	if (RT_FAILURE(rc))
1343	return rc;
1344	rc = RTStrToUInt32Ex(pszValueUnquoted, NULL, 10, puValue);
1345	RTMemTmpFree(pszValueUnquoted);
1346	return rc;
1347	}
1348
1349	static int vmdkDescDDBGetUuid(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1350	const char *pszKey, PRTUUID pUuid)
1351	{
1352	const char *pszValue;
1353	char *pszValueUnquoted;
1354
1355	if (!vmdkDescGetStr(pDescriptor, pDescriptor->uFirstDDB, pszKey,
1356	&pszValue))
1357	return VERR_VDI_VALUE_NOT_FOUND;
1358	int rc = vmdkStringUnquote(pImage, pszValue, &pszValueUnquoted, NULL);
1359	if (RT_FAILURE(rc))
1360	return rc;
1361	rc = RTUuidFromStr(pUuid, pszValueUnquoted);
1362	RTMemTmpFree(pszValueUnquoted);
1363	return rc;
1364	}
1365
1366	static int vmdkDescDDBSetStr(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1367	const char pszKey, const char pszVal)
1368	{
1369	int rc;
1370	char *pszValQuoted;
1371
1372	if (pszVal)
1373	{
1374	rc = RTStrAPrintf(&pszValQuoted, "\"%s\"", pszVal);
1375	if (RT_FAILURE(rc))
1376	return rc;
1377	}
1378	else
1379	pszValQuoted = NULL;
1380	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1381	pszValQuoted);
1382	if (pszValQuoted)
1383	RTStrFree(pszValQuoted);
1384	return rc;
1385	}
1386
1387	static int vmdkDescDDBSetUuid(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1388	const char *pszKey, PCRTUUID pUuid)
1389	{
1390	char *pszUuid;
1391
1392	int rc = RTStrAPrintf(&pszUuid, "\"%RTuuid\"", pUuid);
1393	if (RT_FAILURE(rc))
1394	return rc;
1395	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1396	pszUuid);
1397	RTStrFree(pszUuid);
1398	return rc;
1399	}
1400
1401	static int vmdkDescDDBSetU32(PVMDKIMAGE pImage, PVMDKDESCRIPTOR pDescriptor,
1402	const char *pszKey, uint32_t uValue)
1403	{
1404	char *pszValue;
1405
1406	int rc = RTStrAPrintf(&pszValue, "\"%d\"", uValue);
1407	if (RT_FAILURE(rc))
1408	return rc;
1409	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDDB, pszKey,
1410	pszValue);
1411	RTStrFree(pszValue);
1412	return rc;
1413	}
1414
1415	static int vmdkPreprocessDescriptor(PVMDKIMAGE pImage, char *pDescData,
1416	size_t cbDescData,
1417	PVMDKDESCRIPTOR pDescriptor)
1418	{
1419	int rc = VINF_SUCCESS;
1420	unsigned cLine = 0, uLastNonEmptyLine = 0;
1421	char *pTmp = pDescData;
1422
1423	pDescriptor->cbDescAlloc = cbDescData;
1424	while (*pTmp != '\0')
1425	{
1426	pDescriptor->aLines[cLine++] = pTmp;
1427	if (cLine >= VMDK_DESCRIPTOR_LINES_MAX)
1428	{
1429	rc = vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor too big in '%s'"), pImage->pszFilename);
1430	goto out;
1431	}
1432
1433	while (pTmp != '\0' && pTmp != '\n')
1434	{
1435	if (*pTmp == '\r')
1436	{
1437	if (*(pTmp + 1) != '\n')
1438	{
1439	rc = vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: unsupported end of line in descriptor in '%s'"), pImage->pszFilename);
1440	goto out;
1441	}
1442	else
1443	{
1444	/* Get rid of CR character. */
1445	*pTmp = '\0';
1446	}
1447	}
1448	pTmp++;
1449	}
1450	/* Get rid of LF character. */
1451	if (*pTmp == '\n')
1452	{
1453	*pTmp = '\0';
1454	pTmp++;
1455	}
1456	}
1457	pDescriptor->cLines = cLine;
1458	/* Pointer right after the end of the used part of the buffer. */
1459	pDescriptor->aLines[cLine] = pTmp;
1460
1461	if ( strcmp(pDescriptor->aLines[0], "# Disk DescriptorFile")
1462	&& strcmp(pDescriptor->aLines[0], "# Disk Descriptor File"))
1463	{
1464	rc = vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: descriptor does not start as expected in '%s'"), pImage->pszFilename);
1465	goto out;
1466	}
1467
1468	/* Initialize those, because we need to be able to reopen an image. */
1469	pDescriptor->uFirstDesc = 0;
1470	pDescriptor->uFirstExtent = 0;
1471	pDescriptor->uFirstDDB = 0;
1472	for (unsigned i = 0; i < cLine; i++)
1473	{
1474	if (pDescriptor->aLines[i] != '#' && pDescriptor->aLines[i] != '\0')
1475	{
1476	if ( !strncmp(pDescriptor->aLines[i], "RW", 2)
1477	\|\| !strncmp(pDescriptor->aLines[i], "RDONLY", 6)
1478	\|\| !strncmp(pDescriptor->aLines[i], "NOACCESS", 8) )
1479	{
1480	/* An extent descriptor. */
1481	if (!pDescriptor->uFirstDesc \|\| pDescriptor->uFirstDDB)
1482	{
1483	/* Incorrect ordering of entries. */
1484	rc = vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
1485	goto out;
1486	}
1487	if (!pDescriptor->uFirstExtent)
1488	{
1489	pDescriptor->uFirstExtent = i;
1490	uLastNonEmptyLine = 0;
1491	}
1492	}
1493	else if (!strncmp(pDescriptor->aLines[i], "ddb.", 4))
1494	{
1495	/* A disk database entry. */
1496	if (!pDescriptor->uFirstDesc \|\| !pDescriptor->uFirstExtent)
1497	{
1498	/* Incorrect ordering of entries. */
1499	rc = vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
1500	goto out;
1501	}
1502	if (!pDescriptor->uFirstDDB)
1503	{
1504	pDescriptor->uFirstDDB = i;
1505	uLastNonEmptyLine = 0;
1506	}
1507	}
1508	else
1509	{
1510	/* A normal entry. */
1511	if (pDescriptor->uFirstExtent \|\| pDescriptor->uFirstDDB)
1512	{
1513	/* Incorrect ordering of entries. */
1514	rc = vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrect ordering of entries in descriptor in '%s'"), pImage->pszFilename);
1515	goto out;
1516	}
1517	if (!pDescriptor->uFirstDesc)
1518	{
1519	pDescriptor->uFirstDesc = i;
1520	uLastNonEmptyLine = 0;
1521	}
1522	}
1523	if (uLastNonEmptyLine)
1524	pDescriptor->aNextLines[uLastNonEmptyLine] = i;
1525	uLastNonEmptyLine = i;
1526	}
1527	}
1528
1529	out:
1530	return rc;
1531	}
1532
1533	static int vmdkDescSetPCHSGeometry(PVMDKIMAGE pImage,
1534	PCPDMMEDIAGEOMETRY pPCHSGeometry)
1535	{
1536	int rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1537	VMDK_DDB_GEO_PCHS_CYLINDERS,
1538	pPCHSGeometry->cCylinders);
1539	if (RT_FAILURE(rc))
1540	return rc;
1541	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1542	VMDK_DDB_GEO_PCHS_HEADS,
1543	pPCHSGeometry->cHeads);
1544	if (RT_FAILURE(rc))
1545	return rc;
1546	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1547	VMDK_DDB_GEO_PCHS_SECTORS,
1548	pPCHSGeometry->cSectors);
1549	return rc;
1550	}
1551
1552	static int vmdkDescSetLCHSGeometry(PVMDKIMAGE pImage,
1553	PCPDMMEDIAGEOMETRY pLCHSGeometry)
1554	{
1555	int rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1556	VMDK_DDB_GEO_LCHS_CYLINDERS,
1557	pLCHSGeometry->cCylinders);
1558	if (RT_FAILURE(rc))
1559	return rc;
1560	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1561	VMDK_DDB_GEO_LCHS_HEADS,
1562	pLCHSGeometry->cHeads);
1563	if (RT_FAILURE(rc))
1564	return rc;
1565	rc = vmdkDescDDBSetU32(pImage, &pImage->Descriptor,
1566	VMDK_DDB_GEO_LCHS_SECTORS,
1567	pLCHSGeometry->cSectors);
1568	return rc;
1569	}
1570
1571	static int vmdkCreateDescriptor(PVMDKIMAGE pImage, char *pDescData,
1572	size_t cbDescData, PVMDKDESCRIPTOR pDescriptor)
1573	{
1574	int rc;
1575
1576	pDescriptor->uFirstDesc = 0;
1577	pDescriptor->uFirstExtent = 0;
1578	pDescriptor->uFirstDDB = 0;
1579	pDescriptor->cLines = 0;
1580	pDescriptor->cbDescAlloc = cbDescData;
1581	pDescriptor->fDirty = false;
1582	pDescriptor->aLines[pDescriptor->cLines] = pDescData;
1583	memset(pDescriptor->aNextLines, '\0', sizeof(pDescriptor->aNextLines));
1584
1585	rc = vmdkDescInitStr(pImage, pDescriptor, "# Disk DescriptorFile");
1586	if (RT_FAILURE(rc))
1587	goto out;
1588	rc = vmdkDescInitStr(pImage, pDescriptor, "version=1");
1589	if (RT_FAILURE(rc))
1590	goto out;
1591	pDescriptor->uFirstDesc = pDescriptor->cLines - 1;
1592	rc = vmdkDescInitStr(pImage, pDescriptor, "");
1593	if (RT_FAILURE(rc))
1594	goto out;
1595	rc = vmdkDescInitStr(pImage, pDescriptor, "# Extent description");
1596	if (RT_FAILURE(rc))
1597	goto out;
1598	rc = vmdkDescInitStr(pImage, pDescriptor, "NOACCESS 0 ZERO ");
1599	if (RT_FAILURE(rc))
1600	goto out;
1601	pDescriptor->uFirstExtent = pDescriptor->cLines - 1;
1602	rc = vmdkDescInitStr(pImage, pDescriptor, "");
1603	if (RT_FAILURE(rc))
1604	goto out;
1605	/* The trailing space is created by VMware, too. */
1606	rc = vmdkDescInitStr(pImage, pDescriptor, "# The disk Data Base ");
1607	if (RT_FAILURE(rc))
1608	goto out;
1609	rc = vmdkDescInitStr(pImage, pDescriptor, "#DDB");
1610	if (RT_FAILURE(rc))
1611	goto out;
1612	rc = vmdkDescInitStr(pImage, pDescriptor, "");
1613	if (RT_FAILURE(rc))
1614	goto out;
1615	rc = vmdkDescInitStr(pImage, pDescriptor, "ddb.virtualHWVersion = \"4\"");
1616	if (RT_FAILURE(rc))
1617	goto out;
1618	pDescriptor->uFirstDDB = pDescriptor->cLines - 1;
1619
1620	/* Now that the framework is in place, use the normal functions to insert
1621	* the remaining keys. */
1622	char szBuf[9];
1623	RTStrPrintf(szBuf, sizeof(szBuf), "%08x", RTRandU32());
1624	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc,
1625	"CID", szBuf);
1626	if (RT_FAILURE(rc))
1627	goto out;
1628	rc = vmdkDescSetStr(pImage, pDescriptor, pDescriptor->uFirstDesc,
1629	"parentCID", "ffffffff");
1630	if (RT_FAILURE(rc))
1631	goto out;
1632
1633	rc = vmdkDescDDBSetStr(pImage, pDescriptor, "ddb.adapterType", "ide");
1634	if (RT_FAILURE(rc))
1635	goto out;
1636
1637	out:
1638	return rc;
1639	}
1640
1641	static int vmdkParseDescriptor(PVMDKIMAGE pImage, char *pDescData,
1642	size_t cbDescData)
1643	{
1644	int rc;
1645	unsigned cExtents;
1646	unsigned uLine;
1647
1648	rc = vmdkPreprocessDescriptor(pImage, pDescData, cbDescData,
1649	&pImage->Descriptor);
1650	if (RT_FAILURE(rc))
1651	return rc;
1652
1653	/* Check version, must be 1. */
1654	uint32_t uVersion;
1655	rc = vmdkDescBaseGetU32(&pImage->Descriptor, "version", &uVersion);
1656	if (RT_FAILURE(rc))
1657	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error finding key 'version' in descriptor in '%s'"), pImage->pszFilename);
1658	if (uVersion != 1)
1659	return vmdkError(pImage, VERR_VDI_UNSUPPORTED_VERSION, RT_SRC_POS, N_("VMDK: unsupported format version in descriptor in '%s'"), pImage->pszFilename);
1660
1661	/* Get image creation type and determine image flags. */
1662	const char *pszCreateType;
1663	rc = vmdkDescBaseGetStr(pImage, &pImage->Descriptor, "createType",
1664	&pszCreateType);
1665	if (RT_FAILURE(rc))
1666	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: cannot get image type from descriptor in '%s'"), pImage->pszFilename);
1667	if ( !strcmp(pszCreateType, "twoGbMaxExtentSparse")
1668	\|\| !strcmp(pszCreateType, "twoGbMaxExtentFlat"))
1669	pImage->uImageFlags = VD_VMDK_IMAGE_FLAGS_SPLIT_2G;
1670	if ( !strcmp(pszCreateType, "partitionedDevice")
1671	\|\| !strcmp(pszCreateType, "fullDevice"))
1672	pImage->uImageFlags = VD_VMDK_IMAGE_FLAGS_RAWDISK;
1673	else
1674	pImage->uImageFlags = 0;
1675	RTStrFree((char )(void )pszCreateType);
1676
1677	/* Count the number of extent config entries. */
1678	for (uLine = pImage->Descriptor.uFirstExtent, cExtents = 0;
1679	uLine != 0;
1680	uLine = pImage->Descriptor.aNextLines[uLine], cExtents++)
1681	/* nothing */;
1682
1683	if (!pImage->pDescData && cExtents != 1)
1684	{
1685	/* Monolithic image, must have only one extent (already opened). */
1686	return vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: monolithic image may only have one extent in '%s'"), pImage->pszFilename);
1687	}
1688
1689	if (pImage->pDescData)
1690	{
1691	/* Non-monolithic image, extents need to be allocated. */
1692	rc = vmdkCreateExtents(pImage, cExtents);
1693	if (RT_FAILURE(rc))
1694	return rc;
1695	}
1696
1697	for (unsigned i = 0, uLine = pImage->Descriptor.uFirstExtent;
1698	i < cExtents; i++, uLine = pImage->Descriptor.aNextLines[uLine])
1699	{
1700	char *pszLine = pImage->Descriptor.aLines[uLine];
1701
1702	/* Access type of the extent. */
1703	if (!strncmp(pszLine, "RW", 2))
1704	{
1705	pImage->pExtents[i].enmAccess = VMDKACCESS_READWRITE;
1706	pszLine += 2;
1707	}
1708	else if (!strncmp(pszLine, "RDONLY", 6))
1709	{
1710	pImage->pExtents[i].enmAccess = VMDKACCESS_READONLY;
1711	pszLine += 6;
1712	}
1713	else if (!strncmp(pszLine, "NOACCESS", 8))
1714	{
1715	pImage->pExtents[i].enmAccess = VMDKACCESS_NOACCESS;
1716	pszLine += 8;
1717	}
1718	else
1719	return vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
1720	if (*pszLine++ != ' ')
1721	return vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
1722
1723	/* Nominal size of the extent. */
1724	rc = RTStrToUInt64Ex(pszLine, &pszLine, 10,
1725	&pImage->pExtents[i].cNominalSectors);
1726	if (RT_FAILURE(rc))
1727	return vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
1728	if (*pszLine++ != ' ')
1729	return vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
1730
1731	/* Type of the extent. */
1732	#ifdef VBOX_WITH_VMDK_ESX
1733	/** @todo Add the ESX extent types. Not necessary for now because
1734	* the ESX extent types are only used inside an ESX server. They are
1735	* automatically converted if the VMDK is exported. */
1736	#endif /* VBOX_WITH_VMDK_ESX */
1737	if (!strncmp(pszLine, "SPARSE", 6))
1738	{
1739	pImage->pExtents[i].enmType = VMDKETYPE_HOSTED_SPARSE;
1740	pszLine += 6;
1741	}
1742	else if (!strncmp(pszLine, "FLAT", 4))
1743	{
1744	pImage->pExtents[i].enmType = VMDKETYPE_FLAT;
1745	pszLine += 4;
1746	}
1747	else if (!strncmp(pszLine, "ZERO", 4))
1748	{
1749	pImage->pExtents[i].enmType = VMDKETYPE_ZERO;
1750	pszLine += 4;
1751	}
1752	else
1753	return vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
1754	if (pImage->pExtents[i].enmType == VMDKETYPE_ZERO)
1755	{
1756	/* This one has no basename or offset. */
1757	if (*pszLine == ' ')
1758	pszLine++;
1759	if (*pszLine != '\0')
1760	return vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
1761	pImage->pExtents[i].pszBasename = NULL;
1762	}
1763	else
1764	{
1765	/* All other extent types have basename and optional offset. */
1766	if (*pszLine++ != ' ')
1767	return vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
1768
1769	/* Basename of the image. Surrounded by quotes. */
1770	char *pszBasename;
1771	rc = vmdkStringUnquote(pImage, pszLine, &pszBasename, &pszLine);
1772	if (RT_FAILURE(rc))
1773	return rc;
1774	pImage->pExtents[i].pszBasename = pszBasename;
1775	if (*pszLine == ' ')
1776	{
1777	pszLine++;
1778	if (*pszLine != '\0')
1779	{
1780	/* Optional offset in extent specified. */
1781	rc = RTStrToUInt64Ex(pszLine, &pszLine, 10,
1782	&pImage->pExtents[i].uSectorOffset);
1783	if (RT_FAILURE(rc))
1784	return vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
1785	}
1786	}
1787
1788	if (*pszLine != '\0')
1789	return vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: parse error in extent description in '%s'"), pImage->pszFilename);
1790	}
1791	}
1792
1793	/* Determine PCHS geometry (autogenerate if necessary). */
1794	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
1795	VMDK_DDB_GEO_PCHS_CYLINDERS,
1796	&pImage->PCHSGeometry.cCylinders);
1797	if (rc == VERR_VDI_VALUE_NOT_FOUND)
1798	pImage->PCHSGeometry.cCylinders = 0;
1799	else if (RT_FAILURE(rc))
1800	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
1801	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
1802	VMDK_DDB_GEO_PCHS_HEADS,
1803	&pImage->PCHSGeometry.cHeads);
1804	if (rc == VERR_VDI_VALUE_NOT_FOUND)
1805	pImage->PCHSGeometry.cHeads = 0;
1806	else if (RT_FAILURE(rc))
1807	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
1808	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
1809	VMDK_DDB_GEO_PCHS_SECTORS,
1810	&pImage->PCHSGeometry.cSectors);
1811	if (rc == VERR_VDI_VALUE_NOT_FOUND)
1812	pImage->PCHSGeometry.cSectors = 0;
1813	else if (RT_FAILURE(rc))
1814	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error getting PCHS geometry from extent description in '%s'"), pImage->pszFilename);
1815	if ( pImage->PCHSGeometry.cCylinders == 0
1816	\|\| pImage->PCHSGeometry.cHeads == 0
1817	\|\| pImage->PCHSGeometry.cHeads > 16
1818	\|\| pImage->PCHSGeometry.cSectors == 0
1819	\|\| pImage->PCHSGeometry.cSectors > 63)
1820	{
1821	/* Mark PCHS geometry as not yet valid (can't do the calculation here
1822	* as the total image size isn't known yet). */
1823	pImage->PCHSGeometry.cCylinders = 0;
1824	pImage->PCHSGeometry.cHeads = 16;
1825	pImage->PCHSGeometry.cSectors = 63;
1826	}
1827
1828	/* Determine LCHS geometry (set to 0 if not specified). */
1829	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
1830	VMDK_DDB_GEO_LCHS_CYLINDERS,
1831	&pImage->LCHSGeometry.cCylinders);
1832	if (rc == VERR_VDI_VALUE_NOT_FOUND)
1833	pImage->LCHSGeometry.cCylinders = 0;
1834	else if (RT_FAILURE(rc))
1835	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
1836	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
1837	VMDK_DDB_GEO_LCHS_HEADS,
1838	&pImage->LCHSGeometry.cHeads);
1839	if (rc == VERR_VDI_VALUE_NOT_FOUND)
1840	pImage->LCHSGeometry.cHeads = 0;
1841	else if (RT_FAILURE(rc))
1842	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
1843	rc = vmdkDescDDBGetU32(pImage, &pImage->Descriptor,
1844	VMDK_DDB_GEO_LCHS_SECTORS,
1845	&pImage->LCHSGeometry.cSectors);
1846	if (rc == VERR_VDI_VALUE_NOT_FOUND)
1847	pImage->LCHSGeometry.cSectors = 0;
1848	else if (RT_FAILURE(rc))
1849	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error getting LCHS geometry from extent description in '%s'"), pImage->pszFilename);
1850	if ( pImage->LCHSGeometry.cCylinders == 0
1851	\|\| pImage->LCHSGeometry.cHeads == 0
1852	\|\| pImage->LCHSGeometry.cSectors == 0)
1853	{
1854	pImage->LCHSGeometry.cCylinders = 0;
1855	pImage->LCHSGeometry.cHeads = 0;
1856	pImage->LCHSGeometry.cSectors = 0;
1857	}
1858
1859	/* Get image UUID. */
1860	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor, VMDK_DDB_IMAGE_UUID,
1861	&pImage->ImageUuid);
1862	if (rc == VERR_VDI_VALUE_NOT_FOUND)
1863	{
1864	/* Image without UUID. Probably created by VMware and not yet used
1865	* by VirtualBox. Can only be added for images opened in read/write
1866	* mode, so don't bother producing a sensible UUID otherwise. */
1867	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
1868	RTUuidClear(&pImage->ImageUuid);
1869	else
1870	{
1871	rc = RTUuidCreate(&pImage->ImageUuid);
1872	if (RT_FAILURE(rc))
1873	return rc;
1874	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
1875	VMDK_DDB_IMAGE_UUID, &pImage->ImageUuid);
1876	if (RT_FAILURE(rc))
1877	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error storing image UUID in descriptor in '%s'"), pImage->pszFilename);
1878	}
1879	}
1880	else if (RT_FAILURE(rc))
1881	return rc;
1882
1883	/* Get image modification UUID. */
1884	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor,
1885	VMDK_DDB_MODIFICATION_UUID,
1886	&pImage->ModificationUuid);
1887	if (rc == VERR_VDI_VALUE_NOT_FOUND)
1888	{
1889	/* Image without UUID. Probably created by VMware and not yet used
1890	* by VirtualBox. Can only be added for images opened in read/write
1891	* mode, so don't bother producing a sensible UUID otherwise. */
1892	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
1893	RTUuidClear(&pImage->ModificationUuid);
1894	else
1895	{
1896	rc = RTUuidCreate(&pImage->ModificationUuid);
1897	if (RT_FAILURE(rc))
1898	return rc;
1899	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
1900	VMDK_DDB_MODIFICATION_UUID,
1901	&pImage->ModificationUuid);
1902	if (RT_FAILURE(rc))
1903	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error storing image modification UUID in descriptor in '%s'"), pImage->pszFilename);
1904	}
1905	}
1906	else if (RT_FAILURE(rc))
1907	return rc;
1908
1909	/* Get UUID of parent image. */
1910	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor, VMDK_DDB_PARENT_UUID,
1911	&pImage->ParentUuid);
1912	if (rc == VERR_VDI_VALUE_NOT_FOUND)
1913	{
1914	/* Image without UUID. Probably created by VMware and not yet used
1915	* by VirtualBox. Can only be added for images opened in read/write
1916	* mode, so don't bother producing a sensible UUID otherwise. */
1917	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
1918	RTUuidClear(&pImage->ParentUuid);
1919	else
1920	{
1921	rc = RTUuidClear(&pImage->ParentUuid);
1922	if (RT_FAILURE(rc))
1923	return rc;
1924	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
1925	VMDK_DDB_PARENT_UUID, &pImage->ParentUuid);
1926	if (RT_FAILURE(rc))
1927	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error storing parent UUID in descriptor in '%s'"), pImage->pszFilename);
1928	}
1929	}
1930	else if (RT_FAILURE(rc))
1931	return rc;
1932
1933	/* Get parent image modification UUID. */
1934	rc = vmdkDescDDBGetUuid(pImage, &pImage->Descriptor,
1935	VMDK_DDB_PARENT_MODIFICATION_UUID,
1936	&pImage->ParentModificationUuid);
1937	if (rc == VERR_VDI_VALUE_NOT_FOUND)
1938	{
1939	/* Image without UUID. Probably created by VMware and not yet used
1940	* by VirtualBox. Can only be added for images opened in read/write
1941	* mode, so don't bother producing a sensible UUID otherwise. */
1942	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
1943	RTUuidClear(&pImage->ParentModificationUuid);
1944	else
1945	{
1946	rc = RTUuidCreate(&pImage->ParentModificationUuid);
1947	if (RT_FAILURE(rc))
1948	return rc;
1949	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
1950	VMDK_DDB_PARENT_MODIFICATION_UUID,
1951	&pImage->ParentModificationUuid);
1952	if (RT_FAILURE(rc))
1953	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error storing parent modification UUID in descriptor in '%s'"), pImage->pszFilename);
1954	}
1955	}
1956	else if (RT_FAILURE(rc))
1957	return rc;
1958
1959	return VINF_SUCCESS;
1960	}
1961
1962	/**
1963	* Internal: write/update the descriptor part of the image.
1964	*/
1965	static int vmdkWriteDescriptor(PVMDKIMAGE pImage)
1966	{
1967	int rc = VINF_SUCCESS;
1968	uint64_t cbLimit;
1969	uint64_t uOffset;
1970	PVMDKFILE pDescFile;
1971
1972	if (pImage->pDescData)
1973	{
1974	/* Separate descriptor file. */
1975	uOffset = 0;
1976	cbLimit = 0;
1977	pDescFile = pImage->pFile;
1978	}
1979	else
1980	{
1981	/* Embedded descriptor file. */
1982	uOffset = VMDK_SECTOR2BYTE(pImage->pExtents[0].uDescriptorSector);
1983	cbLimit = VMDK_SECTOR2BYTE(pImage->pExtents[0].cDescriptorSectors);
1984	cbLimit += uOffset;
1985	pDescFile = pImage->pExtents[0].pFile;
1986	}
1987	for (unsigned i = 0; i < pImage->Descriptor.cLines; i++)
1988	{
1989	const char *psz = pImage->Descriptor.aLines[i];
1990	size_t cb = strlen(psz);
1991
1992	if (cbLimit && uOffset + cb + 1 > cbLimit)
1993	return vmdkError(pImage, VERR_BUFFER_OVERFLOW, RT_SRC_POS, N_("VMDK: descriptor too long in '%s'"), pImage->pszFilename);
1994	rc = vmdkFileWriteAt(pDescFile, uOffset, psz, cb, NULL);
1995	if (RT_FAILURE(rc))
1996	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error writing descriptor in '%s'"), pImage->pszFilename);
1997	uOffset += cb;
1998	rc = vmdkFileWriteAt(pDescFile, uOffset, "\n", 1, NULL);
1999	if (RT_FAILURE(rc))
2000	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error writing descriptor in '%s'"), pImage->pszFilename);
2001	uOffset++;
2002	}
2003	if (cbLimit)
2004	{
2005	/* Inefficient, but simple. */
2006	while (uOffset < cbLimit)
2007	{
2008	rc = vmdkFileWriteAt(pDescFile, uOffset, "", 1, NULL);
2009	if (RT_FAILURE(rc))
2010	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error writing descriptor in '%s'"), pImage->pszFilename);
2011	uOffset++;
2012	}
2013	}
2014	else
2015	{
2016	rc = vmdkFileSetSize(pDescFile, uOffset);
2017	if (RT_FAILURE(rc))
2018	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error truncating descriptor in '%s'"), pImage->pszFilename);
2019	}
2020	pImage->Descriptor.fDirty = false;
2021	return rc;
2022	}
2023
2024	/**
2025	* Internal: read metadata belonging to an extent with binary header, i.e.
2026	* as found in monolithic files.
2027	*/
2028	static int vmdkReadBinaryMetaExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
2029	{
2030	SparseExtentHeader Header;
2031	uint64_t cSectorsPerGDE;
2032
2033	int rc = vmdkFileReadAt(pExtent->pFile, 0, &Header, sizeof(Header), NULL);
2034	AssertRC(rc);
2035	if (RT_FAILURE(rc))
2036	{
2037	rc = vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: error reading extent header in '%s'"), pExtent->pszFullname);
2038	goto out;
2039	}
2040	if ( RT_LE2H_U32(Header.magicNumber) != VMDK_SPARSE_MAGICNUMBER
2041	\|\| RT_LE2H_U32(Header.version) != 1)
2042	{
2043	rc = vmdkError(pExtent->pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrect magic/version in extent header in '%s'"), pExtent->pszFullname);
2044	goto out;
2045	}
2046	if ( (RT_LE2H_U32(Header.flags) & 1)
2047	&& ( Header.singleEndLineChar != '\n'
2048	\|\| Header.nonEndLineChar != ' '
2049	\|\| Header.doubleEndLineChar1 != '\r'
2050	\|\| Header.doubleEndLineChar2 != '\n') )
2051	{
2052	rc = vmdkError(pExtent->pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: corrupted by CR/LF translation in '%s'"), pExtent->pszFullname);
2053	goto out;
2054	}
2055	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE; /* Just dummy value, changed later. */
2056	pExtent->cSectors = RT_LE2H_U64(Header.capacity);
2057	pExtent->cSectorsPerGrain = RT_LE2H_U64(Header.grainSize);
2058	pExtent->uDescriptorSector = RT_LE2H_U64(Header.descriptorOffset);
2059	pExtent->cDescriptorSectors = RT_LE2H_U64(Header.descriptorSize);
2060	if (pExtent->uDescriptorSector && !pExtent->cDescriptorSectors)
2061	{
2062	rc = vmdkError(pExtent->pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: inconsistent embedded descriptor config in '%s'"), pExtent->pszFullname);
2063	goto out;
2064	}
2065	pExtent->cGTEntries = RT_LE2H_U32(Header.numGTEsPerGT);
2066	if (RT_LE2H_U32(Header.flags) & 2)
2067	{
2068	pExtent->uSectorRGD = RT_LE2H_U64(Header.rgdOffset);
2069	pExtent->uSectorGD = RT_LE2H_U64(Header.gdOffset);
2070	}
2071	else
2072	{
2073	/** @todo this is just guesswork, the spec doesn't document this
2074	* properly and I don't have a vmdk without RGD. */
2075	pExtent->uSectorGD = RT_LE2H_U64(Header.rgdOffset);
2076	pExtent->uSectorRGD = 0;
2077	}
2078	pExtent->cOverheadSectors = RT_LE2H_U64(Header.overHead);
2079	pExtent->fUncleanShutdown = !!Header.uncleanShutdown;
2080	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
2081	if (!cSectorsPerGDE \|\| cSectorsPerGDE > UINT32_MAX)
2082	{
2083	rc = vmdkError(pExtent->pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: incorrect grain directory size in '%s'"), pExtent->pszFullname);
2084	goto out;
2085	}
2086	pExtent->cSectorsPerGDE = cSectorsPerGDE;
2087	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
2088
2089	/* Fix up the number of descriptor sectors, as some flat images have
2090	* really just one, and this causes failures when inserting the UUID
2091	* values and other extra information. */
2092	if (pExtent->cDescriptorSectors != 0 && pExtent->cDescriptorSectors < 4)
2093	{
2094	/* Do it the easy way - just fix it for flat images which have no
2095	* other complicated metadata which needs space too. */
2096	if ( pExtent->uDescriptorSector + 4 < pExtent->cOverheadSectors
2097	&& pExtent->cGTEntries * pExtent->cGDEntries == 0)
2098	pExtent->cDescriptorSectors = 4;
2099	}
2100
2101	out:
2102	if (RT_FAILURE(rc))
2103	vmdkFreeExtentData(pImage, pExtent, false);
2104
2105	return rc;
2106	}
2107
2108	/**
2109	* Internal: read additional metadata belonging to an extent. For those
2110	* extents which have no additional metadata just verify the information.
2111	*/
2112	static int vmdkReadMetaExtent(PVMDKIMAGE pImage, PVMDKEXTENT pExtent)
2113	{
2114	int rc = VINF_SUCCESS;
2115	uint64_t cbExtentSize;
2116
2117	/* The image must be a multiple of a sector in size and contain the data
2118	* area (flat images only). If not, it means the image is at least
2119	* truncated, or even seriously garbled. */
2120	rc = vmdkFileGetSize(pExtent->pFile, &cbExtentSize);
2121	if (RT_FAILURE(rc))
2122	{
2123	rc = vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: error getting size in '%s'"), pExtent->pszFullname);
2124	goto out;
2125	}
2126	/* disabled the size check again as there are too many too short vmdks out there */
2127	#ifdef VBOX_WITH_VMDK_STRICT_SIZE_CHECK
2128	if ( cbExtentSize != RT_ALIGN_64(cbExtentSize, 512)
2129	&& (pExtent->enmType != VMDKETYPE_FLAT \|\| pExtent->cNominalSectors + pExtent->uSectorOffset > VMDK_BYTE2SECTOR(cbExtentSize)))
2130	{
2131	rc = vmdkError(pExtent->pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: file size is not a multiple of 512 in '%s', file is truncated or otherwise garbled"), pExtent->pszFullname);
2132	goto out;
2133	}
2134	#endif /* VBOX_WITH_VMDK_STRICT_SIZE_CHECK */
2135	if (pExtent->enmType != VMDKETYPE_HOSTED_SPARSE)
2136	goto out;
2137
2138	/* The spec says that this must be a power of two and greater than 8,
2139	* but probably they meant not less than 8. */
2140	if ( (pExtent->cSectorsPerGrain & (pExtent->cSectorsPerGrain - 1))
2141	\|\| pExtent->cSectorsPerGrain < 8)
2142	{
2143	rc = vmdkError(pExtent->pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: invalid extent grain size %u in '%s'"), pExtent->cSectorsPerGrain, pExtent->pszFullname);
2144	goto out;
2145	}
2146
2147	/* This code requires that a grain table must hold a power of two multiple
2148	* of the number of entries per GT cache entry. */
2149	if ( (pExtent->cGTEntries & (pExtent->cGTEntries - 1))
2150	\|\| pExtent->cGTEntries < VMDK_GT_CACHELINE_SIZE)
2151	{
2152	rc = vmdkError(pExtent->pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: grain table cache size problem in '%s'"), pExtent->pszFullname);
2153	goto out;
2154	}
2155
2156	rc = vmdkReadGrainDirectory(pExtent);
2157
2158	out:
2159	if (RT_FAILURE(rc))
2160	vmdkFreeExtentData(pImage, pExtent, false);
2161
2162	return rc;
2163	}
2164
2165	/**
2166	* Internal: write/update the metadata for a sparse extent.
2167	*/
2168	static int vmdkWriteMetaSparseExtent(PVMDKEXTENT pExtent)
2169	{
2170	SparseExtentHeader Header;
2171
2172	memset(&Header, '\0', sizeof(Header));
2173	Header.magicNumber = RT_H2LE_U32(VMDK_SPARSE_MAGICNUMBER);
2174	Header.version = RT_H2LE_U32(1);
2175	Header.flags = RT_H2LE_U32(1 \| ((pExtent->pRGD) ? 2 : 0));
2176	Header.capacity = RT_H2LE_U64(pExtent->cSectors);
2177	Header.grainSize = RT_H2LE_U64(pExtent->cSectorsPerGrain);
2178	Header.descriptorOffset = RT_H2LE_U64(pExtent->uDescriptorSector);
2179	Header.descriptorSize = RT_H2LE_U64(pExtent->cDescriptorSectors);
2180	Header.numGTEsPerGT = RT_H2LE_U32(pExtent->cGTEntries);
2181	if (pExtent->pRGD)
2182	{
2183	Assert(pExtent->uSectorRGD);
2184	Header.rgdOffset = RT_H2LE_U64(pExtent->uSectorRGD);
2185	Header.gdOffset = RT_H2LE_U64(pExtent->uSectorGD);
2186	}
2187	else
2188	{
2189	/** @todo this is just guesswork, the spec doesn't document this
2190	* properly and I don't have a vmdk without RGD. */
2191	Header.rgdOffset = RT_H2LE_U64(pExtent->uSectorGD);
2192	}
2193	Header.overHead = RT_H2LE_U64(pExtent->cOverheadSectors);
2194	Header.uncleanShutdown = pExtent->fUncleanShutdown;
2195	Header.singleEndLineChar = '\n';
2196	Header.nonEndLineChar = ' ';
2197	Header.doubleEndLineChar1 = '\r';
2198	Header.doubleEndLineChar2 = '\n';
2199
2200	int rc = vmdkFileWriteAt(pExtent->pFile, 0, &Header, sizeof(Header), NULL);
2201	AssertRC(rc);
2202	if (RT_FAILURE(rc))
2203	rc = vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: error writing extent header in '%s'"), pExtent->pszFullname);
2204	return rc;
2205	}
2206
2207	#ifdef VBOX_WITH_VMDK_ESX
2208	/**
2209	* Internal: unused code to read the metadata of a sparse ESX extent.
2210	*
2211	* Such extents never leave ESX server, so this isn't ever used.
2212	*/
2213	static int vmdkReadMetaESXSparseExtent(PVMDKEXTENT pExtent)
2214	{
2215	COWDisk_Header Header;
2216	uint64_t cSectorsPerGDE;
2217
2218	int rc = vmdkFileReadAt(pExtent->pFile, 0, &Header, sizeof(Header), NULL);
2219	AssertRC(rc);
2220	if (RT_FAILURE(rc))
2221	goto out;
2222	if ( RT_LE2H_U32(Header.magicNumber) != VMDK_ESX_SPARSE_MAGICNUMBER
2223	\|\| RT_LE2H_U32(Header.version) != 1
2224	\|\| RT_LE2H_U32(Header.flags) != 3)
2225	{
2226	rc = VERR_VDI_INVALID_HEADER;
2227	goto out;
2228	}
2229	pExtent->enmType = VMDKETYPE_ESX_SPARSE;
2230	pExtent->cSectors = RT_LE2H_U32(Header.numSectors);
2231	pExtent->cSectorsPerGrain = RT_LE2H_U32(Header.grainSize);
2232	/* The spec says that this must be between 1 sector and 1MB. This code
2233	* assumes it's a power of two, so check that requirement, too. */
2234	if ( (pExtent->cSectorsPerGrain & (pExtent->cSectorsPerGrain - 1))
2235	\|\| pExtent->cSectorsPerGrain == 0
2236	\|\| pExtent->cSectorsPerGrain > 2048)
2237	{
2238	rc = VERR_VDI_INVALID_HEADER;
2239	goto out;
2240	}
2241	pExtent->uDescriptorSector = 0;
2242	pExtent->cDescriptorSectors = 0;
2243	pExtent->uSectorGD = RT_LE2H_U32(Header.gdOffset);
2244	pExtent->uSectorRGD = 0;
2245	pExtent->cOverheadSectors = 0;
2246	pExtent->cGTEntries = 4096;
2247	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
2248	if (!cSectorsPerGDE \|\| cSectorsPerGDE > UINT32_MAX)
2249	{
2250	rc = VERR_VDI_INVALID_HEADER;
2251	goto out;
2252	}
2253	pExtent->cSectorsPerGDE = cSectorsPerGDE;
2254	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
2255	if (pExtent->cGDEntries != RT_LE2H_U32(Header.numGDEntries))
2256	{
2257	/* Inconsistency detected. Computed number of GD entries doesn't match
2258	* stored value. Better be safe than sorry. */
2259	rc = VERR_VDI_INVALID_HEADER;
2260	goto out;
2261	}
2262	pExtent->uFreeSector = RT_LE2H_U32(Header.freeSector);
2263	pExtent->fUncleanShutdown = !!Header.uncleanShutdown;
2264
2265	rc = vmdkReadGrainDirectory(pExtent);
2266
2267	out:
2268	if (RT_FAILURE(rc))
2269	vmdkFreeExtentData(pImage, pExtent, false);
2270
2271	return rc;
2272	}
2273	#endif /* VBOX_WITH_VMDK_ESX */
2274
2275	/**
2276	* Internal: free the memory used by the extent data structure, optionally
2277	* deleting the referenced files.
2278	*/
2279	static void vmdkFreeExtentData(PVMDKIMAGE pImage, PVMDKEXTENT pExtent,
2280	bool fDelete)
2281	{
2282	vmdkFreeGrainDirectory(pExtent);
2283	if (pExtent->pDescData)
2284	{
2285	RTMemFree(pExtent->pDescData);
2286	pExtent->pDescData = NULL;
2287	}
2288	if (pExtent->pFile != NULL)
2289	{
2290	vmdkFileClose(pImage, &pExtent->pFile,
2291	fDelete
2292	&& pExtent->pszFullname
2293	&& strcmp(pExtent->pszFullname, pExtent->pszBasename));
2294	}
2295	if (pExtent->pszBasename)
2296	{
2297	RTMemTmpFree((void *)pExtent->pszBasename);
2298	pExtent->pszBasename = NULL;
2299	}
2300	if (pExtent->pszFullname)
2301	{
2302	RTStrFree((char )(void )pExtent->pszFullname);
2303	pExtent->pszFullname = NULL;
2304	}
2305	}
2306
2307	/**
2308	* Internal: allocate grain table cache if necessary for this image.
2309	*/
2310	static int vmdkAllocateGrainTableCache(PVMDKIMAGE pImage)
2311	{
2312	PVMDKEXTENT pExtent;
2313
2314	/* Allocate grain table cache if any sparse extent is present. */
2315	for (unsigned i = 0; i < pImage->cExtents; i++)
2316	{
2317	pExtent = &pImage->pExtents[i];
2318	if ( pExtent->enmType == VMDKETYPE_HOSTED_SPARSE
2319	#ifdef VBOX_WITH_VMDK_ESX
2320	\|\| pExtent->enmType == VMDKETYPE_ESX_SPARSE
2321	#endif /* VBOX_WITH_VMDK_ESX */
2322	)
2323	{
2324	/* Allocate grain table cache. */
2325	pImage->pGTCache = (PVMDKGTCACHE)RTMemAllocZ(sizeof(VMDKGTCACHE));
2326	if (!pImage->pGTCache)
2327	return VERR_NO_MEMORY;
2328	for (unsigned i = 0; i < VMDK_GT_CACHE_SIZE; i++)
2329	{
2330	PVMDKGTCACHEENTRY pGCE = &pImage->pGTCache->aGTCache[i];
2331	pGCE->uExtent = UINT32_MAX;
2332	}
2333	pImage->pGTCache->cEntries = VMDK_GT_CACHE_SIZE;
2334	break;
2335	}
2336	}
2337
2338	return VINF_SUCCESS;
2339	}
2340
2341	/**
2342	* Internal: allocate the given number of extents.
2343	*/
2344	static int vmdkCreateExtents(PVMDKIMAGE pImage, unsigned cExtents)
2345	{
2346	int rc = VINF_SUCCESS;
2347	PVMDKEXTENT pExtents = (PVMDKEXTENT)RTMemAllocZ(cExtents * sizeof(VMDKEXTENT));
2348	if (pImage)
2349	{
2350	for (unsigned i = 0; i < cExtents; i++)
2351	{
2352	pExtents[i].pFile = NULL;
2353	pExtents[i].pszBasename = NULL;
2354	pExtents[i].pszFullname = NULL;
2355	pExtents[i].pGD = NULL;
2356	pExtents[i].pRGD = NULL;
2357	pExtents[i].pDescData = NULL;
2358	pExtents[i].uExtent = i;
2359	pExtents[i].pImage = pImage;
2360	}
2361	pImage->pExtents = pExtents;
2362	pImage->cExtents = cExtents;
2363	}
2364	else
2365	rc = VERR_NO_MEMORY;
2366
2367	return rc;
2368	}
2369
2370	/**
2371	* Internal: Open an image, constructing all necessary data structures.
2372	*/
2373	static int vmdkOpenImage(PVMDKIMAGE pImage, unsigned uOpenFlags)
2374	{
2375	int rc;
2376	uint32_t u32Magic;
2377	PVMDKFILE pFile;
2378	PVMDKEXTENT pExtent;
2379
2380	pImage->uOpenFlags = uOpenFlags;
2381
2382	/* Try to get error interface. */
2383	pImage->pInterfaceError = VDInterfaceGet(pImage->pVDIfsDisk, VDINTERFACETYPE_ERROR);
2384	if (pImage->pInterfaceError)
2385	pImage->pInterfaceErrorCallbacks = VDGetInterfaceError(pImage->pInterfaceError);
2386
2387	/* Try to get async I/O interface. */
2388	pImage->pInterfaceAsyncIO = VDInterfaceGet(pImage->pVDIfsDisk, VDINTERFACETYPE_ASYNCIO);
2389	if (pImage->pInterfaceAsyncIO)
2390	pImage->pInterfaceAsyncIOCallbacks = VDGetInterfaceAsyncIO(pImage->pInterfaceAsyncIO);
2391
2392	/*
2393	* Open the image.
2394	* We don't have to check for asynchronous access because
2395	* we only support raw access and the opened file is a description
2396	* file were no data is stored.
2397	*/
2398	rc = vmdkFileOpen(pImage, &pFile, pImage->pszFilename,
2399	uOpenFlags & VD_OPEN_FLAGS_READONLY
2400	? RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE
2401	: RTFILE_O_READWRITE \| RTFILE_O_OPEN \| RTFILE_O_DENY_WRITE, false);
2402	if (RT_FAILURE(rc))
2403	{
2404	/* Do NOT signal an appropriate error here, as the VD layer has the
2405	* choice of retrying the open if it failed. */
2406	goto out;
2407	}
2408	pImage->pFile = pFile;
2409
2410	/* Read magic (if present). */
2411	rc = vmdkFileReadAt(pFile, 0, &u32Magic, sizeof(u32Magic), NULL);
2412	if (RT_FAILURE(rc))
2413	{
2414	rc = vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error reading the magic number in '%s'"), pImage->pszFilename);
2415	goto out;
2416	}
2417
2418	/* Handle the file according to its magic number. */
2419	if (RT_LE2H_U32(u32Magic) == VMDK_SPARSE_MAGICNUMBER)
2420	{
2421	/* It's a hosted single-extent image. */
2422	rc = vmdkCreateExtents(pImage, 1);
2423	if (RT_FAILURE(rc))
2424	goto out;
2425	/* The opened file is passed to the extent. No separate descriptor
2426	* file, so no need to keep anything open for the image. */
2427	pExtent = &pImage->pExtents[0];
2428	pExtent->pFile = pFile;
2429	pImage->pFile = NULL;
2430	pExtent->pszFullname = RTStrDup(pImage->pszFilename);
2431	if (!pExtent->pszFullname)
2432	{
2433	rc = VERR_NO_MEMORY;
2434	goto out;
2435	}
2436	rc = vmdkReadBinaryMetaExtent(pImage, pExtent);
2437	if (RT_FAILURE(rc))
2438	goto out;
2439	/* As we're dealing with a monolithic image here, there must
2440	* be a descriptor embedded in the image file. */
2441	if (!pExtent->uDescriptorSector \|\| !pExtent->cDescriptorSectors)
2442	{
2443	rc = vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: monolithic image without descriptor in '%s'"), pImage->pszFilename);
2444	goto out;
2445	}
2446	/* Read the descriptor from the extent. */
2447	pExtent->pDescData = (char *)RTMemAllocZ(VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
2448	if (!pExtent->pDescData)
2449	{
2450	rc = VERR_NO_MEMORY;
2451	goto out;
2452	}
2453	rc = vmdkFileReadAt(pExtent->pFile,
2454	VMDK_SECTOR2BYTE(pExtent->uDescriptorSector),
2455	pExtent->pDescData,
2456	VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors), NULL);
2457	AssertRC(rc);
2458	if (RT_FAILURE(rc))
2459	{
2460	rc = vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: read error for descriptor in '%s'"), pExtent->pszFullname);
2461	goto out;
2462	}
2463
2464	rc = vmdkParseDescriptor(pImage, pExtent->pDescData,
2465	VMDK_SECTOR2BYTE(pExtent->cDescriptorSectors));
2466	if (RT_FAILURE(rc))
2467	goto out;
2468
2469	rc = vmdkReadMetaExtent(pImage, pExtent);
2470	if (RT_FAILURE(rc))
2471	goto out;
2472
2473	/* Mark the extent as unclean if opened in read-write mode. */
2474	if (!(uOpenFlags & VD_OPEN_FLAGS_READONLY))
2475	{
2476	pExtent->fUncleanShutdown = true;
2477	pExtent->fMetaDirty = true;
2478	}
2479	}
2480	else
2481	{
2482	pImage->cbDescAlloc = VMDK_SECTOR2BYTE(20);
2483	pImage->pDescData = (char *)RTMemAllocZ(pImage->cbDescAlloc);
2484	if (!pImage->pDescData)
2485	{
2486	rc = VERR_NO_MEMORY;
2487	goto out;
2488	}
2489
2490	size_t cbRead;
2491	rc = vmdkFileReadAt(pImage->pFile, 0, pImage->pDescData,
2492	pImage->cbDescAlloc, &cbRead);
2493	if (RT_FAILURE(rc))
2494	{
2495	rc = vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: read error for descriptor in '%s'"), pImage->pszFilename);
2496	goto out;
2497	}
2498	if (cbRead == pImage->cbDescAlloc)
2499	{
2500	/* Likely the read is truncated. Better fail a bit too early
2501	* (normally the descriptor is much smaller than our buffer). */
2502	rc = vmdkError(pImage, VERR_VDI_INVALID_HEADER, RT_SRC_POS, N_("VMDK: cannot read descriptor in '%s'"), pImage->pszFilename);
2503	goto out;
2504	}
2505
2506	rc = vmdkParseDescriptor(pImage, pImage->pDescData,
2507	pImage->cbDescAlloc);
2508	if (RT_FAILURE(rc))
2509	goto out;
2510
2511	/*
2512	* We have to check for the asynchronous open flag. The
2513	* extents are parsed and the type of all are known now.
2514	* Check if every extent is either FLAT or ZERO.
2515	*/
2516	if (uOpenFlags & VD_OPEN_FLAGS_ASYNC_IO)
2517	{
2518	for (unsigned i = 0; i < pImage->cExtents; i++)
2519	{
2520	PVMDKEXTENT pExtent = &pImage->pExtents[i];
2521
2522	if ( (pExtent->enmType != VMDKETYPE_FLAT)
2523	&& (pExtent->enmType != VMDKETYPE_ZERO))
2524	{
2525	/*
2526	* Opened image contains at least one none flat or zero extent.
2527	* Return error but don't set error message as the caller
2528	* has the chance to open in non async I/O mode.
2529	*/
2530	rc = VERR_NOT_SUPPORTED;
2531	goto out;
2532	}
2533	}
2534	}
2535
2536	for (unsigned i = 0; i < pImage->cExtents; i++)
2537	{
2538	PVMDKEXTENT pExtent = &pImage->pExtents[i];
2539
2540	if (pExtent->pszBasename)
2541	{
2542	/* Hack to figure out whether the specified name in the
2543	* extent descriptor is absolute. Doesn't always work, but
2544	* should be good enough for now. */
2545	char *pszFullname;
2546	/** @todo implement proper path absolute check. */
2547	if (pExtent->pszBasename[0] == RTPATH_SLASH)
2548	{
2549	pszFullname = RTStrDup(pExtent->pszBasename);
2550	if (!pszFullname)
2551	{
2552	rc = VERR_NO_MEMORY;
2553	goto out;
2554	}
2555	}
2556	else
2557	{
2558	size_t cbDirname;
2559	char *pszDirname = RTStrDup(pImage->pszFilename);
2560	if (!pszDirname)
2561	{
2562	rc = VERR_NO_MEMORY;
2563	goto out;
2564	}
2565	RTPathStripFilename(pszDirname);
2566	cbDirname = strlen(pszDirname);
2567	rc = RTStrAPrintf(&pszFullname, "%s%c%s", pszDirname,
2568	RTPATH_SLASH, pExtent->pszBasename);
2569	RTStrFree(pszDirname);
2570	if (RT_FAILURE(rc))
2571	goto out;
2572	}
2573	pExtent->pszFullname = pszFullname;
2574	}
2575	else
2576	pExtent->pszFullname = NULL;
2577
2578	switch (pExtent->enmType)
2579	{
2580	case VMDKETYPE_HOSTED_SPARSE:
2581	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszFullname,
2582	uOpenFlags & VD_OPEN_FLAGS_READONLY
2583	? RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE
2584	: RTFILE_O_READWRITE \| RTFILE_O_OPEN \| RTFILE_O_DENY_WRITE, false);
2585	if (RT_FAILURE(rc))
2586	{
2587	/* Do NOT signal an appropriate error here, as the VD
2588	* layer has the choice of retrying the open if it
2589	* failed. */
2590	goto out;
2591	}
2592	rc = vmdkReadBinaryMetaExtent(pImage, pExtent);
2593	if (RT_FAILURE(rc))
2594	goto out;
2595	rc = vmdkReadMetaExtent(pImage, pExtent);
2596	if (RT_FAILURE(rc))
2597	goto out;
2598
2599	/* Mark extent as unclean if opened in read-write mode. */
2600	if (!(uOpenFlags & VD_OPEN_FLAGS_READONLY))
2601	{
2602	pExtent->fUncleanShutdown = true;
2603	pExtent->fMetaDirty = true;
2604	}
2605	break;
2606	case VMDKETYPE_FLAT:
2607	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszFullname,
2608	uOpenFlags & VD_OPEN_FLAGS_READONLY
2609	? RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE
2610	: RTFILE_O_READWRITE \| RTFILE_O_OPEN \| RTFILE_O_DENY_WRITE, true);
2611	if (RT_FAILURE(rc))
2612	{
2613	/* Do NOT signal an appropriate error here, as the VD
2614	* layer has the choice of retrying the open if it
2615	* failed. */
2616	goto out;
2617	}
2618	break;
2619	case VMDKETYPE_ZERO:
2620	/* Nothing to do. */
2621	break;
2622	default:
2623	AssertMsgFailed(("unknown vmdk extent type %d\n", pExtent->enmType));
2624	}
2625	}
2626	}
2627
2628	/* Make sure this is not reached accidentally with an error status. */
2629	AssertRC(rc);
2630
2631	/* Determine PCHS geometry if not set. */
2632	if (pImage->PCHSGeometry.cCylinders == 0)
2633	{
2634	uint64_t cCylinders = VMDK_BYTE2SECTOR(pImage->cbSize)
2635	/ pImage->PCHSGeometry.cHeads
2636	/ pImage->PCHSGeometry.cSectors;
2637	pImage->PCHSGeometry.cCylinders = (unsigned)RT_MIN(cCylinders, 16383);
2638	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
2639	{
2640	rc = vmdkDescSetPCHSGeometry(pImage, &pImage->PCHSGeometry);
2641	AssertRC(rc);
2642	}
2643	}
2644
2645	/* Update the image metadata now in case has changed. */
2646	rc = vmdkFlushImage(pImage);
2647	if (RT_FAILURE(rc))
2648	goto out;
2649
2650	/* Figure out a few per-image constants from the extents. */
2651	pImage->cbSize = 0;
2652	for (unsigned i = 0; i < pImage->cExtents; i++)
2653	{
2654	pExtent = &pImage->pExtents[i];
2655	if ( pExtent->enmType == VMDKETYPE_HOSTED_SPARSE
2656	#ifdef VBOX_WITH_VMDK_ESX
2657	\|\| pExtent->enmType == VMDKETYPE_ESX_SPARSE
2658	#endif /* VBOX_WITH_VMDK_ESX */
2659	)
2660	{
2661	/* Here used to be a check whether the nominal size of an extent
2662	* is a multiple of the grain size. The spec says that this is
2663	* always the case, but unfortunately some files out there in the
2664	* wild violate the spec (e.g. ReactOS 0.3.1). */
2665	}
2666	pImage->cbSize += VMDK_SECTOR2BYTE(pExtent->cNominalSectors);
2667	}
2668
2669	pImage->enmImageType = VD_IMAGE_TYPE_NORMAL;
2670	for (unsigned i = 0; i < pImage->cExtents; i++)
2671	{
2672	pExtent = &pImage->pExtents[i];
2673	if ( pImage->pExtents[i].enmType == VMDKETYPE_FLAT
2674	\|\| pImage->pExtents[i].enmType == VMDKETYPE_ZERO)
2675	{
2676	pImage->enmImageType = VD_IMAGE_TYPE_FIXED;
2677	break;
2678	}
2679	}
2680
2681	rc = vmdkAllocateGrainTableCache(pImage);
2682	if (RT_FAILURE(rc))
2683	goto out;
2684
2685	out:
2686	if (RT_FAILURE(rc))
2687	vmdkFreeImage(pImage, false);
2688	return rc;
2689	}
2690
2691	/**
2692	* Internal: create VMDK images for raw disk/partition access.
2693	*/
2694	static int vmdkCreateRawImage(PVMDKIMAGE pImage, const PVBOXHDDRAW pRaw,
2695	uint64_t cbSize)
2696	{
2697	int rc = VINF_SUCCESS;
2698	PVMDKEXTENT pExtent;
2699
2700	if (pRaw->fRawDisk)
2701	{
2702	/* Full raw disk access. This requires setting up a descriptor
2703	* file and open the (flat) raw disk. */
2704	rc = vmdkCreateExtents(pImage, 1);
2705	if (RT_FAILURE(rc))
2706	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
2707	pExtent = &pImage->pExtents[0];
2708	/* Create raw disk descriptor file. */
2709	rc = vmdkFileOpen(pImage, &pImage->pFile, pImage->pszFilename,
2710	RTFILE_O_READWRITE \| RTFILE_O_CREATE \| RTFILE_O_DENY_WRITE \| RTFILE_O_NOT_CONTENT_INDEXED,
2711	false);
2712	if (RT_FAILURE(rc))
2713	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pImage->pszFilename);
2714
2715	/* Set up basename for extent description. Cannot use StrDup. */
2716	size_t cbBasename = strlen(pRaw->pszRawDisk) + 1;
2717	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
2718	if (!pszBasename)
2719	return VERR_NO_MEMORY;
2720	memcpy(pszBasename, pRaw->pszRawDisk, cbBasename);
2721	pExtent->pszBasename = pszBasename;
2722	/* For raw disks the full name is identical to the base name. */
2723	pExtent->pszFullname = RTStrDup(pszBasename);
2724	if (!pExtent->pszFullname)
2725	return VERR_NO_MEMORY;
2726	pExtent->enmType = VMDKETYPE_FLAT;
2727	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize);
2728	pExtent->uSectorOffset = 0;
2729	pExtent->enmAccess = VMDKACCESS_READWRITE;
2730	pExtent->fMetaDirty = false;
2731
2732	/* Open flat image, the raw disk. */
2733	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszFullname,
2734	RTFILE_O_READWRITE \| RTFILE_O_OPEN \| RTFILE_O_DENY_WRITE, false);
2735	if (RT_FAILURE(rc))
2736	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not open raw disk file '%s'"), pExtent->pszFullname);
2737	}
2738	else
2739	{
2740	/* Raw partition access. This requires setting up a descriptor
2741	* file, write the partition information to a flat extent and
2742	* open all the (flat) raw disk partitions. */
2743
2744	/* First pass over the partitions to determine how many
2745	* extents we need. One partition can require up to 4 extents.
2746	* One to skip over unpartitioned space, one for the
2747	* partitioning data, one to skip over unpartitioned space
2748	* and one for the partition data. */
2749	unsigned cExtents = 0;
2750	uint64_t uStart = 0;
2751	for (unsigned i = 0; i < pRaw->cPartitions; i++)
2752	{
2753	PVBOXHDDRAWPART pPart = &pRaw->pPartitions[i];
2754	if (pPart->cbPartitionData)
2755	{
2756	if (uStart > pPart->uPartitionDataStart)
2757	return vmdkError(pImage, VERR_INVALID_PARAMETER, RT_SRC_POS, N_("VMDK: cannot go backwards for partitioning information in '%s'"), pImage->pszFilename);
2758	else if (uStart != pPart->uPartitionDataStart)
2759	cExtents++;
2760	uStart = pPart->uPartitionDataStart + pPart->cbPartitionData;
2761	cExtents++;
2762	}
2763	if (pPart->cbPartition)
2764	{
2765	if (uStart > pPart->uPartitionStart)
2766	return vmdkError(pImage, VERR_INVALID_PARAMETER, RT_SRC_POS, N_("VMDK: cannot go backwards for partition data in '%s'"), pImage->pszFilename);
2767	else if (uStart != pPart->uPartitionStart)
2768	cExtents++;
2769	uStart = pPart->uPartitionStart + pPart->cbPartition;
2770	cExtents++;
2771	}
2772	}
2773	/* Another extent for filling up the rest of the image. */
2774	if (uStart != cbSize)
2775	cExtents++;
2776
2777	rc = vmdkCreateExtents(pImage, cExtents);
2778	if (RT_FAILURE(rc))
2779	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
2780
2781	/* Create raw partition descriptor file. */
2782	rc = vmdkFileOpen(pImage, &pImage->pFile, pImage->pszFilename,
2783	RTFILE_O_READWRITE \| RTFILE_O_CREATE \| RTFILE_O_DENY_WRITE \| RTFILE_O_NOT_CONTENT_INDEXED,
2784	false);
2785	if (RT_FAILURE(rc))
2786	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pImage->pszFilename);
2787
2788	/* Create base filename for the partition table extent. */
2789	/** @todo remove fixed buffer without creating memory leaks. */
2790	char pszPartition[1024];
2791	const char *pszBase = RTPathFilename(pImage->pszFilename);
2792	const char *pszExt = RTPathExt(pszBase);
2793	if (pszExt == NULL)
2794	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: invalid filename '%s'"), pImage->pszFilename);
2795	char *pszBaseBase = RTStrDup(pszBase);
2796	if (!pszBaseBase)
2797	return VERR_NO_MEMORY;
2798	RTPathStripExt(pszBaseBase);
2799	RTStrPrintf(pszPartition, sizeof(pszPartition), "%s-pt%s",
2800	pszBaseBase, pszExt);
2801	RTStrFree(pszBaseBase);
2802
2803	/* Second pass over the partitions, now define all extents. */
2804	uint64_t uPartOffset = 0;
2805	cExtents = 0;
2806	uStart = 0;
2807	for (unsigned i = 0; i < pRaw->cPartitions; i++)
2808	{
2809	PVBOXHDDRAWPART pPart = &pRaw->pPartitions[i];
2810	if (pPart->cbPartitionData)
2811	{
2812	if (uStart != pPart->uPartitionDataStart)
2813	{
2814	pExtent = &pImage->pExtents[cExtents++];
2815	pExtent->pszBasename = NULL;
2816	pExtent->pszFullname = NULL;
2817	pExtent->enmType = VMDKETYPE_ZERO;
2818	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->uPartitionDataStart - uStart);
2819	pExtent->uSectorOffset = 0;
2820	pExtent->enmAccess = VMDKACCESS_READWRITE;
2821	pExtent->fMetaDirty = false;
2822	}
2823	uStart = pPart->uPartitionDataStart + pPart->cbPartitionData;
2824	pExtent = &pImage->pExtents[cExtents++];
2825	/* Set up basename for extent description. Can't use StrDup. */
2826	size_t cbBasename = strlen(pszPartition) + 1;
2827	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
2828	if (!pszBasename)
2829	return VERR_NO_MEMORY;
2830	memcpy(pszBasename, pszPartition, cbBasename);
2831	pExtent->pszBasename = pszBasename;
2832
2833	/* Set up full name for partition extent. */
2834	size_t cbDirname;
2835	char *pszDirname = RTStrDup(pImage->pszFilename);
2836	if (!pszDirname)
2837	return VERR_NO_MEMORY;
2838	RTPathStripFilename(pszDirname);
2839	cbDirname = strlen(pszDirname);
2840	char *pszFullname;
2841	rc = RTStrAPrintf(&pszFullname, "%s%c%s", pszDirname,
2842	RTPATH_SLASH, pExtent->pszBasename);
2843	RTStrFree(pszDirname);
2844	if (RT_FAILURE(rc))
2845	return rc;
2846	pExtent->pszFullname = pszFullname;
2847	pExtent->enmType = VMDKETYPE_FLAT;
2848	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbPartitionData);
2849	pExtent->uSectorOffset = uPartOffset;
2850	pExtent->enmAccess = VMDKACCESS_READWRITE;
2851	pExtent->fMetaDirty = false;
2852
2853	/* Create partition table flat image. */
2854	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszFullname,
2855	RTFILE_O_READWRITE \| RTFILE_O_CREATE \| RTFILE_O_DENY_WRITE \| RTFILE_O_NOT_CONTENT_INDEXED,
2856	false);
2857	if (RT_FAILURE(rc))
2858	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not create new partition data file '%s'"), pExtent->pszFullname);
2859	rc = vmdkFileWriteAt(pExtent->pFile,
2860	VMDK_SECTOR2BYTE(uPartOffset),
2861	pPart->pvPartitionData,
2862	pPart->cbPartitionData, NULL);
2863	if (RT_FAILURE(rc))
2864	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not write partition data to '%s'"), pExtent->pszFullname);
2865	uPartOffset += VMDK_BYTE2SECTOR(pPart->cbPartitionData);
2866	}
2867	if (pPart->cbPartition)
2868	{
2869	if (uStart != pPart->uPartitionStart)
2870	{
2871	pExtent = &pImage->pExtents[cExtents++];
2872	pExtent->pszBasename = NULL;
2873	pExtent->pszFullname = NULL;
2874	pExtent->enmType = VMDKETYPE_ZERO;
2875	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->uPartitionStart - uStart);
2876	pExtent->uSectorOffset = 0;
2877	pExtent->enmAccess = VMDKACCESS_READWRITE;
2878	pExtent->fMetaDirty = false;
2879	}
2880	uStart = pPart->uPartitionStart + pPart->cbPartition;
2881	pExtent = &pImage->pExtents[cExtents++];
2882	if (pPart->pszRawDevice)
2883	{
2884	/* Set up basename for extent descr. Can't use StrDup. */
2885	size_t cbBasename = strlen(pPart->pszRawDevice) + 1;
2886	char pszBasename = (char )RTMemTmpAlloc(cbBasename);
2887	if (!pszBasename)
2888	return VERR_NO_MEMORY;
2889	memcpy(pszBasename, pPart->pszRawDevice, cbBasename);
2890	pExtent->pszBasename = pszBasename;
2891	/* For raw disks full name is identical to base name. */
2892	pExtent->pszFullname = RTStrDup(pszBasename);
2893	if (!pExtent->pszFullname)
2894	return VERR_NO_MEMORY;
2895	pExtent->enmType = VMDKETYPE_FLAT;
2896	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbPartition);
2897	pExtent->uSectorOffset = VMDK_BYTE2SECTOR(pPart->uPartitionStartOffset);
2898	pExtent->enmAccess = VMDKACCESS_READWRITE;
2899	pExtent->fMetaDirty = false;
2900
2901	/* Open flat image, the raw partition. */
2902	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszFullname,
2903	RTFILE_O_READWRITE \| RTFILE_O_OPEN \| RTFILE_O_DENY_WRITE,
2904	false);
2905	if (RT_FAILURE(rc))
2906	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not open raw partition file '%s'"), pExtent->pszFullname);
2907	}
2908	else
2909	{
2910	pExtent->pszBasename = NULL;
2911	pExtent->pszFullname = NULL;
2912	pExtent->enmType = VMDKETYPE_ZERO;
2913	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(pPart->cbPartition);
2914	pExtent->uSectorOffset = 0;
2915	pExtent->enmAccess = VMDKACCESS_READWRITE;
2916	pExtent->fMetaDirty = false;
2917	}
2918	}
2919	}
2920	/* Another extent for filling up the rest of the image. */
2921	if (uStart != cbSize)
2922	{
2923	pExtent = &pImage->pExtents[cExtents++];
2924	pExtent->pszBasename = NULL;
2925	pExtent->pszFullname = NULL;
2926	pExtent->enmType = VMDKETYPE_ZERO;
2927	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbSize - uStart);
2928	pExtent->uSectorOffset = 0;
2929	pExtent->enmAccess = VMDKACCESS_READWRITE;
2930	pExtent->fMetaDirty = false;
2931	}
2932	}
2933
2934	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
2935	pRaw->fRawDisk ?
2936	"fullDevice" : "partitionedDevice");
2937	if (RT_FAILURE(rc))
2938	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
2939	return rc;
2940	}
2941
2942	/**
2943	* Internal: create a regular (i.e. file-backed) VMDK image.
2944	*/
2945	static int vmdkCreateRegularImage(PVMDKIMAGE pImage, VDIMAGETYPE enmType,
2946	uint64_t cbSize, unsigned uImageFlags,
2947	PFNVMPROGRESS pfnProgress, void *pvUser,
2948	unsigned uPercentStart, unsigned uPercentSpan)
2949	{
2950	int rc = VINF_SUCCESS;
2951	unsigned cExtents = 1;
2952	uint64_t cbOffset = 0;
2953	uint64_t cbRemaining = cbSize;
2954
2955	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
2956	{
2957	cExtents = cbSize / VMDK_2G_SPLIT_SIZE;
2958	/* Do proper extent computation: need one smaller extent if the total
2959	* size isn't evenly divisible by the split size. */
2960	if (cbSize % VMDK_2G_SPLIT_SIZE)
2961	cExtents++;
2962	}
2963	rc = vmdkCreateExtents(pImage, cExtents);
2964	if (RT_FAILURE(rc))
2965	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not create new extent list in '%s'"), pImage->pszFilename);
2966
2967	/* Basename strings needed for constructing the extent names. */
2968	char *pszBasenameSubstr = RTPathFilename(pImage->pszFilename);
2969	Assert(pszBasenameSubstr);
2970	size_t cbBasenameSubstr = strlen(pszBasenameSubstr) + 1;
2971
2972	/* Create searate descriptor file if necessary. */
2973	if (cExtents != 1 \|\| enmType == VD_IMAGE_TYPE_FIXED)
2974	{
2975	rc = vmdkFileOpen(pImage, &pImage->pFile, pImage->pszFilename,
2976	RTFILE_O_READWRITE \| RTFILE_O_CREATE \| RTFILE_O_DENY_WRITE \| RTFILE_O_NOT_CONTENT_INDEXED,
2977	false);
2978	if (RT_FAILURE(rc))
2979	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not create new sparse descriptor file '%s'"), pImage->pszFilename);
2980	pImage->pszFilename = RTStrDup(pImage->pszFilename);
2981	}
2982	else
2983	pImage->pFile = NULL;
2984
2985	/* Set up all extents. */
2986	for (unsigned i = 0; i < cExtents; i++)
2987	{
2988	PVMDKEXTENT pExtent = &pImage->pExtents[i];
2989	uint64_t cbExtent = cbRemaining;
2990
2991	/* Set up fullname/basename for extent description. Cannot use StrDup
2992	* for basename, as it is not guaranteed that the memory can be freed
2993	* with RTMemTmpFree, which must be used as in other code paths
2994	* StrDup is not usable. */
2995	if (cExtents == 1 && enmType != VD_IMAGE_TYPE_FIXED)
2996	{
2997	char pszBasename = (char )RTMemTmpAlloc(cbBasenameSubstr);
2998	if (!pszBasename)
2999	return VERR_NO_MEMORY;
3000	memcpy(pszBasename, pszBasenameSubstr, cbBasenameSubstr);
3001	pExtent->pszBasename = pszBasename;
3002	}
3003	else
3004	{
3005	char *pszBasenameExt = RTPathExt(pszBasenameSubstr);
3006	char *pszBasenameBase = RTStrDup(pszBasenameSubstr);
3007	RTPathStripExt(pszBasenameBase);
3008	char *pszTmp;
3009	size_t cbTmp;
3010	if (enmType == VD_IMAGE_TYPE_FIXED)
3011	{
3012	if (cExtents == 1)
3013	rc = RTStrAPrintf(&pszTmp, "%s-flat%s", pszBasenameBase,
3014	pszBasenameExt);
3015	else
3016	rc = RTStrAPrintf(&pszTmp, "%s-f%03d%s", pszBasenameBase,
3017	i+1, pszBasenameExt);
3018	}
3019	else
3020	rc = RTStrAPrintf(&pszTmp, "%s-s%03d%s", pszBasenameBase, i+1,
3021	pszBasenameExt);
3022	RTStrFree(pszBasenameBase);
3023	if (RT_FAILURE(rc))
3024	return rc;
3025	cbTmp = strlen(pszTmp) + 1;
3026	char pszBasename = (char )RTMemTmpAlloc(cbTmp);
3027	if (!pszBasename)
3028	return VERR_NO_MEMORY;
3029	memcpy(pszBasename, pszTmp, cbTmp);
3030	RTStrFree(pszTmp);
3031	pExtent->pszBasename = pszBasename;
3032	if (uImageFlags & VD_VMDK_IMAGE_FLAGS_SPLIT_2G)
3033	cbExtent = RT_MIN(cbRemaining, VMDK_2G_SPLIT_SIZE);
3034	}
3035	char *pszBasedirectory = RTStrDup(pImage->pszFilename);
3036	RTPathStripFilename(pszBasedirectory);
3037	char *pszFullname;
3038	rc = RTStrAPrintf(&pszFullname, "%s%c%s", pszBasedirectory,
3039	RTPATH_SLASH, pExtent->pszBasename);
3040	RTStrFree(pszBasedirectory);
3041	if (RT_FAILURE(rc))
3042	return rc;
3043	pExtent->pszFullname = pszFullname;
3044
3045	/* Create file for extent. */
3046	rc = vmdkFileOpen(pImage, &pExtent->pFile, pExtent->pszFullname,
3047	RTFILE_O_READWRITE \| RTFILE_O_CREATE \| RTFILE_O_DENY_WRITE \| RTFILE_O_NOT_CONTENT_INDEXED,
3048	false);
3049	if (RT_FAILURE(rc))
3050	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not create new file '%s'"), pExtent->pszFullname);
3051	if (enmType == VD_IMAGE_TYPE_FIXED)
3052	{
3053	rc = vmdkFileSetSize(pExtent->pFile, cbExtent);
3054	if (RT_FAILURE(rc))
3055	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not set size of new file '%s'"), pExtent->pszFullname);
3056
3057	/* Fill image with zeroes. We do this for every fixed-size image since on some systems
3058	* (for example Windows Vista), it takes ages to write a block near the end of a sparse
3059	* file and the guest could complain about an ATA timeout. */
3060
3061	/** @todo Starting with Linux 2.6.23, there is an fallocate() system call.
3062	* Currently supported file systems are ext4 and ocfs2. */
3063
3064	/* Allocate a temporary zero-filled buffer. Use a bigger block size to optimize writing */
3065	const size_t cbBuf = 128 * _1K;
3066	void *pvBuf = RTMemTmpAllocZ(cbBuf);
3067	if (!pvBuf)
3068	return VERR_NO_MEMORY;
3069
3070	uint64_t uOff = 0;
3071	/* Write data to all image blocks. */
3072	while (uOff < cbExtent)
3073	{
3074	unsigned cbChunk = (unsigned)RT_MIN(cbExtent, cbBuf);
3075
3076	rc = vmdkFileWriteAt(pExtent->pFile, uOff, pvBuf, cbChunk, NULL);
3077	if (RT_FAILURE(rc))
3078	{
3079	RTMemFree(pvBuf);
3080	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: writing block failed for '%s'"), pImage->pszFilename);
3081	}
3082
3083	uOff += cbChunk;
3084
3085	if (pfnProgress)
3086	{
3087	rc = pfnProgress(NULL /* WARNING! pVM=NULL */,
3088	uPercentStart + uOff * uPercentSpan / cbExtent,
3089	pvUser);
3090	if (RT_FAILURE(rc))
3091	{
3092	RTMemFree(pvBuf);
3093	return rc;
3094	}
3095	}
3096	}
3097	RTMemTmpFree(pvBuf);
3098	}
3099
3100	/* Place descriptor file information (where integrated). */
3101	if (cExtents == 1 && enmType != VD_IMAGE_TYPE_FIXED)
3102	{
3103	pExtent->uDescriptorSector = 1;
3104	pExtent->cDescriptorSectors = VMDK_BYTE2SECTOR(pImage->cbDescAlloc);
3105	/* The descriptor is part of the (only) extent. */
3106	pExtent->pDescData = pImage->pDescData;
3107	pImage->pDescData = NULL;
3108	}
3109
3110	if (enmType == VD_IMAGE_TYPE_NORMAL)
3111	{
3112	uint64_t cSectorsPerGDE, cSectorsPerGD;
3113	pExtent->enmType = VMDKETYPE_HOSTED_SPARSE;
3114	pExtent->cSectors = VMDK_BYTE2SECTOR(RT_ALIGN_64(cbExtent, 65536));
3115	pExtent->cSectorsPerGrain = VMDK_BYTE2SECTOR(65536);
3116	pExtent->cGTEntries = 512;
3117	cSectorsPerGDE = pExtent->cGTEntries * pExtent->cSectorsPerGrain;
3118	pExtent->cSectorsPerGDE = cSectorsPerGDE;
3119	pExtent->cGDEntries = (pExtent->cSectors + cSectorsPerGDE - 1) / cSectorsPerGDE;
3120	cSectorsPerGD = (pExtent->cGDEntries + (512 / sizeof(uint32_t) - 1)) / (512 / sizeof(uint32_t));
3121	}
3122	else
3123	pExtent->enmType = VMDKETYPE_FLAT;
3124
3125	pExtent->enmAccess = VMDKACCESS_READWRITE;
3126	pExtent->fUncleanShutdown = true;
3127	pExtent->cNominalSectors = VMDK_BYTE2SECTOR(cbExtent);
3128	pExtent->uSectorOffset = VMDK_BYTE2SECTOR(cbOffset);
3129	pExtent->fMetaDirty = true;
3130
3131	if (enmType == VD_IMAGE_TYPE_NORMAL)
3132	{
3133	rc = vmdkCreateGrainDirectory(pExtent,
3134	RT_MAX( pExtent->uDescriptorSector
3135	+ pExtent->cDescriptorSectors,
3136	1),
3137	true);
3138	if (RT_FAILURE(rc))
3139	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not create new grain directory in '%s'"), pExtent->pszFullname);
3140	}
3141
3142	if (RT_SUCCESS(rc) && pfnProgress)
3143	pfnProgress(NULL /* WARNING! pVM=NULL */,
3144	uPercentStart + i * uPercentSpan / cExtents,
3145	pvUser);
3146
3147	cbRemaining -= cbExtent;
3148	cbOffset += cbExtent;
3149	}
3150
3151	const char *pszDescType = NULL;
3152	if (enmType == VD_IMAGE_TYPE_FIXED)
3153	{
3154	pszDescType = (cExtents == 1)
3155	? "monolithicFlat" : "twoGbMaxExtentFlat";
3156	}
3157	else if (enmType == VD_IMAGE_TYPE_NORMAL)
3158	{
3159	pszDescType = (cExtents == 1)
3160	? "monolithicSparse" : "twoGbMaxExtentSparse";
3161	}
3162	else
3163	AssertMsgFailed(("invalid image type %d\n", enmType));
3164	rc = vmdkDescBaseSetStr(pImage, &pImage->Descriptor, "createType",
3165	pszDescType);
3166	if (RT_FAILURE(rc))
3167	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not set the image type in '%s'"), pImage->pszFilename);
3168	return rc;
3169	}
3170
3171	/**
3172	* Internal: The actual code for creating any VMDK variant currently in
3173	* existence on hosted environments.
3174	*/
3175	static int vmdkCreateImage(PVMDKIMAGE pImage, VDIMAGETYPE enmType,
3176	uint64_t cbSize, unsigned uImageFlags,
3177	const char *pszComment,
3178	PCPDMMEDIAGEOMETRY pPCHSGeometry,
3179	PCPDMMEDIAGEOMETRY pLCHSGeometry, PCRTUUID pUuid,
3180	PFNVMPROGRESS pfnProgress, void *pvUser,
3181	unsigned uPercentStart, unsigned uPercentSpan)
3182	{
3183	int rc;
3184
3185	pImage->uImageFlags = uImageFlags;
3186
3187	/* Try to get error interface. */
3188	pImage->pInterfaceError = VDInterfaceGet(pImage->pVDIfsDisk, VDINTERFACETYPE_ERROR);
3189	if (pImage->pInterfaceError)
3190	pImage->pInterfaceErrorCallbacks = VDGetInterfaceError(pImage->pInterfaceError);
3191
3192	/* Try to get async I/O interface. */
3193	pImage->pInterfaceAsyncIO = VDInterfaceGet(pImage->pVDIfsDisk, VDINTERFACETYPE_ASYNCIO);
3194	if (pImage->pInterfaceAsyncIO)
3195	pImage->pInterfaceAsyncIOCallbacks = VDGetInterfaceAsyncIO(pImage->pInterfaceAsyncIO);
3196
3197	rc = vmdkCreateDescriptor(pImage, pImage->pDescData, pImage->cbDescAlloc,
3198	&pImage->Descriptor);
3199	if (RT_FAILURE(rc))
3200	{
3201	rc = vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not create new descriptor in '%s'"), pImage->pszFilename);
3202	goto out;
3203	}
3204
3205	if ( enmType == VD_IMAGE_TYPE_FIXED
3206	&& (uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK))
3207	{
3208	/* Raw disk image (includes raw partition). */
3209	const PVBOXHDDRAW pRaw = (const PVBOXHDDRAW)pszComment;
3210	/* As the comment is misused, zap it so that no garbage comment
3211	* is set below. */
3212	pszComment = NULL;
3213	rc = vmdkCreateRawImage(pImage, pRaw, cbSize);
3214	}
3215	else if ( enmType == VD_IMAGE_TYPE_FIXED
3216	\|\| enmType == VD_IMAGE_TYPE_NORMAL)
3217	{
3218	/* Regular fixed or sparse image (monolithic or split). */
3219	rc = vmdkCreateRegularImage(pImage, enmType, cbSize, uImageFlags,
3220	pfnProgress, pvUser, uPercentStart,
3221	uPercentSpan * 95 / 100);
3222	}
3223	else
3224	{
3225	/* Unknown/invalid image type. */
3226	rc = VERR_NOT_IMPLEMENTED;
3227	}
3228
3229	if (RT_FAILURE(rc))
3230	goto out;
3231
3232	if (RT_SUCCESS(rc) && pfnProgress)
3233	pfnProgress(NULL /* WARNING! pVM=NULL */,
3234	uPercentStart + uPercentSpan * 98 / 100, pvUser);
3235
3236	pImage->enmImageType = enmType;
3237	pImage->cbSize = cbSize;
3238
3239	for (unsigned i = 0; i < pImage->cExtents; i++)
3240	{
3241	PVMDKEXTENT pExtent = &pImage->pExtents[i];
3242
3243	rc = vmdkDescExtInsert(pImage, &pImage->Descriptor, pExtent->enmAccess,
3244	pExtent->cNominalSectors, pExtent->enmType,
3245	pExtent->pszBasename, pExtent->uSectorOffset);
3246	if (RT_FAILURE(rc))
3247	{
3248	rc = vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: could not insert the extent list into descriptor in '%s'"), pImage->pszFilename);
3249	goto out;
3250	}
3251	}
3252	vmdkDescExtRemoveDummy(pImage, &pImage->Descriptor);
3253
3254	if ( pPCHSGeometry->cCylinders != 0
3255	&& pPCHSGeometry->cHeads != 0
3256	&& pPCHSGeometry->cSectors != 0)
3257	{
3258	rc = vmdkDescSetPCHSGeometry(pImage, pPCHSGeometry);
3259	if (RT_FAILURE(rc))
3260	goto out;
3261	}
3262	if ( pLCHSGeometry->cCylinders != 0
3263	&& pLCHSGeometry->cHeads != 0
3264	&& pLCHSGeometry->cSectors != 0)
3265	{
3266	rc = vmdkDescSetLCHSGeometry(pImage, pLCHSGeometry);
3267	if (RT_FAILURE(rc))
3268	goto out;
3269	}
3270
3271	pImage->LCHSGeometry = *pLCHSGeometry;
3272	pImage->PCHSGeometry = *pPCHSGeometry;
3273
3274	pImage->ImageUuid = *pUuid;
3275	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
3276	VMDK_DDB_IMAGE_UUID, &pImage->ImageUuid);
3277	if (RT_FAILURE(rc))
3278	{
3279	rc = vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error storing image UUID in new descriptor in '%s'"), pImage->pszFilename);
3280	goto out;
3281	}
3282	RTUuidClear(&pImage->ParentUuid);
3283	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
3284	VMDK_DDB_PARENT_UUID, &pImage->ParentUuid);
3285	if (RT_FAILURE(rc))
3286	{
3287	rc = vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error storing parent image UUID in new descriptor in '%s'"), pImage->pszFilename);
3288	goto out;
3289	}
3290	RTUuidClear(&pImage->ModificationUuid);
3291	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
3292	VMDK_DDB_MODIFICATION_UUID,
3293	&pImage->ModificationUuid);
3294	if (RT_FAILURE(rc))
3295	{
3296	rc = vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error storing modification UUID in new descriptor in '%s'"), pImage->pszFilename);
3297	goto out;
3298	}
3299	RTUuidClear(&pImage->ParentModificationUuid);
3300	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
3301	VMDK_DDB_PARENT_MODIFICATION_UUID,
3302	&pImage->ParentModificationUuid);
3303	if (RT_FAILURE(rc))
3304	{
3305	rc = vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error storing parent modification UUID in new descriptor in '%s'"), pImage->pszFilename);
3306	goto out;
3307	}
3308
3309	rc = vmdkAllocateGrainTableCache(pImage);
3310	if (RT_FAILURE(rc))
3311	goto out;
3312
3313	rc = vmdkSetImageComment(pImage, pszComment);
3314	if (RT_FAILURE(rc))
3315	{
3316	rc = vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: cannot set image comment in '%s'"), pImage->pszFilename);
3317	goto out;
3318	}
3319
3320	if (RT_SUCCESS(rc) && pfnProgress)
3321	pfnProgress(NULL /* WARNING! pVM=NULL */,
3322	uPercentStart + uPercentSpan * 99 / 100, pvUser);
3323
3324	rc = vmdkFlushImage(pImage);
3325
3326	out:
3327	if (RT_SUCCESS(rc) && pfnProgress)
3328	pfnProgress(NULL /* WARNING! pVM=NULL */,
3329	uPercentStart + uPercentSpan, pvUser);
3330
3331	if (RT_FAILURE(rc))
3332	vmdkFreeImage(pImage, rc != VERR_ALREADY_EXISTS);
3333	return rc;
3334	}
3335
3336	/**
3337	* Internal: Update image comment.
3338	*/
3339	static int vmdkSetImageComment(PVMDKIMAGE pImage, const char *pszComment)
3340	{
3341	char *pszCommentEncoded;
3342	if (pszComment)
3343	{
3344	pszCommentEncoded = vmdkEncodeString(pszComment);
3345	if (!pszCommentEncoded)
3346	return VERR_NO_MEMORY;
3347	}
3348	else
3349	pszCommentEncoded = NULL;
3350	int rc = vmdkDescDDBSetStr(pImage, &pImage->Descriptor,
3351	"ddb.comment", pszCommentEncoded);
3352	if (pszComment)
3353	RTStrFree(pszCommentEncoded);
3354	if (RT_FAILURE(rc))
3355	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error storing image comment in descriptor in '%s'"), pImage->pszFilename);
3356	return VINF_SUCCESS;
3357	}
3358
3359	/**
3360	* Internal. Free all allocated space for representing an image, and optionally
3361	* delete the image from disk.
3362	*/
3363	static void vmdkFreeImage(PVMDKIMAGE pImage, bool fDelete)
3364	{
3365	Assert(pImage);
3366
3367	if (pImage->enmImageType)
3368	{
3369	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
3370	{
3371	/* Mark all extents as clean. */
3372	for (unsigned i = 0; i < pImage->cExtents; i++)
3373	{
3374	if (( pImage->pExtents[i].enmType == VMDKETYPE_HOSTED_SPARSE
3375	#ifdef VBOX_WITH_VMDK_ESX
3376	\|\| pImage->pExtents[i].enmType == VMDKETYPE_ESX_SPARSE
3377	#endif /* VBOX_WITH_VMDK_ESX */
3378	)
3379	&& pImage->pExtents[i].fUncleanShutdown)
3380	{
3381	pImage->pExtents[i].fUncleanShutdown = false;
3382	pImage->pExtents[i].fMetaDirty = true;
3383	}
3384	}
3385	}
3386	(void)vmdkFlushImage(pImage);
3387	}
3388	if (pImage->pExtents != NULL)
3389	{
3390	for (unsigned i = 0 ; i < pImage->cExtents; i++)
3391	vmdkFreeExtentData(pImage, &pImage->pExtents[i], fDelete);
3392	RTMemFree(pImage->pExtents);
3393	pImage->pExtents = NULL;
3394	}
3395	if (pImage->pFile != NULL)
3396	vmdkFileClose(pImage, &pImage->pFile, fDelete);
3397	vmdkFileCheckAllClose(pImage);
3398	}
3399
3400	/**
3401	* Internal. Flush image data (and metadata) to disk.
3402	*/
3403	static int vmdkFlushImage(PVMDKIMAGE pImage)
3404	{
3405	PVMDKEXTENT pExtent;
3406	int rc = VINF_SUCCESS;
3407
3408	/* Update descriptor if changed. */
3409	if (pImage->Descriptor.fDirty)
3410	{
3411	rc = vmdkWriteDescriptor(pImage);
3412	if (RT_FAILURE(rc))
3413	goto out;
3414	}
3415
3416	for (unsigned i = 0; i < pImage->cExtents; i++)
3417	{
3418	pExtent = &pImage->pExtents[i];
3419	if (pExtent->pFile != NULL && pExtent->fMetaDirty)
3420	{
3421	switch (pExtent->enmType)
3422	{
3423	case VMDKETYPE_HOSTED_SPARSE:
3424	rc = vmdkWriteMetaSparseExtent(pExtent);
3425	if (RT_FAILURE(rc))
3426	goto out;
3427	break;
3428	#ifdef VBOX_WITH_VMDK_ESX
3429	case VMDKETYPE_ESX_SPARSE:
3430	/** @todo update the header. */
3431	break;
3432	#endif /* VBOX_WITH_VMDK_ESX */
3433	case VMDKETYPE_FLAT:
3434	/* Nothing to do. */
3435	break;
3436	case VMDKETYPE_ZERO:
3437	default:
3438	AssertMsgFailed(("extent with type %d marked as dirty\n",
3439	pExtent->enmType));
3440	break;
3441	}
3442	}
3443	switch (pExtent->enmType)
3444	{
3445	case VMDKETYPE_HOSTED_SPARSE:
3446	#ifdef VBOX_WITH_VMDK_ESX
3447	case VMDKETYPE_ESX_SPARSE:
3448	#endif /* VBOX_WITH_VMDK_ESX */
3449	case VMDKETYPE_FLAT:
3450	/** @todo implement proper path absolute check. */
3451	if ( pExtent->pFile != NULL
3452	&& !(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
3453	&& !(pExtent->pszBasename[0] == RTPATH_SLASH))
3454	rc = vmdkFileFlush(pExtent->pFile);
3455	break;
3456	case VMDKETYPE_ZERO:
3457	/* No need to do anything for this extent. */
3458	break;
3459	default:
3460	AssertMsgFailed(("unknown extent type %d\n", pExtent->enmType));
3461	break;
3462	}
3463	}
3464
3465	out:
3466	return rc;
3467	}
3468
3469	/**
3470	* Internal. Find extent corresponding to the sector number in the disk.
3471	*/
3472	static int vmdkFindExtent(PVMDKIMAGE pImage, uint64_t offSector,
3473	PVMDKEXTENT ppExtent, uint64_t puSectorInExtent)
3474	{
3475	PVMDKEXTENT pExtent = NULL;
3476	int rc = VINF_SUCCESS;
3477
3478	for (unsigned i = 0; i < pImage->cExtents; i++)
3479	{
3480	if (offSector < pImage->pExtents[i].cNominalSectors)
3481	{
3482	pExtent = &pImage->pExtents[i];
3483	*puSectorInExtent = offSector + pImage->pExtents[i].uSectorOffset;
3484	break;
3485	}
3486	offSector -= pImage->pExtents[i].cNominalSectors;
3487	}
3488
3489	if (pExtent)
3490	*ppExtent = pExtent;
3491	else
3492	rc = VERR_IO_SECTOR_NOT_FOUND;
3493
3494	return rc;
3495	}
3496
3497	/**
3498	* Internal. Hash function for placing the grain table hash entries.
3499	*/
3500	static uint32_t vmdkGTCacheHash(PVMDKGTCACHE pCache, uint64_t uSector,
3501	unsigned uExtent)
3502	{
3503	/** @todo this hash function is quite simple, maybe use a better one which
3504	* scrambles the bits better. */
3505	return (uSector + uExtent) % pCache->cEntries;
3506	}
3507
3508	/**
3509	* Internal. Get sector number in the extent file from the relative sector
3510	* number in the extent.
3511	*/
3512	static int vmdkGetSector(PVMDKGTCACHE pCache, PVMDKEXTENT pExtent,
3513	uint64_t uSector, uint64_t *puExtentSector)
3514	{
3515	uint64_t uGDIndex, uGTSector, uGTBlock;
3516	uint32_t uGTHash, uGTBlockIndex;
3517	PVMDKGTCACHEENTRY pGTCacheEntry;
3518	uint32_t aGTDataTmp[VMDK_GT_CACHELINE_SIZE];
3519	int rc;
3520
3521	uGDIndex = uSector / pExtent->cSectorsPerGDE;
3522	if (uGDIndex >= pExtent->cGDEntries)
3523	return VERR_OUT_OF_RANGE;
3524	uGTSector = pExtent->pGD[uGDIndex];
3525	if (!uGTSector)
3526	{
3527	/* There is no grain table referenced by this grain directory
3528	* entry. So there is absolutely no data in this area. */
3529	*puExtentSector = 0;
3530	return VINF_SUCCESS;
3531	}
3532
3533	uGTBlock = uSector / (pExtent->cSectorsPerGrain * VMDK_GT_CACHELINE_SIZE);
3534	uGTHash = vmdkGTCacheHash(pCache, uGTBlock, pExtent->uExtent);
3535	pGTCacheEntry = &pCache->aGTCache[uGTHash];
3536	if ( pGTCacheEntry->uExtent != pExtent->uExtent
3537	\|\| pGTCacheEntry->uGTBlock != uGTBlock)
3538	{
3539	/* Cache miss, fetch data from disk. */
3540	rc = vmdkFileReadAt(pExtent->pFile,
3541	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
3542	aGTDataTmp, sizeof(aGTDataTmp), NULL);
3543	if (RT_FAILURE(rc))
3544	return vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: cannot read grain table entry in '%s'"), pExtent->pszFullname);
3545	pGTCacheEntry->uExtent = pExtent->uExtent;
3546	pGTCacheEntry->uGTBlock = uGTBlock;
3547	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
3548	pGTCacheEntry->aGTData[i] = RT_LE2H_U32(aGTDataTmp[i]);
3549	}
3550	uGTBlockIndex = (uSector / pExtent->cSectorsPerGrain) % VMDK_GT_CACHELINE_SIZE;
3551	uint64_t uGrainSector = pGTCacheEntry->aGTData[uGTBlockIndex];
3552	if (uGrainSector)
3553	*puExtentSector = uGrainSector + uSector % pExtent->cSectorsPerGrain;
3554	else
3555	*puExtentSector = 0;
3556	return VINF_SUCCESS;
3557	}
3558
3559	/**
3560	* Internal. Allocates a new grain table (if necessary), writes the grain
3561	* and updates the grain table. The cache is also updated by this operation.
3562	* This is separate from vmdkGetSector, because that should be as fast as
3563	* possible. Most code from vmdkGetSector also appears here.
3564	*/
3565	static int vmdkAllocGrain(PVMDKGTCACHE pCache, PVMDKEXTENT pExtent,
3566	uint64_t uSector, const void *pvBuf,
3567	uint64_t cbWrite)
3568	{
3569	uint64_t uGDIndex, uGTSector, uRGTSector, uGTBlock;
3570	uint64_t cbExtentSize;
3571	uint32_t uGTHash, uGTBlockIndex;
3572	PVMDKGTCACHEENTRY pGTCacheEntry;
3573	uint32_t aGTDataTmp[VMDK_GT_CACHELINE_SIZE];
3574	int rc;
3575
3576	uGDIndex = uSector / pExtent->cSectorsPerGDE;
3577	if (uGDIndex >= pExtent->cGDEntries)
3578	return VERR_OUT_OF_RANGE;
3579	uGTSector = pExtent->pGD[uGDIndex];
3580	if (pExtent->pRGD)
3581	uRGTSector = pExtent->pRGD[uGDIndex];
3582	else
3583	uRGTSector = 0; /*< avoid compiler warning /
3584	if (!uGTSector)
3585	{
3586	/* There is no grain table referenced by this grain directory
3587	* entry. So there is absolutely no data in this area. Allocate
3588	* a new grain table and put the reference to it in the GDs. */
3589	rc = vmdkFileGetSize(pExtent->pFile, &cbExtentSize);
3590	if (RT_FAILURE(rc))
3591	return vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: error getting size in '%s'"), pExtent->pszFullname);
3592	Assert(!(cbExtentSize % 512));
3593	cbExtentSize = RT_ALIGN_64(cbExtentSize, 512);
3594	uGTSector = VMDK_BYTE2SECTOR(cbExtentSize);
3595	/* Normally the grain table is preallocated for hosted sparse extents
3596	* that support more than 32 bit sector numbers. So this shouldn't
3597	* ever happen on a valid extent. */
3598	if (uGTSector > UINT32_MAX)
3599	return VERR_VDI_INVALID_HEADER;
3600	/* Write grain table by writing the required number of grain table
3601	* cache chunks. Avoids dynamic memory allocation, but is a bit
3602	* slower. But as this is a pretty infrequently occurring case it
3603	* should be acceptable. */
3604	memset(aGTDataTmp, '\0', sizeof(aGTDataTmp));
3605	for (unsigned i = 0;
3606	i < pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE;
3607	i++)
3608	{
3609	rc = vmdkFileWriteAt(pExtent->pFile,
3610	VMDK_SECTOR2BYTE(uGTSector) + i * sizeof(aGTDataTmp),
3611	aGTDataTmp, sizeof(aGTDataTmp), NULL);
3612	if (RT_FAILURE(rc))
3613	return vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: cannot write grain table allocation in '%s'"), pExtent->pszFullname);
3614	}
3615	if (pExtent->pRGD)
3616	{
3617	AssertReturn(!uRGTSector, VERR_VDI_INVALID_HEADER);
3618	rc = vmdkFileGetSize(pExtent->pFile, &cbExtentSize);
3619	if (RT_FAILURE(rc))
3620	return vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: error getting size in '%s'"), pExtent->pszFullname);
3621	Assert(!(cbExtentSize % 512));
3622	uRGTSector = VMDK_BYTE2SECTOR(cbExtentSize);
3623	/* Normally the redundant grain table is preallocated for hosted
3624	* sparse extents that support more than 32 bit sector numbers. So
3625	* this shouldn't ever happen on a valid extent. */
3626	if (uRGTSector > UINT32_MAX)
3627	return VERR_VDI_INVALID_HEADER;
3628	/* Write backup grain table by writing the required number of grain
3629	* table cache chunks. Avoids dynamic memory allocation, but is a
3630	* bit slower. But as this is a pretty infrequently occurring case
3631	* it should be acceptable. */
3632	for (unsigned i = 0;
3633	i < pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE;
3634	i++)
3635	{
3636	rc = vmdkFileWriteAt(pExtent->pFile,
3637	VMDK_SECTOR2BYTE(uRGTSector) + i * sizeof(aGTDataTmp),
3638	aGTDataTmp, sizeof(aGTDataTmp), NULL);
3639	if (RT_FAILURE(rc))
3640	return vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: cannot write backup grain table allocation in '%s'"), pExtent->pszFullname);
3641	}
3642	}
3643
3644	/* Update the grain directory on disk (doing it before writing the
3645	* grain table will result in a garbled extent if the operation is
3646	* aborted for some reason. Otherwise the worst that can happen is
3647	* some unused sectors in the extent. */
3648	uint32_t uGTSectorLE = RT_H2LE_U64(uGTSector);
3649	rc = vmdkFileWriteAt(pExtent->pFile,
3650	VMDK_SECTOR2BYTE(pExtent->uSectorGD) + uGDIndex * sizeof(uGTSectorLE),
3651	&uGTSectorLE, sizeof(uGTSectorLE), NULL);
3652	if (RT_FAILURE(rc))
3653	return vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: cannot write grain directory entry in '%s'"), pExtent->pszFullname);
3654	if (pExtent->pRGD)
3655	{
3656	uint32_t uRGTSectorLE = RT_H2LE_U64(uRGTSector);
3657	rc = vmdkFileWriteAt(pExtent->pFile,
3658	VMDK_SECTOR2BYTE(pExtent->uSectorRGD) + uGDIndex * sizeof(uRGTSectorLE),
3659	&uRGTSectorLE, sizeof(uRGTSectorLE), NULL);
3660	if (RT_FAILURE(rc))
3661	return vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: cannot write backup grain directory entry in '%s'"), pExtent->pszFullname);
3662	}
3663
3664	/* As the final step update the in-memory copy of the GDs. */
3665	pExtent->pGD[uGDIndex] = uGTSector;
3666	if (pExtent->pRGD)
3667	pExtent->pRGD[uGDIndex] = uRGTSector;
3668	}
3669
3670	rc = vmdkFileGetSize(pExtent->pFile, &cbExtentSize);
3671	if (RT_FAILURE(rc))
3672	return vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: error getting size in '%s'"), pExtent->pszFullname);
3673	Assert(!(cbExtentSize % 512));
3674
3675	/* Write the data. */
3676	rc = vmdkFileWriteAt(pExtent->pFile, cbExtentSize, pvBuf, cbWrite, NULL);
3677	if (RT_FAILURE(rc))
3678	return vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: cannot write allocated data block in '%s'"), pExtent->pszFullname);
3679
3680	/* Update the grain table (and the cache). */
3681	uGTBlock = uSector / (pExtent->cSectorsPerGrain * VMDK_GT_CACHELINE_SIZE);
3682	uGTHash = vmdkGTCacheHash(pCache, uGTBlock, pExtent->uExtent);
3683	pGTCacheEntry = &pCache->aGTCache[uGTHash];
3684	if ( pGTCacheEntry->uExtent != pExtent->uExtent
3685	\|\| pGTCacheEntry->uGTBlock != uGTBlock)
3686	{
3687	/* Cache miss, fetch data from disk. */
3688	rc = vmdkFileReadAt(pExtent->pFile,
3689	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
3690	aGTDataTmp, sizeof(aGTDataTmp), NULL);
3691	if (RT_FAILURE(rc))
3692	return vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: cannot read allocated grain table entry in '%s'"), pExtent->pszFullname);
3693	pGTCacheEntry->uExtent = pExtent->uExtent;
3694	pGTCacheEntry->uGTBlock = uGTBlock;
3695	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
3696	pGTCacheEntry->aGTData[i] = RT_LE2H_U32(aGTDataTmp[i]);
3697	}
3698	else
3699	{
3700	/* Cache hit. Convert grain table block back to disk format, otherwise
3701	* the code below will write garbage for all but the updated entry. */
3702	for (unsigned i = 0; i < VMDK_GT_CACHELINE_SIZE; i++)
3703	aGTDataTmp[i] = RT_H2LE_U32(pGTCacheEntry->aGTData[i]);
3704	}
3705	uGTBlockIndex = (uSector / pExtent->cSectorsPerGrain) % VMDK_GT_CACHELINE_SIZE;
3706	aGTDataTmp[uGTBlockIndex] = RT_H2LE_U32(VMDK_BYTE2SECTOR(cbExtentSize));
3707	pGTCacheEntry->aGTData[uGTBlockIndex] = VMDK_BYTE2SECTOR(cbExtentSize);
3708	/* Update grain table on disk. */
3709	rc = vmdkFileWriteAt(pExtent->pFile,
3710	VMDK_SECTOR2BYTE(uGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
3711	aGTDataTmp, sizeof(aGTDataTmp), NULL);
3712	if (RT_FAILURE(rc))
3713	return vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: cannot write updated grain table in '%s'"), pExtent->pszFullname);
3714	if (pExtent->pRGD)
3715	{
3716	/* Update backup grain table on disk. */
3717	rc = vmdkFileWriteAt(pExtent->pFile,
3718	VMDK_SECTOR2BYTE(uRGTSector) + (uGTBlock % (pExtent->cGTEntries / VMDK_GT_CACHELINE_SIZE)) * sizeof(aGTDataTmp),
3719	aGTDataTmp, sizeof(aGTDataTmp), NULL);
3720	if (RT_FAILURE(rc))
3721	return vmdkError(pExtent->pImage, rc, RT_SRC_POS, N_("VMDK: cannot write updated backup grain table in '%s'"), pExtent->pszFullname);
3722	}
3723	#ifdef VBOX_WITH_VMDK_ESX
3724	if (RT_SUCCESS(rc) && pExtent->enmType == VMDKETYPE_ESX_SPARSE)
3725	{
3726	pExtent->uFreeSector = uGTSector + VMDK_BYTE2SECTOR(cbWrite);
3727	pExtent->fMetaDirty = true;
3728	}
3729	#endif /* VBOX_WITH_VMDK_ESX */
3730	return rc;
3731	}
3732
3733
3734	/** @copydoc VBOXHDDBACKEND::pfnCheckIfValid */
3735	static int vmdkCheckIfValid(const char *pszFilename)
3736	{
3737	LogFlowFunc(("pszFilename=\"%s\"\n", pszFilename));
3738	int rc = VINF_SUCCESS;
3739	PVMDKIMAGE pImage;
3740
3741	if ( !pszFilename
3742	\|\| !*pszFilename
3743	\|\| strchr(pszFilename, '"'))
3744	{
3745	rc = VERR_INVALID_PARAMETER;
3746	goto out;
3747	}
3748
3749	pImage = (PVMDKIMAGE)RTMemAllocZ(sizeof(VMDKIMAGE));
3750	if (!pImage)
3751	{
3752	rc = VERR_NO_MEMORY;
3753	goto out;
3754	}
3755	pImage->pszFilename = pszFilename;
3756	pImage->pFile = NULL;
3757	pImage->pExtents = NULL;
3758	pImage->pFiles = NULL;
3759	pImage->pGTCache = NULL;
3760	pImage->pDescData = NULL;
3761	pImage->pVDIfsDisk = NULL;
3762	/** @todo speed up this test open (VD_OPEN_FLAGS_INFO) by skipping as
3763	* much as possible in vmdkOpenImage. */
3764	rc = vmdkOpenImage(pImage, VD_OPEN_FLAGS_INFO \| VD_OPEN_FLAGS_READONLY);
3765	vmdkFreeImage(pImage, false);
3766
3767	out:
3768	LogFlowFunc(("returns %Rrc\n", rc));
3769	return rc;
3770	}
3771
3772	/** @copydoc VBOXHDDBACKEND::pfnOpen */
3773	static int vmdkOpen(const char *pszFilename, unsigned uOpenFlags,
3774	PVDINTERFACE pVDIfsDisk, PVDINTERFACE pVDIfsImage,
3775	void **ppBackendData)
3776	{
3777	LogFlowFunc(("pszFilename=\"%s\" uOpenFlags=%#x pVDIfsDisk=%#p pVDIfsImage=%#p ppBackendData=%#p\n", pszFilename, uOpenFlags, pVDIfsDisk, pVDIfsImage, ppBackendData));
3778	int rc;
3779	PVMDKIMAGE pImage;
3780
3781	/* Check open flags. All valid flags are supported. */
3782	if (uOpenFlags & ~VD_OPEN_FLAGS_MASK)
3783	{
3784	rc = VERR_INVALID_PARAMETER;
3785	goto out;
3786	}
3787
3788	/* Check remaining arguments. */
3789	if ( !VALID_PTR(pszFilename)
3790	\|\| !*pszFilename
3791	\|\| strchr(pszFilename, '"'))
3792	{
3793	rc = VERR_INVALID_PARAMETER;
3794	goto out;
3795	}
3796
3797
3798	pImage = (PVMDKIMAGE)RTMemAllocZ(sizeof(VMDKIMAGE));
3799	if (!pImage)
3800	{
3801	rc = VERR_NO_MEMORY;
3802	goto out;
3803	}
3804	pImage->pszFilename = pszFilename;
3805	pImage->pFile = NULL;
3806	pImage->pExtents = NULL;
3807	pImage->pFiles = NULL;
3808	pImage->pGTCache = NULL;
3809	pImage->pDescData = NULL;
3810	pImage->pVDIfsDisk = pVDIfsDisk;
3811
3812	rc = vmdkOpenImage(pImage, uOpenFlags);
3813	if (RT_SUCCESS(rc))
3814	*ppBackendData = pImage;
3815
3816	out:
3817	LogFlowFunc(("returns %Rrc (pBackendData=%#p)\n", rc, *ppBackendData));
3818	return rc;
3819	}
3820
3821	/** @copydoc VBOXHDDBACKEND::pfnCreate */
3822	static int vmdkCreate(const char *pszFilename, VDIMAGETYPE enmType,
3823	uint64_t cbSize, unsigned uImageFlags,
3824	const char *pszComment,
3825	PCPDMMEDIAGEOMETRY pPCHSGeometry,
3826	PCPDMMEDIAGEOMETRY pLCHSGeometry, PCRTUUID pUuid,
3827	unsigned uOpenFlags, unsigned uPercentStart,
3828	unsigned uPercentSpan, PVDINTERFACE pVDIfsDisk,
3829	PVDINTERFACE pVDIfsImage, PVDINTERFACE pVDIfsOperation,
3830	void **ppBackendData)
3831	{
3832	LogFlowFunc(("pszFilename=\"%s\" enmType=%d cbSize=%llu uImageFlags=%#x pszComment=\"%s\" pPCHSGeometry=%#p pLCHSGeometry=%#p Uuid=%RTuuid uOpenFlags=%#x uPercentStart=%u uPercentSpan=%u pVDIfsDisk=%#p pVDIfsImage=%#p pVDIfsOperation=%#p ppBackendData=%#p", pszFilename, enmType, cbSize, uImageFlags, pszComment, pPCHSGeometry, pLCHSGeometry, pUuid, uOpenFlags, uPercentStart, uPercentSpan, pVDIfsDisk, pVDIfsImage, pVDIfsOperation, ppBackendData));
3833	int rc;
3834	PVMDKIMAGE pImage;
3835
3836	PFNVMPROGRESS pfnProgress = NULL;
3837	void *pvUser = NULL;
3838	PVDINTERFACE pIfProgress = VDInterfaceGet(pVDIfsOperation,
3839	VDINTERFACETYPE_PROGRESS);
3840	PVDINTERFACEPROGRESS pCbProgress = NULL;
3841	if (pIfProgress)
3842	{
3843	pCbProgress = VDGetInterfaceProgress(pIfProgress);
3844	pfnProgress = pCbProgress->pfnProgress;
3845	pvUser = pIfProgress->pvUser;
3846	}
3847
3848	/* Check open flags. All valid flags are supported. */
3849	if (uOpenFlags & ~VD_OPEN_FLAGS_MASK)
3850	{
3851	rc = VERR_INVALID_PARAMETER;
3852	goto out;
3853	}
3854
3855	/* @todo A quick hack to support differencing images in VMDK. */
3856	if (enmType == VD_IMAGE_TYPE_DIFF)
3857	enmType = VD_IMAGE_TYPE_NORMAL;
3858
3859	/* Check remaining arguments. */
3860	if ( !VALID_PTR(pszFilename)
3861	\|\| !*pszFilename
3862	\|\| strchr(pszFilename, '"')
3863	\|\| (enmType != VD_IMAGE_TYPE_NORMAL && enmType != VD_IMAGE_TYPE_FIXED)
3864	\|\| !VALID_PTR(pPCHSGeometry)
3865	\|\| !VALID_PTR(pLCHSGeometry))
3866	{
3867	rc = VERR_INVALID_PARAMETER;
3868	goto out;
3869	}
3870
3871	pImage = (PVMDKIMAGE)RTMemAllocZ(sizeof(VMDKIMAGE));
3872	if (!pImage)
3873	{
3874	rc = VERR_NO_MEMORY;
3875	goto out;
3876	}
3877	pImage->pszFilename = pszFilename;
3878	pImage->pFile = NULL;
3879	pImage->pExtents = NULL;
3880	pImage->pFiles = NULL;
3881	pImage->pGTCache = NULL;
3882	pImage->pDescData = NULL;
3883	pImage->pVDIfsDisk = NULL;
3884	pImage->cbDescAlloc = VMDK_SECTOR2BYTE(20);
3885	pImage->pDescData = (char *)RTMemAllocZ(pImage->cbDescAlloc);
3886	if (!pImage->pDescData)
3887	{
3888	rc = VERR_NO_MEMORY;
3889	goto out;
3890	}
3891
3892	rc = vmdkCreateImage(pImage, enmType, cbSize, uImageFlags, pszComment,
3893	pPCHSGeometry, pLCHSGeometry, pUuid,
3894	pfnProgress, pvUser, uPercentStart, uPercentSpan);
3895	if (RT_SUCCESS(rc))
3896	{
3897	/* So far the image is opened in read/write mode. Make sure the
3898	* image is opened in read-only mode if the caller requested that. */
3899	if (uOpenFlags & VD_OPEN_FLAGS_READONLY)
3900	{
3901	vmdkFreeImage(pImage, false);
3902	rc = vmdkOpenImage(pImage, uOpenFlags);
3903	if (RT_FAILURE(rc))
3904	goto out;
3905	}
3906	*ppBackendData = pImage;
3907	}
3908
3909	out:
3910	LogFlowFunc(("returns %Rrc (pBackendData=%#p)\n", rc, *ppBackendData));
3911	return rc;
3912	}
3913
3914	/**
3915	* Replaces a fragment of a string with the specified string.
3916	*
3917	* @returns Pointer to the allocated UTF-8 string.
3918	* @param pszWhere UTF-8 string to search in.
3919	* @param pszWhat UTF-8 string to search for.
3920	* @param pszByWhat UTF-8 string to replace the found string with.
3921	*/
3922	static char * vmdkStrReplace(const char pszWhere, const char pszWhat, const char *pszByWhat)
3923	{
3924	Assert(VALID_PTR(pszWhere));
3925	Assert(VALID_PTR(pszWhat));
3926	Assert(VALID_PTR(pszByWhat));
3927	const char *pszFoundStr = strstr(pszWhere, pszWhat);
3928	if (!pszFoundStr)
3929	return NULL;
3930	size_t cFinal = strlen(pszWhere) + 1 + strlen(pszByWhat) - strlen(pszWhat);
3931	char pszNewStr = (char )RTMemAlloc(cFinal);
3932	if (pszNewStr)
3933	{
3934	char *pszTmp = pszNewStr;
3935	memcpy(pszTmp, pszWhere, pszFoundStr - pszWhere);
3936	pszTmp += pszFoundStr - pszWhere;
3937	memcpy(pszTmp, pszByWhat, strlen(pszByWhat));
3938	pszTmp += strlen(pszByWhat);
3939	strcpy(pszTmp, pszFoundStr + strlen(pszWhat));
3940	}
3941	return pszNewStr;
3942	}
3943
3944	/** @copydoc VBOXHDDBACKEND::pfnRename */
3945	static int vmdkRename(void pBackendData, const char pszFilename)
3946	{
3947	LogFlowFunc(("pBackendData=%#p pszFilename=%#p\n", pBackendData, pszFilename));
3948
3949	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
3950	int rc = VINF_SUCCESS;
3951	char **apszOldName = NULL;
3952	char **apszNewName = NULL;
3953	char **apszNewLines = NULL;
3954	char *pszOldDescName = NULL;
3955	bool fImageFreed = false;
3956	bool fEmbeddedDesc = false;
3957	unsigned cExtents = pImage->cExtents;
3958	char *pszNewBasename;
3959	char *pszOldBasename;
3960	unsigned i, line;
3961	VMDKDESCRIPTOR DescriptorCopy;
3962	VMDKEXTENT ExtentCopy;
3963
3964	memset(&DescriptorCopy, 0, sizeof(DescriptorCopy));
3965
3966	/* Check arguments. */
3967	if ( !pImage
3968	\|\| (pImage->uImageFlags & VD_VMDK_IMAGE_FLAGS_RAWDISK)
3969	\|\| !VALID_PTR(pszFilename)
3970	\|\| !*pszFilename)
3971	{
3972	rc = VERR_INVALID_PARAMETER;
3973	goto out;
3974	}
3975
3976	/*
3977	* Allocate an array to store both old and new names of renamed files
3978	* in case we have to roll back the changes. Arrays are initialized
3979	* with zeros. We actually save stuff when and if we change it.
3980	*/
3981	apszOldName = (char *)RTMemTmpAllocZ((cExtents + 1) sizeof(char*));
3982	apszNewName = (char *)RTMemTmpAllocZ((cExtents + 1) sizeof(char*));
3983	apszNewLines = (char *)RTMemTmpAllocZ((cExtents) sizeof(char*));
3984	if (!apszOldName \|\| !apszNewName \|\| !apszNewLines)
3985	{
3986	rc = VERR_NO_MEMORY;
3987	goto out;
3988	}
3989
3990	/* Save the descriptor size and position. */
3991	if (pImage->pDescData)
3992	{
3993	/* Separate descriptor file. */
3994	fEmbeddedDesc = false;
3995	}
3996	else
3997	{
3998	/* Embedded descriptor file. */
3999	ExtentCopy = pImage->pExtents[0];
4000	fEmbeddedDesc = true;
4001	}
4002	/* Save the descriptor content. */
4003	DescriptorCopy.cLines = pImage->Descriptor.cLines;
4004	for (i = 0; i < DescriptorCopy.cLines; i++)
4005	{
4006	DescriptorCopy.aLines[i] = RTStrDup(pImage->Descriptor.aLines[i]);
4007	if (!DescriptorCopy.aLines[i])
4008	{
4009	rc = VERR_NO_MEMORY;
4010	goto out;
4011	}
4012	}
4013
4014	/* Prepare both old and new base names used for string replacement. */
4015	pszNewBasename = RTStrDup(RTPathFilename(pszFilename));
4016	RTPathStripExt(pszNewBasename);
4017	pszOldBasename = RTStrDup(RTPathFilename(pImage->pszFilename));
4018	RTPathStripExt(pszOldBasename);
4019
4020	/* --- Up to this point we have not done any damage yet. --- */
4021
4022	/* Save the old name for easy access to the old descriptor file. */
4023	pszOldDescName = RTStrDup(pImage->pszFilename);
4024
4025	/* Rename the extents. */
4026	for (i = 0, line = pImage->Descriptor.uFirstExtent;
4027	i < cExtents;
4028	i++, line = pImage->Descriptor.aNextLines[line])
4029	{
4030	PVMDKEXTENT pExtent = &pImage->pExtents[i];
4031	/* Assume that vmdkStrReplace will fail. */
4032	rc = VERR_NO_MEMORY;
4033	/* Update the descriptor. */
4034	apszNewLines[i] = vmdkStrReplace(pImage->Descriptor.aLines[line],
4035	pszOldBasename, pszNewBasename);
4036	if (!apszNewLines[i])
4037	goto rollback;
4038	pImage->Descriptor.aLines[line] = apszNewLines[i];
4039	/* Compose new name for the extent. */
4040	apszNewName[i] = vmdkStrReplace(pExtent->pszFullname,
4041	pszOldBasename, pszNewBasename);
4042	if (!apszNewName[i])
4043	goto rollback;
4044	/* Close the extent file. */
4045	vmdkFileClose(pImage, &pExtent->pFile, false);
4046	/* Rename the extent file. */
4047	rc = RTFileMove(pExtent->pszFullname, apszNewName[i], 0);
4048	if (RT_FAILURE(rc))
4049	goto rollback;
4050	/* Remember the old name. */
4051	apszOldName[i] = RTStrDup(pExtent->pszFullname);
4052	}
4053
4054	/* Make sure the descriptor gets written back. */
4055	pImage->Descriptor.fDirty = true;
4056	/* Release all old stuff and write back the descriptor. */
4057	vmdkFreeImage(pImage, false);
4058
4059	fImageFreed = true;
4060
4061	/* Last elements of new/old name arrays are intended for
4062	* storing descriptor's names.
4063	*/
4064	apszNewName[cExtents] = RTPathFilename(pszFilename);
4065	/* Rename the descriptor file if it's separate. */
4066	if (!fEmbeddedDesc)
4067	{
4068	rc = RTFileMove(pImage->pszFilename, apszNewName[cExtents], 0);
4069	if (RT_FAILURE(rc))
4070	goto rollback;
4071	/* Save old name only if we may need to change it back. */
4072	apszOldName[cExtents] = RTStrDup(pszFilename);
4073	}
4074
4075	/* Update pImage with the new information. */
4076	pImage->pszFilename = pszFilename;
4077
4078	/* Open the new image. */
4079	rc = vmdkOpenImage(pImage, pImage->uOpenFlags);
4080	if (RT_SUCCESS(rc))
4081	goto out;
4082
4083	rollback:
4084	/* Roll back all changes in case of failure. */
4085	if (RT_FAILURE(rc))
4086	{
4087	int rrc;
4088	if (!fImageFreed)
4089	{
4090	/*
4091	* Some extents may have been closed, close the rest. We will
4092	* re-open the whole thing later.
4093	*/
4094	vmdkFreeImage(pImage, false);
4095	}
4096	/* Rename files back and free the memory. */
4097	for (i = 0; i < cExtents + 1; i++)
4098	{
4099	if (apszOldName[i])
4100	{
4101	rrc = RTFileMove(apszNewName[i], apszOldName[i], 0);
4102	AssertRC(rrc);
4103	RTStrFree(apszOldName[i]);
4104	}
4105	if (apszNewName[i])
4106	RTStrFree(apszNewName[i]);
4107	if (apszNewLines[i])
4108	RTStrFree(apszNewLines[i]);
4109	}
4110	/* Restore the old descriptor. */
4111	PVMDKFILE pFile;
4112	rrc = vmdkFileOpen(pImage, &pFile, pszOldDescName,
4113	RTFILE_O_READWRITE \| RTFILE_O_OPEN \| RTFILE_O_DENY_WRITE, false);
4114	AssertRC(rrc);
4115	if (fEmbeddedDesc)
4116	{
4117	ExtentCopy.pFile = pFile;
4118	pImage->pExtents = &ExtentCopy;
4119	}
4120	else
4121	pImage->pFile = pFile;
4122	pImage->Descriptor = DescriptorCopy;
4123	vmdkWriteDescriptor(pImage);
4124	vmdkFileClose(pImage, &pFile, false);
4125	RTStrFree(pszOldDescName);
4126	/* Re-open the image back. */
4127	rrc = vmdkOpenImage(pImage, pImage->uOpenFlags);
4128	AssertRC(rrc);
4129	}
4130
4131	out:
4132	for (i = 0; i < DescriptorCopy.cLines; i++)
4133	if (DescriptorCopy.aLines[i])
4134	RTStrFree(DescriptorCopy.aLines[i]);
4135	if (apszOldName)
4136	RTMemTmpFree(apszOldName);
4137	if (apszNewName)
4138	RTMemTmpFree(apszNewName);
4139	if (apszNewLines)
4140	RTMemTmpFree(apszNewLines);
4141	LogFlowFunc(("returns %Rrc\n", rc));
4142	return rc;
4143	}
4144
4145	/** @copydoc VBOXHDDBACKEND::pfnClose */
4146	static int vmdkClose(void *pBackendData, bool fDelete)
4147	{
4148	LogFlowFunc(("pBackendData=%#p fDelete=%d\n", pBackendData, fDelete));
4149	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4150	int rc = VINF_SUCCESS;
4151
4152	/* Freeing a never allocated image (e.g. because the open failed) is
4153	* not signalled as an error. After all nothing bad happens. */
4154	if (pImage)
4155	vmdkFreeImage(pImage, fDelete);
4156
4157	LogFlowFunc(("returns %Rrc\n", rc));
4158	return rc;
4159	}
4160
4161	/** @copydoc VBOXHDDBACKEND::pfnRead */
4162	static int vmdkRead(void pBackendData, uint64_t uOffset, void pvBuf,
4163	size_t cbToRead, size_t *pcbActuallyRead)
4164	{
4165	LogFlowFunc(("pBackendData=%#p uOffset=%llu pvBuf=%#p cbToRead=%zu pcbActuallyRead=%#p\n", pBackendData, uOffset, pvBuf, cbToRead, pcbActuallyRead));
4166	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4167	PVMDKEXTENT pExtent;
4168	uint64_t uSectorExtentRel;
4169	uint64_t uSectorExtentAbs;
4170	int rc;
4171
4172	Assert(pImage);
4173	Assert(uOffset % 512 == 0);
4174	Assert(cbToRead % 512 == 0);
4175
4176	if ( uOffset + cbToRead > pImage->cbSize
4177	\|\| cbToRead == 0)
4178	{
4179	rc = VERR_INVALID_PARAMETER;
4180	goto out;
4181	}
4182
4183	rc = vmdkFindExtent(pImage, VMDK_BYTE2SECTOR(uOffset),
4184	&pExtent, &uSectorExtentRel);
4185	if (RT_FAILURE(rc))
4186	goto out;
4187
4188	/* Check access permissions as defined in the extent descriptor. */
4189	if (pExtent->enmAccess == VMDKACCESS_NOACCESS)
4190	{
4191	rc = VERR_VDI_INVALID_STATE;
4192	goto out;
4193	}
4194
4195	/* Clip read range to remain in this extent. */
4196	cbToRead = RT_MIN(cbToRead, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
4197
4198	/* Handle the read according to the current extent type. */
4199	switch (pExtent->enmType)
4200	{
4201	case VMDKETYPE_HOSTED_SPARSE:
4202	#ifdef VBOX_WITH_VMDK_ESX
4203	case VMDKETYPE_ESX_SPARSE:
4204	#endif /* VBOX_WITH_VMDK_ESX */
4205	rc = vmdkGetSector(pImage->pGTCache, pExtent, uSectorExtentRel,
4206	&uSectorExtentAbs);
4207	if (RT_FAILURE(rc))
4208	goto out;
4209	/* Clip read range to at most the rest of the grain. */
4210	cbToRead = RT_MIN(cbToRead, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain - uSectorExtentRel % pExtent->cSectorsPerGrain));
4211	Assert(!(cbToRead % 512));
4212	if (uSectorExtentAbs == 0)
4213	rc = VERR_VDI_BLOCK_FREE;
4214	else
4215	rc = vmdkFileReadAt(pExtent->pFile,
4216	VMDK_SECTOR2BYTE(uSectorExtentAbs),
4217	pvBuf, cbToRead, NULL);
4218	break;
4219	case VMDKETYPE_FLAT:
4220	rc = vmdkFileReadAt(pExtent->pFile,
4221	VMDK_SECTOR2BYTE(uSectorExtentRel),
4222	pvBuf, cbToRead, NULL);
4223	break;
4224	case VMDKETYPE_ZERO:
4225	memset(pvBuf, '\0', cbToRead);
4226	break;
4227	}
4228	*pcbActuallyRead = cbToRead;
4229
4230	out:
4231	LogFlowFunc(("returns %Rrc\n", rc));
4232	return rc;
4233	}
4234
4235	/** @copydoc VBOXHDDBACKEND::pfnWrite */
4236	static int vmdkWrite(void pBackendData, uint64_t uOffset, const void pvBuf,
4237	size_t cbToWrite, size_t *pcbWriteProcess,
4238	size_t pcbPreRead, size_t pcbPostRead, unsigned fWrite)
4239	{
4240	LogFlowFunc(("pBackendData=%#p uOffset=%llu pvBuf=%#p cbToWrite=%zu pcbWriteProcess=%#p pcbPreRead=%#p pcbPostRead=%#p\n", pBackendData, uOffset, pvBuf, cbToWrite, pcbWriteProcess, pcbPreRead, pcbPostRead));
4241	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4242	PVMDKEXTENT pExtent;
4243	uint64_t uSectorExtentRel;
4244	uint64_t uSectorExtentAbs;
4245	int rc;
4246
4247	Assert(pImage);
4248	Assert(uOffset % 512 == 0);
4249	Assert(cbToWrite % 512 == 0);
4250
4251	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
4252	{
4253	rc = VERR_VDI_IMAGE_READ_ONLY;
4254	goto out;
4255	}
4256
4257	if (cbToWrite == 0)
4258	{
4259	rc = VERR_INVALID_PARAMETER;
4260	goto out;
4261	}
4262
4263	/* No size check here, will do that later when the extent is located.
4264	* There are sparse images out there which according to the spec are
4265	* invalid, because the total size is not a multiple of the grain size.
4266	* Also for sparse images which are stitched together in odd ways (not at
4267	* grain boundaries, and with the nominal size not being a multiple of the
4268	* grain size), this would prevent writing to the last grain. */
4269
4270	rc = vmdkFindExtent(pImage, VMDK_BYTE2SECTOR(uOffset),
4271	&pExtent, &uSectorExtentRel);
4272	if (RT_FAILURE(rc))
4273	goto out;
4274
4275	/* Check access permissions as defined in the extent descriptor. */
4276	if (pExtent->enmAccess != VMDKACCESS_READWRITE)
4277	{
4278	rc = VERR_VDI_INVALID_STATE;
4279	goto out;
4280	}
4281
4282	/* Handle the write according to the current extent type. */
4283	switch (pExtent->enmType)
4284	{
4285	case VMDKETYPE_HOSTED_SPARSE:
4286	#ifdef VBOX_WITH_VMDK_ESX
4287	case VMDKETYPE_ESX_SPARSE:
4288	#endif /* VBOX_WITH_VMDK_ESX */
4289	rc = vmdkGetSector(pImage->pGTCache, pExtent, uSectorExtentRel,
4290	&uSectorExtentAbs);
4291	if (RT_FAILURE(rc))
4292	goto out;
4293	/* Clip write range to at most the rest of the grain. */
4294	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain - uSectorExtentRel % pExtent->cSectorsPerGrain));
4295	if (uSectorExtentAbs == 0)
4296	{
4297	if (cbToWrite == VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain))
4298	{
4299	/* Full block write to a previously unallocated block.
4300	* Check if the caller wants to avoid this. */
4301	if (!(fWrite & VD_WRITE_NO_ALLOC))
4302	{
4303	/* Allocate GT and find out where to store the grain. */
4304	rc = vmdkAllocGrain(pImage->pGTCache, pExtent,
4305	uSectorExtentRel, pvBuf, cbToWrite);
4306	}
4307	else
4308	rc = VERR_VDI_BLOCK_FREE;
4309	*pcbPreRead = 0;
4310	*pcbPostRead = 0;
4311	}
4312	else
4313	{
4314	/* Clip write range to remain in this extent. */
4315	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
4316	*pcbPreRead = VMDK_SECTOR2BYTE(uSectorExtentRel % pExtent->cSectorsPerGrain);
4317	pcbPostRead = VMDK_SECTOR2BYTE(pExtent->cSectorsPerGrain) - cbToWrite - pcbPreRead;
4318	rc = VERR_VDI_BLOCK_FREE;
4319	}
4320	}
4321	else
4322	rc = vmdkFileWriteAt(pExtent->pFile,
4323	VMDK_SECTOR2BYTE(uSectorExtentAbs),
4324	pvBuf, cbToWrite, NULL);
4325	break;
4326	case VMDKETYPE_FLAT:
4327	/* Clip write range to remain in this extent. */
4328	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
4329	rc = vmdkFileWriteAt(pExtent->pFile,
4330	VMDK_SECTOR2BYTE(uSectorExtentRel),
4331	pvBuf, cbToWrite, NULL);
4332	break;
4333	case VMDKETYPE_ZERO:
4334	/* Clip write range to remain in this extent. */
4335	cbToWrite = RT_MIN(cbToWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
4336	break;
4337	}
4338	if (pcbWriteProcess)
4339	*pcbWriteProcess = cbToWrite;
4340
4341	out:
4342	LogFlowFunc(("returns %Rrc\n", rc));
4343	return rc;
4344	}
4345
4346	/** @copydoc VBOXHDDBACKEND::pfnFlush */
4347	static int vmdkFlush(void *pBackendData)
4348	{
4349	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
4350	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4351	int rc;
4352
4353	Assert(pImage);
4354
4355	rc = vmdkFlushImage(pImage);
4356	LogFlowFunc(("returns %Rrc\n", rc));
4357	return rc;
4358	}
4359
4360	/** @copydoc VBOXHDDBACKEND::pfnGetVersion */
4361	static unsigned vmdkGetVersion(void *pBackendData)
4362	{
4363	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
4364	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4365
4366	Assert(pImage);
4367
4368	if (pImage)
4369	return VMDK_IMAGE_VERSION;
4370	else
4371	return 0;
4372	}
4373
4374	/** @copydoc VBOXHDDBACKEND::pfnGetImageType */
4375	static int vmdkGetImageType(void *pBackendData, PVDIMAGETYPE penmImageType)
4376	{
4377	LogFlowFunc(("pBackendData=%#p penmImageType=%#p\n", pBackendData, penmImageType));
4378	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4379	int rc = VINF_SUCCESS;
4380
4381	Assert(pImage);
4382	Assert(penmImageType);
4383
4384	if (pImage && pImage->cExtents != 0)
4385	*penmImageType = pImage->enmImageType;
4386	else
4387	rc = VERR_VDI_NOT_OPENED;
4388
4389	LogFlowFunc(("returns %Rrc enmImageType=%u\n", rc, *penmImageType));
4390	return rc;
4391	}
4392
4393	/** @copydoc VBOXHDDBACKEND::pfnGetSize */
4394	static uint64_t vmdkGetSize(void *pBackendData)
4395	{
4396	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
4397	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4398
4399	Assert(pImage);
4400
4401	if (pImage)
4402	return pImage->cbSize;
4403	else
4404	return 0;
4405	}
4406
4407	/** @copydoc VBOXHDDBACKEND::pfnGetFileSize */
4408	static uint64_t vmdkGetFileSize(void *pBackendData)
4409	{
4410	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
4411	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4412	uint64_t cb = 0;
4413
4414	Assert(pImage);
4415
4416	if (pImage)
4417	{
4418	uint64_t cbFile;
4419	if (pImage->pFile != NULL)
4420	{
4421	int rc = vmdkFileGetSize(pImage->pFile, &cbFile);
4422	if (RT_SUCCESS(rc))
4423	cb += cbFile;
4424	for (unsigned i = 0; i <= pImage->cExtents; i++)
4425	{
4426	rc = vmdkFileGetSize(pImage->pFile, &cbFile);
4427	if (RT_SUCCESS(rc))
4428	cb += cbFile;
4429	}
4430	}
4431	}
4432
4433	LogFlowFunc(("returns %lld\n", cb));
4434	return cb;
4435	}
4436
4437	/** @copydoc VBOXHDDBACKEND::pfnGetPCHSGeometry */
4438	static int vmdkGetPCHSGeometry(void *pBackendData,
4439	PPDMMEDIAGEOMETRY pPCHSGeometry)
4440	{
4441	LogFlowFunc(("pBackendData=%#p pPCHSGeometry=%#p\n", pBackendData, pPCHSGeometry));
4442	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4443	int rc;
4444
4445	Assert(pImage);
4446
4447	if (pImage)
4448	{
4449	if (pImage->PCHSGeometry.cCylinders)
4450	{
4451	*pPCHSGeometry = pImage->PCHSGeometry;
4452	rc = VINF_SUCCESS;
4453	}
4454	else
4455	rc = VERR_VDI_GEOMETRY_NOT_SET;
4456	}
4457	else
4458	rc = VERR_VDI_NOT_OPENED;
4459
4460	LogFlowFunc(("returns %Rrc (PCHS=%u/%u/%u)\n", rc, pPCHSGeometry->cCylinders, pPCHSGeometry->cHeads, pPCHSGeometry->cSectors));
4461	return rc;
4462	}
4463
4464	/** @copydoc VBOXHDDBACKEND::pfnSetPCHSGeometry */
4465	static int vmdkSetPCHSGeometry(void *pBackendData,
4466	PCPDMMEDIAGEOMETRY pPCHSGeometry)
4467	{
4468	LogFlowFunc(("pBackendData=%#p pPCHSGeometry=%#p PCHS=%u/%u/%u\n", pBackendData, pPCHSGeometry, pPCHSGeometry->cCylinders, pPCHSGeometry->cHeads, pPCHSGeometry->cSectors));
4469	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4470	int rc;
4471
4472	Assert(pImage);
4473
4474	if (pImage)
4475	{
4476	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
4477	{
4478	rc = VERR_VDI_IMAGE_READ_ONLY;
4479	goto out;
4480	}
4481	rc = vmdkDescSetPCHSGeometry(pImage, pPCHSGeometry);
4482	if (RT_FAILURE(rc))
4483	goto out;
4484
4485	pImage->PCHSGeometry = *pPCHSGeometry;
4486	rc = VINF_SUCCESS;
4487	}
4488	else
4489	rc = VERR_VDI_NOT_OPENED;
4490
4491	out:
4492	LogFlowFunc(("returns %Rrc\n", rc));
4493	return rc;
4494	}
4495
4496	/** @copydoc VBOXHDDBACKEND::pfnGetLCHSGeometry */
4497	static int vmdkGetLCHSGeometry(void *pBackendData,
4498	PPDMMEDIAGEOMETRY pLCHSGeometry)
4499	{
4500	LogFlowFunc(("pBackendData=%#p pLCHSGeometry=%#p\n", pBackendData, pLCHSGeometry));
4501	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4502	int rc;
4503
4504	Assert(pImage);
4505
4506	if (pImage)
4507	{
4508	if (pImage->LCHSGeometry.cCylinders)
4509	{
4510	*pLCHSGeometry = pImage->LCHSGeometry;
4511	rc = VINF_SUCCESS;
4512	}
4513	else
4514	rc = VERR_VDI_GEOMETRY_NOT_SET;
4515	}
4516	else
4517	rc = VERR_VDI_NOT_OPENED;
4518
4519	LogFlowFunc(("returns %Rrc (LCHS=%u/%u/%u)\n", rc, pLCHSGeometry->cCylinders, pLCHSGeometry->cHeads, pLCHSGeometry->cSectors));
4520	return rc;
4521	}
4522
4523	/** @copydoc VBOXHDDBACKEND::pfnSetLCHSGeometry */
4524	static int vmdkSetLCHSGeometry(void *pBackendData,
4525	PCPDMMEDIAGEOMETRY pLCHSGeometry)
4526	{
4527	LogFlowFunc(("pBackendData=%#p pLCHSGeometry=%#p LCHS=%u/%u/%u\n", pBackendData, pLCHSGeometry, pLCHSGeometry->cCylinders, pLCHSGeometry->cHeads, pLCHSGeometry->cSectors));
4528	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4529	int rc;
4530
4531	Assert(pImage);
4532
4533	if (pImage)
4534	{
4535	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
4536	{
4537	rc = VERR_VDI_IMAGE_READ_ONLY;
4538	goto out;
4539	}
4540	rc = vmdkDescSetLCHSGeometry(pImage, pLCHSGeometry);
4541	if (RT_FAILURE(rc))
4542	goto out;
4543
4544	pImage->LCHSGeometry = *pLCHSGeometry;
4545	rc = VINF_SUCCESS;
4546	}
4547	else
4548	rc = VERR_VDI_NOT_OPENED;
4549
4550	out:
4551	LogFlowFunc(("returns %Rrc\n", rc));
4552	return rc;
4553	}
4554
4555	/** @copydoc VBOXHDDBACKEND::pfnGetImageFlags */
4556	static unsigned vmdkGetImageFlags(void *pBackendData)
4557	{
4558	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
4559	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4560	unsigned uImageFlags;
4561
4562	Assert(pImage);
4563
4564	if (pImage)
4565	uImageFlags = pImage->uImageFlags;
4566	else
4567	uImageFlags = 0;
4568
4569	LogFlowFunc(("returns %#x\n", uImageFlags));
4570	return uImageFlags;
4571	}
4572
4573	/** @copydoc VBOXHDDBACKEND::pfnGetOpenFlags */
4574	static unsigned vmdkGetOpenFlags(void *pBackendData)
4575	{
4576	LogFlowFunc(("pBackendData=%#p\n", pBackendData));
4577	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4578	unsigned uOpenFlags;
4579
4580	Assert(pImage);
4581
4582	if (pImage)
4583	uOpenFlags = pImage->uOpenFlags;
4584	else
4585	uOpenFlags = 0;
4586
4587	LogFlowFunc(("returns %#x\n", uOpenFlags));
4588	return uOpenFlags;
4589	}
4590
4591	/** @copydoc VBOXHDDBACKEND::pfnSetOpenFlags */
4592	static int vmdkSetOpenFlags(void *pBackendData, unsigned uOpenFlags)
4593	{
4594	LogFlowFunc(("pBackendData=%#p\n uOpenFlags=%#x", pBackendData, uOpenFlags));
4595	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4596	int rc;
4597
4598	/* Image must be opened and the new flags must be valid. Just readonly flag
4599	* is supported. */
4600	if (!pImage \|\| uOpenFlags & ~(VD_OPEN_FLAGS_READONLY \| VD_OPEN_FLAGS_ASYNC_IO))
4601	{
4602	rc = VERR_INVALID_PARAMETER;
4603	goto out;
4604	}
4605
4606	/* Implement this operation via reopening the image. */
4607	vmdkFreeImage(pImage, false);
4608	rc = vmdkOpenImage(pImage, uOpenFlags);
4609
4610	out:
4611	LogFlowFunc(("returns %Rrc\n", rc));
4612	return rc;
4613	}
4614
4615	/** @copydoc VBOXHDDBACKEND::pfnGetComment */
4616	static int vmdkGetComment(void pBackendData, char pszComment,
4617	size_t cbComment)
4618	{
4619	LogFlowFunc(("pBackendData=%#p pszComment=%#p cbComment=%zu\n", pBackendData, pszComment, cbComment));
4620	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4621	int rc;
4622
4623	Assert(pImage);
4624
4625	if (pImage)
4626	{
4627	const char *pszCommentEncoded = NULL;
4628	rc = vmdkDescDDBGetStr(pImage, &pImage->Descriptor,
4629	"ddb.comment", &pszCommentEncoded);
4630	if (rc == VERR_VDI_VALUE_NOT_FOUND)
4631	pszCommentEncoded = NULL;
4632	else if (RT_FAILURE(rc))
4633	goto out;
4634
4635	if (pszComment)
4636	rc = vmdkDecodeString(pszCommentEncoded, pszComment, cbComment);
4637	else
4638	rc = VINF_SUCCESS;
4639	if (pszCommentEncoded)
4640	RTStrFree((char )(void )pszCommentEncoded);
4641	}
4642	else
4643	rc = VERR_VDI_NOT_OPENED;
4644
4645	out:
4646	LogFlowFunc(("returns %Rrc comment='%s'\n", rc, pszComment));
4647	return rc;
4648	}
4649
4650	/** @copydoc VBOXHDDBACKEND::pfnSetComment */
4651	static int vmdkSetComment(void pBackendData, const char pszComment)
4652	{
4653	LogFlowFunc(("pBackendData=%#p pszComment=\"%s\"\n", pBackendData, pszComment));
4654	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4655	int rc;
4656
4657	Assert(pImage);
4658
4659	if (pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY)
4660	{
4661	rc = VERR_VDI_IMAGE_READ_ONLY;
4662	goto out;
4663	}
4664
4665	if (pImage)
4666	rc = vmdkSetImageComment(pImage, pszComment);
4667	else
4668	rc = VERR_VDI_NOT_OPENED;
4669
4670	out:
4671	LogFlowFunc(("returns %Rrc\n", rc));
4672	return rc;
4673	}
4674
4675	/** @copydoc VBOXHDDBACKEND::pfnGetUuid */
4676	static int vmdkGetUuid(void *pBackendData, PRTUUID pUuid)
4677	{
4678	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
4679	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4680	int rc;
4681
4682	Assert(pImage);
4683
4684	if (pImage)
4685	{
4686	*pUuid = pImage->ImageUuid;
4687	rc = VINF_SUCCESS;
4688	}
4689	else
4690	rc = VERR_VDI_NOT_OPENED;
4691
4692	LogFlowFunc(("returns %Rrc (%RTuuid)\n", rc, pUuid));
4693	return rc;
4694	}
4695
4696	/** @copydoc VBOXHDDBACKEND::pfnSetUuid */
4697	static int vmdkSetUuid(void *pBackendData, PCRTUUID pUuid)
4698	{
4699	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
4700	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4701	int rc;
4702
4703	LogFlowFunc(("%RTuuid\n", pUuid));
4704	Assert(pImage);
4705
4706	if (pImage)
4707	{
4708	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
4709	{
4710	pImage->ImageUuid = *pUuid;
4711	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
4712	VMDK_DDB_IMAGE_UUID, pUuid);
4713	if (RT_FAILURE(rc))
4714	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error storing image UUID in descriptor in '%s'"), pImage->pszFilename);
4715	rc = VINF_SUCCESS;
4716	}
4717	else
4718	rc = VERR_VDI_IMAGE_READ_ONLY;
4719	}
4720	else
4721	rc = VERR_VDI_NOT_OPENED;
4722
4723	LogFlowFunc(("returns %Rrc\n", rc));
4724	return rc;
4725	}
4726
4727	/** @copydoc VBOXHDDBACKEND::pfnGetModificationUuid */
4728	static int vmdkGetModificationUuid(void *pBackendData, PRTUUID pUuid)
4729	{
4730	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
4731	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4732	int rc;
4733
4734	Assert(pImage);
4735
4736	if (pImage)
4737	{
4738	*pUuid = pImage->ModificationUuid;
4739	rc = VINF_SUCCESS;
4740	}
4741	else
4742	rc = VERR_VDI_NOT_OPENED;
4743
4744	LogFlowFunc(("returns %Rrc (%RTuuid)\n", rc, pUuid));
4745	return rc;
4746	}
4747
4748	/** @copydoc VBOXHDDBACKEND::pfnSetModificationUuid */
4749	static int vmdkSetModificationUuid(void *pBackendData, PCRTUUID pUuid)
4750	{
4751	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
4752	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4753	int rc;
4754
4755	Assert(pImage);
4756
4757	if (pImage)
4758	{
4759	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
4760	{
4761	pImage->ModificationUuid = *pUuid;
4762	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
4763	VMDK_DDB_MODIFICATION_UUID, pUuid);
4764	if (RT_FAILURE(rc))
4765	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error storing modification UUID in descriptor in '%s'"), pImage->pszFilename);
4766	rc = VINF_SUCCESS;
4767	}
4768	else
4769	rc = VERR_VDI_IMAGE_READ_ONLY;
4770	}
4771	else
4772	rc = VERR_VDI_NOT_OPENED;
4773
4774	LogFlowFunc(("returns %Rrc\n", rc));
4775	return rc;
4776	}
4777
4778	/** @copydoc VBOXHDDBACKEND::pfnGetParentUuid */
4779	static int vmdkGetParentUuid(void *pBackendData, PRTUUID pUuid)
4780	{
4781	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
4782	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4783	int rc;
4784
4785	Assert(pImage);
4786
4787	if (pImage)
4788	{
4789	*pUuid = pImage->ParentUuid;
4790	rc = VINF_SUCCESS;
4791	}
4792	else
4793	rc = VERR_VDI_NOT_OPENED;
4794
4795	LogFlowFunc(("returns %Rrc (%RTuuid)\n", rc, pUuid));
4796	return rc;
4797	}
4798
4799	/** @copydoc VBOXHDDBACKEND::pfnSetParentUuid */
4800	static int vmdkSetParentUuid(void *pBackendData, PCRTUUID pUuid)
4801	{
4802	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
4803	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4804	int rc;
4805
4806	Assert(pImage);
4807
4808	if (pImage)
4809	{
4810	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
4811	{
4812	pImage->ParentUuid = *pUuid;
4813	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
4814	VMDK_DDB_PARENT_UUID, pUuid);
4815	if (RT_FAILURE(rc))
4816	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error storing parent image UUID in descriptor in '%s'"), pImage->pszFilename);
4817	rc = VINF_SUCCESS;
4818	}
4819	else
4820	rc = VERR_VDI_IMAGE_READ_ONLY;
4821	}
4822	else
4823	rc = VERR_VDI_NOT_OPENED;
4824
4825	LogFlowFunc(("returns %Rrc\n", rc));
4826	return rc;
4827	}
4828
4829	/** @copydoc VBOXHDDBACKEND::pfnGetParentModificationUuid */
4830	static int vmdkGetParentModificationUuid(void *pBackendData, PRTUUID pUuid)
4831	{
4832	LogFlowFunc(("pBackendData=%#p pUuid=%#p\n", pBackendData, pUuid));
4833	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4834	int rc;
4835
4836	Assert(pImage);
4837
4838	if (pImage)
4839	{
4840	*pUuid = pImage->ParentModificationUuid;
4841	rc = VINF_SUCCESS;
4842	}
4843	else
4844	rc = VERR_VDI_NOT_OPENED;
4845
4846	LogFlowFunc(("returns %Rrc (%RTuuid)\n", rc, pUuid));
4847	return rc;
4848	}
4849
4850	/** @copydoc VBOXHDDBACKEND::pfnSetParentModificationUuid */
4851	static int vmdkSetParentModificationUuid(void *pBackendData, PCRTUUID pUuid)
4852	{
4853	LogFlowFunc(("pBackendData=%#p Uuid=%RTuuid\n", pBackendData, pUuid));
4854	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4855	int rc;
4856
4857	Assert(pImage);
4858
4859	if (pImage)
4860	{
4861	if (!(pImage->uOpenFlags & VD_OPEN_FLAGS_READONLY))
4862	{
4863	pImage->ParentModificationUuid = *pUuid;
4864	rc = vmdkDescDDBSetUuid(pImage, &pImage->Descriptor,
4865	VMDK_DDB_PARENT_MODIFICATION_UUID, pUuid);
4866	if (RT_FAILURE(rc))
4867	return vmdkError(pImage, rc, RT_SRC_POS, N_("VMDK: error storing parent image UUID in descriptor in '%s'"), pImage->pszFilename);
4868	rc = VINF_SUCCESS;
4869	}
4870	else
4871	rc = VERR_VDI_IMAGE_READ_ONLY;
4872	}
4873	else
4874	rc = VERR_VDI_NOT_OPENED;
4875
4876	LogFlowFunc(("returns %Rrc\n", rc));
4877	return rc;
4878	}
4879
4880	/** @copydoc VBOXHDDBACKEND::pfnDump */
4881	static void vmdkDump(void *pBackendData)
4882	{
4883	PVMDKIMAGE pImage = (PVMDKIMAGE)pBackendData;
4884
4885	Assert(pImage);
4886	if (pImage)
4887	{
4888	RTLogPrintf("Header: Geometry PCHS=%u/%u/%u LCHS=%u/%u/%u cbSector=%llu\n",
4889	pImage->PCHSGeometry.cCylinders, pImage->PCHSGeometry.cHeads, pImage->PCHSGeometry.cSectors,
4890	pImage->LCHSGeometry.cCylinders, pImage->LCHSGeometry.cHeads, pImage->LCHSGeometry.cSectors,
4891	VMDK_BYTE2SECTOR(pImage->cbSize));
4892	RTLogPrintf("Header: uuidCreation={%RTuuid}\n", &pImage->ImageUuid);
4893	RTLogPrintf("Header: uuidModification={%RTuuid}\n", &pImage->ModificationUuid);
4894	RTLogPrintf("Header: uuidParent={%RTuuid}\n", &pImage->ParentUuid);
4895	RTLogPrintf("Header: uuidParentModification={%RTuuid}\n", &pImage->ParentModificationUuid);
4896	}
4897	}
4898
4899
4900	static int vmdkGetTimeStamp(void *pvBackendData, PRTTIMESPEC pTimeStamp)
4901	{
4902	int rc = VERR_NOT_IMPLEMENTED;
4903	LogFlow(("%s: returned %Rrc\n", __FUNCTION__, rc));
4904	return rc;
4905	}
4906
4907	static int vmdkGetParentTimeStamp(void *pvBackendData, PRTTIMESPEC pTimeStamp)
4908	{
4909	int rc = VERR_NOT_IMPLEMENTED;
4910	LogFlow(("%s: returned %Rrc\n", __FUNCTION__, rc));
4911	return rc;
4912	}
4913
4914	static int vmdkSetParentTimeStamp(void *pvBackendData, PCRTTIMESPEC pTimeStamp)
4915	{
4916	int rc = VERR_NOT_IMPLEMENTED;
4917	LogFlow(("%s: returned %Rrc\n", __FUNCTION__, rc));
4918	return rc;
4919	}
4920
4921	static int vmdkGetParentFilename(void pvBackendData, char *ppszParentFilename)
4922	{
4923	int rc = VERR_NOT_IMPLEMENTED;
4924	LogFlow(("%s: returned %Rrc\n", __FUNCTION__, rc));
4925	return rc;
4926	}
4927
4928	static int vmdkSetParentFilename(void pvBackendData, const char pszParentFilename)
4929	{
4930	int rc = VERR_NOT_IMPLEMENTED;
4931	LogFlow(("%s: returned %Rrc\n", __FUNCTION__, rc));
4932	return rc;
4933	}
4934
4935	static bool vmdkIsAsyncIOSupported(void *pvBackendData)
4936	{
4937	PVMDKIMAGE pImage = (PVMDKIMAGE)pvBackendData;
4938	bool fAsyncIOSupported = false;
4939
4940	if (pImage)
4941	{
4942	/* We only support async I/O support if the image only consists of FLAT or ZERO extents. */
4943	fAsyncIOSupported = true;
4944	for (unsigned i = 0; i < pImage->cExtents; i++)
4945	{
4946	if ( (pImage->pExtents[i].enmType != VMDKETYPE_FLAT)
4947	&& (pImage->pExtents[i].enmType != VMDKETYPE_ZERO))
4948	{
4949	fAsyncIOSupported = false;
4950	break; /* Stop search */
4951	}
4952	}
4953	}
4954
4955	return fAsyncIOSupported;
4956	}
4957
4958	static int vmdkAsyncRead(void *pvBackendData, uint64_t uOffset, size_t cbRead,
4959	PPDMDATASEG paSeg, unsigned cSeg, void *pvUser)
4960	{
4961	PVMDKIMAGE pImage = (PVMDKIMAGE)pvBackendData;
4962	PVMDKEXTENT pExtent;
4963	int rc = VINF_SUCCESS;
4964	unsigned cTasksToSubmit = 0;
4965	PPDMDATASEG paSegCurrent = paSeg;
4966	unsigned cbLeftInCurrentSegment = paSegCurrent->cbSeg;
4967	unsigned uOffsetInCurrentSegment = 0;
4968
4969	Assert(pImage);
4970	Assert(uOffset % 512 == 0);
4971	Assert(cbRead % 512 == 0);
4972
4973	if ( uOffset + cbRead > pImage->cbSize
4974	\|\| cbRead == 0)
4975	{
4976	rc = VERR_INVALID_PARAMETER;
4977	goto out;
4978	}
4979
4980	while (cbRead && cSeg)
4981	{
4982	unsigned cbToRead;
4983	uint64_t uSectorExtentRel;
4984
4985	rc = vmdkFindExtent(pImage, VMDK_BYTE2SECTOR(uOffset),
4986	&pExtent, &uSectorExtentRel);
4987	if (RT_FAILURE(rc))
4988	goto out;
4989
4990	/* Check access permissions as defined in the extent descriptor. */
4991	if (pExtent->enmAccess == VMDKACCESS_NOACCESS)
4992	{
4993	rc = VERR_VDI_INVALID_STATE;
4994	goto out;
4995	}
4996
4997	/* Clip read range to remain in this extent. */
4998	cbToRead = RT_MIN(cbRead, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
4999	/* Clip read range to remain into current data segment. */
5000	cbToRead = RT_MIN(cbToRead, cbLeftInCurrentSegment);
5001
5002	switch (pExtent->enmType)
5003	{
5004	case VMDKETYPE_FLAT:
5005	{
5006	/* Setup new task. */
5007	void *pTask;
5008	rc = pImage->pInterfaceAsyncIOCallbacks->pfnPrepareRead(pImage->pInterfaceAsyncIO->pvUser, pExtent->pFile->pStorage,
5009	VMDK_SECTOR2BYTE(uSectorExtentRel),
5010	(uint8_t *)paSegCurrent->pvSeg + uOffsetInCurrentSegment,
5011	cbToRead, &pTask);
5012	if (RT_FAILURE(rc))
5013	{
5014	AssertMsgFailed(("Preparing read failed rc=%Rrc\n", rc));
5015	goto out;
5016	}
5017
5018	/* Check for enough room first. */
5019	if (cTasksToSubmit >= pImage->cTask)
5020	{
5021	/* We reached maximum, resize array. Try to realloc memory first. */
5022	void apTaskNew = (void )RTMemRealloc(pImage->apTask, (cTasksToSubmit + 10)sizeof(void ));
5023
5024	if (!apTaskNew)
5025	{
5026	/* We failed. Allocate completely new. */
5027	apTaskNew = (void *)RTMemAllocZ((cTasksToSubmit + 10) sizeof(void *));
5028	if (!apTaskNew)
5029	{
5030	/* Damn, we are out of memory. */
5031	rc = VERR_NO_MEMORY;
5032	goto out;
5033	}
5034
5035	/* Copy task handles over. */
5036	for (unsigned i = 0; i < cTasksToSubmit; i++)
5037	apTaskNew[i] = pImage->apTask[i];
5038
5039	/* Free old memory. */
5040	RTMemFree(pImage->apTask);
5041	}
5042
5043	pImage->cTask = cTasksToSubmit + 10;
5044	pImage->apTask = apTaskNew;
5045	}
5046
5047	pImage->apTask[cTasksToSubmit] = pTask;
5048	cTasksToSubmit++;
5049	break;
5050	}
5051	case VMDKETYPE_ZERO:
5052	memset((uint8_t *)paSegCurrent->pvSeg + uOffsetInCurrentSegment, 0, cbToRead);
5053	break;
5054	default:
5055	AssertMsgFailed(("Unsupported extent type %u\n", pExtent->enmType));
5056	}
5057
5058	cbRead -= cbToRead;
5059	uOffset += cbToRead;
5060	cbLeftInCurrentSegment -= cbToRead;
5061	uOffsetInCurrentSegment += cbToRead;
5062	/* Go to next extent if there is no space left in current one. */
5063	if (!cbLeftInCurrentSegment)
5064	{
5065	uOffsetInCurrentSegment = 0;
5066	paSegCurrent++;
5067	cSeg--;
5068	cbLeftInCurrentSegment = paSegCurrent->cbSeg;
5069	}
5070	}
5071
5072	AssertMsg((cSeg >= 0) && (cbRead == 0), ("No segment left but there is still data to read\n"));
5073
5074	if (cTasksToSubmit == 0)
5075	{
5076	/* The request was completely in a ZERO extent nothing to do. */
5077	rc = VINF_VDI_ASYNC_IO_FINISHED;
5078	}
5079	else
5080	{
5081	/* Submit tasks. */
5082	rc = pImage->pInterfaceAsyncIOCallbacks->pfnTasksSubmit(pImage->pInterfaceAsyncIO->pvUser,
5083	pImage->apTask, cTasksToSubmit,
5084	NULL, pvUser,
5085	NULL /* Nothing required after read. */);
5086	AssertMsg(RT_SUCCESS(rc), ("Failed to enqueue tasks rc=%Rrc\n", rc));
5087	}
5088
5089	out:
5090	LogFlowFunc(("returns %Rrc\n", rc));
5091	return rc;
5092	}
5093
5094	static int vmdkAsyncWrite(void *pvBackendData, uint64_t uOffset, size_t cbWrite,
5095	PPDMDATASEG paSeg, unsigned cSeg, void *pvUser)
5096	{
5097	PVMDKIMAGE pImage = (PVMDKIMAGE)pvBackendData;
5098	PVMDKEXTENT pExtent;
5099	int rc = VINF_SUCCESS;
5100	unsigned cTasksToSubmit = 0;
5101	PPDMDATASEG paSegCurrent = paSeg;
5102	unsigned cbLeftInCurrentSegment = paSegCurrent->cbSeg;
5103	unsigned uOffsetInCurrentSegment = 0;
5104
5105	Assert(pImage);
5106	Assert(uOffset % 512 == 0);
5107	Assert(cbWrite % 512 == 0);
5108
5109	if ( uOffset + cbWrite > pImage->cbSize
5110	\|\| cbWrite == 0)
5111	{
5112	rc = VERR_INVALID_PARAMETER;
5113	goto out;
5114	}
5115
5116	while (cbWrite && cSeg)
5117	{
5118	unsigned cbToWrite;
5119	uint64_t uSectorExtentRel;
5120
5121	rc = vmdkFindExtent(pImage, VMDK_BYTE2SECTOR(uOffset),
5122	&pExtent, &uSectorExtentRel);
5123	if (RT_FAILURE(rc))
5124	goto out;
5125
5126	/* Check access permissions as defined in the extent descriptor. */
5127	if (pExtent->enmAccess == VMDKACCESS_NOACCESS)
5128	{
5129	rc = VERR_VDI_INVALID_STATE;
5130	goto out;
5131	}
5132
5133	/* Clip write range to remain in this extent. */
5134	cbToWrite = RT_MIN(cbWrite, VMDK_SECTOR2BYTE(pExtent->uSectorOffset + pExtent->cNominalSectors - uSectorExtentRel));
5135	/* Clip write range to remain into current data segment. */
5136	cbToWrite = RT_MIN(cbToWrite, cbLeftInCurrentSegment);
5137
5138	switch (pExtent->enmType)
5139	{
5140	case VMDKETYPE_FLAT:
5141	{
5142	/* Setup new task. */
5143	void *pTask;
5144	rc = pImage->pInterfaceAsyncIOCallbacks->pfnPrepareWrite(pImage->pInterfaceAsyncIO->pvUser, pExtent->pFile->pStorage,
5145	VMDK_SECTOR2BYTE(uSectorExtentRel),
5146	(uint8_t *)paSegCurrent->pvSeg + uOffsetInCurrentSegment,
5147	cbToWrite, &pTask);
5148	if (RT_FAILURE(rc))
5149	{
5150	AssertMsgFailed(("Preparing read failed rc=%Rrc\n", rc));
5151	goto out;
5152	}
5153
5154	/* Check for enough room first. */
5155	if (cTasksToSubmit >= pImage->cTask)
5156	{
5157	/* We reached maximum, resize array. Try to realloc memory first. */
5158	void apTaskNew = (void )RTMemRealloc(pImage->apTask, (cTasksToSubmit + 10)sizeof(void ));
5159
5160	if (!apTaskNew)
5161	{
5162	/* We failed. Allocate completely new. */
5163	apTaskNew = (void *)RTMemAllocZ((cTasksToSubmit + 10) sizeof(void *));
5164	if (!apTaskNew)
5165	{
5166	/* Damn, we are out of memory. */
5167	rc = VERR_NO_MEMORY;
5168	goto out;
5169	}
5170
5171	/* Copy task handles over. */
5172	for (unsigned i = 0; i < cTasksToSubmit; i++)
5173	apTaskNew[i] = pImage->apTask[i];
5174
5175	/* Free old memory. */
5176	RTMemFree(pImage->apTask);
5177	}
5178
5179	pImage->cTask = cTasksToSubmit + 10;
5180	pImage->apTask = apTaskNew;
5181	}
5182
5183	pImage->apTask[cTasksToSubmit] = pTask;
5184	cTasksToSubmit++;
5185	break;
5186	}
5187	case VMDKETYPE_ZERO:
5188	/* Nothing left to do. */
5189	break;
5190	default:
5191	AssertMsgFailed(("Unsupported extent type %u\n", pExtent->enmType));
5192	}
5193
5194	cbWrite -= cbToWrite;
5195	uOffset += cbToWrite;
5196	cbLeftInCurrentSegment -= cbToWrite;
5197	uOffsetInCurrentSegment += cbToWrite;
5198	/* Go to next extent if there is no space left in current one. */
5199	if (!cbLeftInCurrentSegment)
5200	{
5201	uOffsetInCurrentSegment = 0;
5202	paSegCurrent++;
5203	cSeg--;
5204	cbLeftInCurrentSegment = paSegCurrent->cbSeg;
5205	}
5206	}
5207
5208	AssertMsg((cSeg >= 0) && (cbWrite == 0), ("No segment left but there is still data to read\n"));
5209
5210	if (cTasksToSubmit == 0)
5211	{
5212	/* The request was completely in a ZERO extent nothing to do. */
5213	rc = VINF_VDI_ASYNC_IO_FINISHED;
5214	}
5215	else
5216	{
5217	/* Submit tasks. */
5218	rc = pImage->pInterfaceAsyncIOCallbacks->pfnTasksSubmit(pImage->pInterfaceAsyncIO->pvUser,
5219	pImage->apTask, cTasksToSubmit,
5220	NULL, pvUser,
5221	NULL /* Nothing required after read. */);
5222	AssertMsg(RT_SUCCESS(rc), ("Failed to enqueue tasks rc=%Rrc\n", rc));
5223	}
5224
5225	out:
5226	LogFlowFunc(("returns %Rrc\n", rc));
5227	return rc;
5228
5229	}
5230
5231
5232	VBOXHDDBACKEND g_VmdkBackend =
5233	{
5234	/* pszBackendName */
5235	"VMDK",
5236	/* cbSize */
5237	sizeof(VBOXHDDBACKEND),
5238	/* uBackendCaps */
5239	VD_CAP_UUID \| VD_CAP_CREATE_FIXED \| VD_CAP_CREATE_DYNAMIC
5240	\| VD_CAP_CREATE_SPLIT_2G \| VD_CAP_DIFF \| VD_CAP_FILE \|VD_CAP_ASYNC,
5241	/* papszFileExtensions */
5242	s_apszVmdkFileExtensions,
5243	/* paConfigInfo */
5244	NULL,
5245	/* pfnCheckIfValid */
5246	vmdkCheckIfValid,
5247	/* pfnOpen */
5248	vmdkOpen,
5249	/* pfnCreate */
5250	vmdkCreate,
5251	/* pfnRename */
5252	vmdkRename,
5253	/* pfnClose */
5254	vmdkClose,
5255	/* pfnRead */
5256	vmdkRead,
5257	/* pfnWrite */
5258	vmdkWrite,
5259	/* pfnFlush */
5260	vmdkFlush,
5261	/* pfnGetVersion */
5262	vmdkGetVersion,
5263	/* pfnGetImageType */
5264	vmdkGetImageType,
5265	/* pfnGetSize */
5266	vmdkGetSize,
5267	/* pfnGetFileSize */
5268	vmdkGetFileSize,
5269	/* pfnGetPCHSGeometry */
5270	vmdkGetPCHSGeometry,
5271	/* pfnSetPCHSGeometry */
5272	vmdkSetPCHSGeometry,
5273	/* pfnGetLCHSGeometry */
5274	vmdkGetLCHSGeometry,
5275	/* pfnSetLCHSGeometry */
5276	vmdkSetLCHSGeometry,
5277	/* pfnGetImageFlags */
5278	vmdkGetImageFlags,
5279	/* pfnGetOpenFlags */
5280	vmdkGetOpenFlags,
5281	/* pfnSetOpenFlags */
5282	vmdkSetOpenFlags,
5283	/* pfnGetComment */
5284	vmdkGetComment,
5285	/* pfnSetComment */
5286	vmdkSetComment,
5287	/* pfnGetUuid */
5288	vmdkGetUuid,
5289	/* pfnSetUuid */
5290	vmdkSetUuid,
5291	/* pfnGetModificationUuid */
5292	vmdkGetModificationUuid,
5293	/* pfnSetModificationUuid */
5294	vmdkSetModificationUuid,
5295	/* pfnGetParentUuid */
5296	vmdkGetParentUuid,
5297	/* pfnSetParentUuid */
5298	vmdkSetParentUuid,
5299	/* pfnGetParentModificationUuid */
5300	vmdkGetParentModificationUuid,
5301	/* pfnSetParentModificationUuid */
5302	vmdkSetParentModificationUuid,
5303	/* pfnDump */
5304	vmdkDump,
5305	/* pfnGetTimeStamp */
5306	vmdkGetTimeStamp,
5307	/* pfnGetParentTimeStamp */
5308	vmdkGetParentTimeStamp,
5309	/* pfnSetParentTimeStamp */
5310	vmdkSetParentTimeStamp,
5311	/* pfnGetParentFilename */
5312	vmdkGetParentFilename,
5313	/* pfnSetParentFilename */
5314	vmdkSetParentFilename,
5315	/* pfnIsAsyncIOSupported */
5316	vmdkIsAsyncIOSupported,
5317	/* pfnAsyncRead */
5318	vmdkAsyncRead,
5319	/* pfnAsyncWrite */
5320	vmdkAsyncWrite
5321	};

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

source: vbox/trunk/src/VBox/Devices/Storage/VmdkHDDCore.cpp@ 12367

Download in other formats: