DBGPlugInLinux.cpp@ 86719

Last change on this file since 86719 was 84989, checked in by vboxsync, 4 years ago
Debugger/DBGPlugInLinux: Straighten the code to adjust the kernel log buffer size to not require a loop, bugref:1098
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 118.0 KB

Line
1	/* $Id: DBGPlugInLinux.cpp 84989 2020-06-29 11:12:40Z vboxsync $ */
2	/** @file
3	* DBGPlugInLinux - Debugger and Guest OS Digger Plugin For Linux.
4	*/
5
6	/*
7	* Copyright (C) 2008-2020 Oracle Corporation
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
18
19	/*********************************************************************************************************************************
20	* Header Files *
21	*********************************************************************************************************************************/
22	#define LOG_GROUP LOG_GROUP_DBGF /// @todo add new log group.
23	#include "DBGPlugIns.h"
24	#include "DBGPlugInCommonELF.h"
25	#include <VBox/vmm/dbgf.h>
26	#include <VBox/dis.h>
27	#include <iprt/ctype.h>
28	#include <iprt/file.h>
29	#include <iprt/err.h>
30	#include <iprt/mem.h>
31	#include <iprt/stream.h>
32	#include <iprt/string.h>
33	#include <iprt/vfs.h>
34	#include <iprt/zip.h>
35
36
37	/*********************************************************************************************************************************
38	* Structures and Typedefs *
39	*********************************************************************************************************************************/
40
41	/** @name InternalLinux structures
42	* @{ */
43
44
45	/** @} */
46
47
48	/**
49	* Config item type.
50	*/
51	typedef enum DBGDIGGERLINUXCFGITEMTYPE
52	{
53	/** Invalid type. */
54	DBGDIGGERLINUXCFGITEMTYPE_INVALID = 0,
55	/** String. */
56	DBGDIGGERLINUXCFGITEMTYPE_STRING,
57	/** Number. */
58	DBGDIGGERLINUXCFGITEMTYPE_NUMBER,
59	/** Flag whether this feature is included in the
60	* kernel or as a module. */
61	DBGDIGGERLINUXCFGITEMTYPE_FLAG
62	} DBGDIGGERLINUXCFGITEMTYPE;
63
64	/**
65	* Item in the config database.
66	*/
67	typedef struct DBGDIGGERLINUXCFGITEM
68	{
69	/** String space core. */
70	RTSTRSPACECORE Core;
71	/** Config item type. */
72	DBGDIGGERLINUXCFGITEMTYPE enmType;
73	/** Data based on the type. */
74	union
75	{
76	/** Number. */
77	int64_t i64Num;
78	/** Flag. */
79	bool fModule;
80	/** String - variable in size. */
81	char aszString[1];
82	} u;
83	} DBGDIGGERLINUXCFGITEM;
84	/** Pointer to a config database item. */
85	typedef DBGDIGGERLINUXCFGITEM *PDBGDIGGERLINUXCFGITEM;
86	/** Pointer to a const config database item. */
87	typedef const DBGDIGGERLINUXCFGITEM *PCDBGDIGGERLINUXCFGITEM;
88
89	/**
90	* Linux guest OS digger instance data.
91	*/
92	typedef struct DBGDIGGERLINUX
93	{
94	/** Whether the information is valid or not.
95	* (For fending off illegal interface method calls.) */
96	bool fValid;
97	/** Set if 64-bit, clear if 32-bit. */
98	bool f64Bit;
99	/** Set if the kallsyms table uses relative addressing, clear
100	* if absolute addresses are used. */
101	bool fRelKrnlAddr;
102	/** The relative base when kernel symbols use offsets rather than
103	* absolute addresses. */
104	RTGCUINTPTR uKernelRelativeBase;
105
106	/** The address of the linux banner.
107	* This is set during probing. */
108	DBGFADDRESS AddrLinuxBanner;
109	/** Kernel base address.
110	* This is set during probing, refined during kallsyms parsing. */
111	DBGFADDRESS AddrKernelBase;
112	/** The kernel size. */
113	uint32_t cbKernel;
114
115	/** The number of kernel symbols (kallsyms_num_syms).
116	* This is set during init. */
117	uint32_t cKernelSymbols;
118	/** The size of the kernel name table (sizeof(kallsyms_names)). */
119	uint32_t cbKernelNames;
120	/** Number of entries in the kernel_markers table. */
121	uint32_t cKernelNameMarkers;
122	/** The size of the kernel symbol token table. */
123	uint32_t cbKernelTokenTable;
124	/** The address of the encoded kernel symbol names (kallsyms_names). */
125	DBGFADDRESS AddrKernelNames;
126	/** The address of the kernel symbol addresses (kallsyms_addresses). */
127	DBGFADDRESS AddrKernelAddresses;
128	/** The address of the kernel symbol name markers (kallsyms_markers). */
129	DBGFADDRESS AddrKernelNameMarkers;
130	/** The address of the kernel symbol token table (kallsyms_token_table). */
131	DBGFADDRESS AddrKernelTokenTable;
132	/** The address of the kernel symbol token index table (kallsyms_token_index). */
133	DBGFADDRESS AddrKernelTokenIndex;
134
135	/** The kernel message log interface. */
136	DBGFOSIDMESG IDmesg;
137
138	/** The config database root. */
139	RTSTRSPACE hCfgDb;
140	} DBGDIGGERLINUX;
141	/** Pointer to the linux guest OS digger instance data. */
142	typedef DBGDIGGERLINUX *PDBGDIGGERLINUX;
143
144
145	/**
146	* The current printk_log structure.
147	*/
148	typedef struct LNXPRINTKHDR
149	{
150	/** Monotonic timestamp. */
151	uint64_t nsTimestamp;
152	/** The total size of this message record. */
153	uint16_t cbTotal;
154	/** The size of the text part (immediately follows the header). */
155	uint16_t cbText;
156	/** The size of the optional dictionary part (follows the text). */
157	uint16_t cbDict;
158	/** The syslog facility number. */
159	uint8_t bFacility;
160	/** First 5 bits are internal flags, next 3 bits are log level. */
161	uint8_t fFlagsAndLevel;
162	} LNXPRINTKHDR;
163	AssertCompileSize(LNXPRINTKHDR, 2*sizeof(uint64_t));
164	/** Pointer to linux printk_log header. */
165	typedef LNXPRINTKHDR *PLNXPRINTKHDR;
166	/** Pointer to linux const printk_log header. */
167	typedef LNXPRINTKHDR const *PCLNXPRINTKHDR;
168
169
170	/*********************************************************************************************************************************
171	* Defined Constants And Macros *
172	*********************************************************************************************************************************/
173	/** First kernel map address for 32bit Linux hosts (__START_KERNEL_map). */
174	#define LNX32_KERNEL_ADDRESS_START UINT32_C(0xc0000000)
175	/** First kernel map address for 64bit Linux hosts (__START_KERNEL_map). */
176	#define LNX64_KERNEL_ADDRESS_START UINT64_C(0xffffffff80000000)
177	/** Validates a 32-bit linux kernel address */
178	#define LNX32_VALID_ADDRESS(Addr) ((Addr) > UINT32_C(0x80000000) && (Addr) < UINT32_C(0xfffff000))
179	/** Validates a 64-bit linux kernel address */
180	#define LNX64_VALID_ADDRESS(Addr) ((Addr) > UINT64_C(0xffff800000000000) && (Addr) < UINT64_C(0xfffffffffffff000))
181
182	/** The max kernel size. */
183	#define LNX_MAX_KERNEL_SIZE UINT32_C(0x0f000000)
184	/** Maximum kernel log buffer size. */
185	#define LNX_MAX_KERNEL_LOG_SIZE (16 * _1M)
186
187	/** The maximum size we expect for kallsyms_names. */
188	#define LNX_MAX_KALLSYMS_NAMES_SIZE UINT32_C(0x200000)
189	/** The maximum size we expect for kallsyms_token_table. */
190	#define LNX_MAX_KALLSYMS_TOKEN_TABLE_SIZE UINT32_C(0x10000)
191	/** The minimum number of symbols we expect in kallsyms_num_syms. */
192	#define LNX_MIN_KALLSYMS_SYMBOLS UINT32_C(2048)
193	/** The maximum number of symbols we expect in kallsyms_num_syms. */
194	#define LNX_MAX_KALLSYMS_SYMBOLS UINT32_C(1048576)
195	/** The min length an encoded symbol in kallsyms_names is expected to have. */
196	#define LNX_MIN_KALLSYMS_ENC_LENGTH UINT8_C(1)
197	/** The max length an encoded symbol in kallsyms_names is expected to have.
198	* @todo check real life here. */
199	#define LNX_MAX_KALLSYMS_ENC_LENGTH UINT8_C(28)
200	/** The approximate maximum length of a string token. */
201	#define LNX_MAX_KALLSYMS_TOKEN_LEN UINT16_C(32)
202	/** Maximum compressed config size expected. */
203	#define LNX_MAX_COMPRESSED_CFG_SIZE _1M
204
205	/** Module tag for linux ('linuxmod' on little endian ASCII systems). */
206	#define DIG_LNX_MOD_TAG UINT64_C(0x545f5d78758e898c)
207
208
209	/*********************************************************************************************************************************
210	* Internal Functions *
211	*********************************************************************************************************************************/
212	static DECLCALLBACK(int) dbgDiggerLinuxInit(PUVM pUVM, void *pvData);
213
214
215	/*********************************************************************************************************************************
216	* Global Variables *
217	*********************************************************************************************************************************/
218	/** Table of common linux kernel addresses. */
219	static uint64_t g_au64LnxKernelAddresses[] =
220	{
221	UINT64_C(0xc0100000),
222	UINT64_C(0x90100000),
223	UINT64_C(0xffffffff80200000)
224	};
225
226	static const uint8_t g_abLinuxVersion[] = "Linux version ";
227	/** The needle for searching for the kernel log area (the value is observed in pretty much all 32bit and 64bit x86 kernels).
228	* This needle should appear only once in the memory due to the address being filled in by a format string. */
229	static const uint8_t g_abKrnlLogNeedle[] = "BIOS-e820: [mem 0x0000000000000000";
230
231
232	/**
233	* Tries to resolve the kernel log buffer start and end by searching for needle.
234	*
235	* @returns VBox status code.
236	* @param pThis The Linux digger data.
237	* @param pUVM The VM handle.
238	* @param pGCPtrLogBuf Where to store the start of the kernel log buffer on success.
239	* @param pcbLogBuf Where to store the size of the kernel log buffer on success.
240	*/
241	static int dbgDiggerLinuxKrnlLogBufFindByNeedle(PDBGDIGGERLINUX pThis, PUVM pUVM, RTGCPTR pGCPtrLogBuf, uint32_t pcbLogBuf)
242	{
243	int rc = VINF_SUCCESS;
244
245	/* Try to find the needle, it should be very early in the kernel log buffer. */
246	DBGFADDRESS AddrScan;
247	DBGFADDRESS AddrHit;
248	DBGFR3AddrFromFlat(pUVM, &AddrScan, pThis->f64Bit ? LNX64_KERNEL_ADDRESS_START : LNX32_KERNEL_ADDRESS_START);
249
250	rc = DBGFR3MemScan(pUVM, 0 /idCpu/, &AddrScan, ~(RTGCUINTPTR)0, 1 /uAlign/,
251	g_abKrnlLogNeedle, sizeof(g_abKrnlLogNeedle) - 1, &AddrHit);
252	if (RT_SUCCESS(rc))
253	{
254	uint32_t cbLogBuf = 0;
255	uint64_t tsLastNs = 0;
256	DBGFADDRESS AddrCur;
257
258	DBGFR3AddrSub(&AddrHit, sizeof(LNXPRINTKHDR));
259	AddrCur = AddrHit;
260
261	/* Try to find the end of the kernel log buffer. */
262	for (;;)
263	{
264	if (cbLogBuf >= LNX_MAX_KERNEL_LOG_SIZE)
265	break;
266
267	LNXPRINTKHDR Hdr;
268	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &AddrCur, &Hdr, sizeof(Hdr));
269	if (RT_SUCCESS(rc))
270	{
271	uint32_t const cbLogAlign = 4;
272
273	/*
274	* If the header does not look valid anymore we stop.
275	* Timestamps are monotonically increasing.
276	*/
277	if ( !Hdr.cbTotal /* Zero entry size means there is no record anymore, doesn't make sense to look futher. */
278	\|\| Hdr.cbText + Hdr.cbDict + sizeof(Hdr) > Hdr.cbTotal
279	\|\| (Hdr.cbTotal & (cbLogAlign - 1)) != 0
280	\|\| tsLastNs > Hdr.nsTimestamp)
281	break;
282
283	/** @todo Maybe read text part and verify it is all ASCII. */
284
285	cbLogBuf += Hdr.cbTotal;
286	DBGFR3AddrAdd(&AddrCur, Hdr.cbTotal);
287	}
288
289	if (RT_FAILURE(rc))
290	break;
291	}
292
293	/** @todo Go back to find the start address of the kernel log (or we loose potential kernel log messages). */
294
295	if ( RT_SUCCESS(rc)
296	&& cbLogBuf)
297	{
298	/* Align log buffer size to a power of two. */
299	uint32_t idxBitLast = ASMBitLastSetU32(cbLogBuf);
300	idxBitLast--; /* There is at least one bit set, see check above. */
301
302	if (cbLogBuf & (RT_BIT_32(idxBitLast) - 1))
303	idxBitLast++;
304
305	*pGCPtrLogBuf = AddrHit.FlatPtr;
306	*pcbLogBuf = RT_MIN(RT_BIT_32(idxBitLast), LNX_MAX_KERNEL_LOG_SIZE);
307	}
308	else if (RT_SUCCESS(rc))
309	rc = VERR_NOT_FOUND;
310	}
311
312	return rc;
313	}
314
315
316	/**
317	* Converts a given offset into an absolute address if relative kernel offsets are used for
318	* kallsyms.
319	*
320	* @returns The absolute kernel address.
321	* @param pThis The Linux digger data.
322	* @param uOffset The offset to convert.
323	*/
324	DECLINLINE(RTGCUINTPTR) dbgDiggerLinuxConvOffsetToAddr(PDBGDIGGERLINUX pThis, int32_t uOffset)
325	{
326	RTGCUINTPTR uAddr;
327
328	/*
329	* How the absolute address is calculated from the offset depends on the
330	* CONFIG_KALLSYMS_ABSOLUTE_PERCPU config which is only set for 64bit
331	* SMP kernels (we assume that all 64bit kernels always have SMP enabled too).
332	*/
333	if (pThis->f64Bit)
334	{
335	if (uOffset >= 0)
336	uAddr = uOffset;
337	else
338	uAddr = pThis->uKernelRelativeBase - 1 - uOffset;
339	}
340	else
341	uAddr = pThis->uKernelRelativeBase + (uint32_t)uOffset;
342
343	return uAddr;
344	}
345
346	/**
347	* Disassembles a simple getter returning the value for it.
348	*
349	* @returns VBox status code.
350	* @param pThis The Linux digger data.
351	* @param pUVM The VM handle.
352	* @param hMod The module to use.
353	* @param pszSymbol The symbol of the getter.
354	* @param pvVal Where to store the value on success.
355	* @param cbVal Size of the value in bytes.
356	*/
357	static int dbgDiggerLinuxDisassembleSimpleGetter(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
358	const char pszSymbol, void pvVal, uint32_t cbVal)
359	{
360	int rc = VINF_SUCCESS;
361
362	RTDBGSYMBOL SymInfo;
363	rc = RTDbgModSymbolByName(hMod, pszSymbol, &SymInfo);
364	if (RT_SUCCESS(rc))
365	{
366	/*
367	* Do the diassembling. Disassemble until a ret instruction is encountered
368	* or a limit is reached (don't want to disassemble for too long as the getter
369	* should be short).
370	* push and pop instructions are skipped as well as any mov instructions not
371	* touching the rax or eax register (depending on the size of the value).
372	*/
373	unsigned cInstrDisassembled = 0;
374	uint32_t offInstr = 0;
375	bool fRet = false;
376	DISSTATE DisState;
377	RT_ZERO(DisState);
378
379	do
380	{
381	DBGFADDRESS Addr;
382	RTGCPTR GCPtrCur = (RTGCPTR)SymInfo.Value + pThis->AddrKernelBase.FlatPtr + offInstr;
383	DBGFR3AddrFromFlat(pUVM, &Addr, GCPtrCur);
384
385	/* Prefetch the instruction. */
386	uint8_t abInstr[32];
387	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &Addr, &abInstr[0], sizeof(abInstr));
388	if (RT_SUCCESS(rc))
389	{
390	uint32_t cbInstr = 0;
391
392	rc = DISInstr(&abInstr[0], pThis->f64Bit ? DISCPUMODE_64BIT : DISCPUMODE_32BIT, &DisState, &cbInstr);
393	if (RT_SUCCESS(rc))
394	{
395	switch (DisState.pCurInstr->uOpcode)
396	{
397	case OP_PUSH:
398	case OP_POP:
399	case OP_NOP:
400	case OP_LEA:
401	break;
402	case OP_RETN:
403	/* Getter returned, abort disassembling. */
404	fRet = true;
405	break;
406	case OP_MOV:
407	/*
408	* Check that the destination is either rax or eax depending on the
409	* value size.
410	*
411	* Param1 is the destination and Param2 the source.
412	*/
413	if ( ( ( (DisState.Param1.fUse & (DISUSE_BASE \| DISUSE_REG_GEN32))
414	&& cbVal == sizeof(uint32_t))
415	\|\| ( (DisState.Param1.fUse & (DISUSE_BASE \| DISUSE_REG_GEN64))
416	&& cbVal == sizeof(uint64_t)))
417	&& DisState.Param1.Base.idxGenReg == DISGREG_RAX)
418	{
419	/* Parse the source. */
420	if (DisState.Param2.fUse & (DISUSE_IMMEDIATE32 \| DISUSE_IMMEDIATE64))
421	memcpy(pvVal, &DisState.Param2.uValue, cbVal);
422	else if (DisState.Param2.fUse & (DISUSE_RIPDISPLACEMENT32\|DISUSE_DISPLACEMENT32\|DISUSE_DISPLACEMENT64))
423	{
424	RTGCPTR GCPtrVal = 0;
425
426	if (DisState.Param2.fUse & DISUSE_RIPDISPLACEMENT32)
427	GCPtrVal = GCPtrCur + DisState.Param2.uDisp.i32 + cbInstr;
428	else if (DisState.Param2.fUse & DISUSE_DISPLACEMENT32)
429	GCPtrVal = (RTGCPTR)DisState.Param2.uDisp.u32;
430	else if (DisState.Param2.fUse & DISUSE_DISPLACEMENT64)
431	GCPtrVal = (RTGCPTR)DisState.Param2.uDisp.u64;
432	else
433	AssertMsgFailedBreakStmt(("Invalid displacement\n"), rc = VERR_INVALID_STATE);
434
435	DBGFADDRESS AddrVal;
436	rc = DBGFR3MemRead(pUVM, 0 /idCpu/,
437	DBGFR3AddrFromFlat(pUVM, &AddrVal, GCPtrVal),
438	pvVal, cbVal);
439	}
440	}
441	break;
442	default:
443	/* All other instructions will cause an error for now (playing safe here). */
444	rc = VERR_INVALID_PARAMETER;
445	break;
446	}
447	cInstrDisassembled++;
448	offInstr += cbInstr;
449	}
450	}
451	} while ( RT_SUCCESS(rc)
452	&& cInstrDisassembled < 20
453	&& !fRet);
454	}
455
456	return rc;
457	}
458
459	/**
460	* Try to get at the log buffer starting address and size by disassembling emit_log_char.
461	*
462	* @returns VBox status code.
463	* @param pThis The Linux digger data.
464	* @param pUVM The VM handle.
465	* @param hMod The module to use.
466	* @param pGCPtrLogBuf Where to store the log buffer pointer on success.
467	* @param pcbLogBuf Where to store the size of the log buffer on success.
468	*/
469	static int dbgDiggerLinuxQueryAsciiLogBufferPtrs(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
470	RTGCPTR pGCPtrLogBuf, uint32_t pcbLogBuf)
471	{
472	int rc = VINF_SUCCESS;
473
474	/**
475	* We disassemble emit_log_char to get at the log buffer address and size.
476	* This is used in case the symbols are not exported in kallsyms.
477	*
478	* This is what it typically looks like:
479	* vmlinux!emit_log_char:
480	* %00000000c01204a1 56 push esi
481	* %00000000c01204a2 8b 35 d0 1c 34 c0 mov esi, dword [0c0341cd0h]
482	* %00000000c01204a8 53 push ebx
483	* %00000000c01204a9 8b 1d 74 3b 3e c0 mov ebx, dword [0c03e3b74h]
484	* %00000000c01204af 8b 0d d8 1c 34 c0 mov ecx, dword [0c0341cd8h]
485	* %00000000c01204b5 8d 56 ff lea edx, [esi-001h]
486	* %00000000c01204b8 21 da and edx, ebx
487	* %00000000c01204ba 88 04 11 mov byte [ecx+edx], al
488	* %00000000c01204bd 8d 53 01 lea edx, [ebx+001h]
489	* %00000000c01204c0 89 d0 mov eax, edx
490	* [...]
491	*/
492	RTDBGSYMBOL SymInfo;
493	rc = RTDbgModSymbolByName(hMod, "emit_log_char", &SymInfo);
494	if (RT_SUCCESS(rc))
495	{
496	/*
497	* Do the diassembling. Disassemble until a ret instruction is encountered
498	* or a limit is reached (don't want to disassemble for too long as the getter
499	* should be short). Certain instructions found are ignored (push, nop, etc.).
500	*/
501	unsigned cInstrDisassembled = 0;
502	uint32_t offInstr = 0;
503	bool fRet = false;
504	DISSTATE DisState;
505	unsigned cAddressesUsed = 0;
506	struct { size_t cb; RTGCPTR GCPtrOrigSrc; } aAddresses[5];
507	RT_ZERO(DisState);
508	RT_ZERO(aAddresses);
509
510	do
511	{
512	DBGFADDRESS Addr;
513	RTGCPTR GCPtrCur = (RTGCPTR)SymInfo.Value + pThis->AddrKernelBase.FlatPtr + offInstr;
514	DBGFR3AddrFromFlat(pUVM, &Addr, GCPtrCur);
515
516	/* Prefetch the instruction. */
517	uint8_t abInstr[32];
518	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &Addr, &abInstr[0], sizeof(abInstr));
519	if (RT_SUCCESS(rc))
520	{
521	uint32_t cbInstr = 0;
522
523	rc = DISInstr(&abInstr[0], pThis->f64Bit ? DISCPUMODE_64BIT : DISCPUMODE_32BIT, &DisState, &cbInstr);
524	if (RT_SUCCESS(rc))
525	{
526	switch (DisState.pCurInstr->uOpcode)
527	{
528	case OP_PUSH:
529	case OP_POP:
530	case OP_NOP:
531	case OP_LEA:
532	case OP_AND:
533	case OP_CBW:
534	case OP_DEC:
535	break;
536	case OP_RETN:
537	/* emit_log_char returned, abort disassembling. */
538	rc = VERR_NOT_FOUND;
539	fRet = true;
540	break;
541	case OP_MOV:
542	case OP_MOVSXD:
543	/*
544	* If a mov is encountered writing to memory with al (or dil for amd64) being the source the
545	* character is stored and we can infer the base address and size of the log buffer from
546	* the source addresses.
547	*/
548	if ( (DisState.Param2.fUse & DISUSE_REG_GEN8)
549	&& ( (DisState.Param2.Base.idxGenReg == DISGREG_AL && !pThis->f64Bit)
550	\|\| (DisState.Param2.Base.idxGenReg == DISGREG_DIL && pThis->f64Bit))
551	&& DISUSE_IS_EFFECTIVE_ADDR(DisState.Param1.fUse))
552	{
553	RTGCPTR GCPtrLogBuf = 0;
554	uint32_t cbLogBuf = 0;
555
556	/*
557	* We can stop disassembling now and inspect all registers, look for a valid kernel address first.
558	* Only one of the accessed registers should hold a valid kernel address.
559	* For the log size look for the biggest non kernel address.
560	*/
561	for (unsigned i = 0; i < cAddressesUsed; i++)
562	{
563	DBGFADDRESS AddrVal;
564	union { uint8_t abVal[8]; uint32_t u32Val; uint64_t u64Val; } Val;
565
566	rc = DBGFR3MemRead(pUVM, 0 /idCpu/,
567	DBGFR3AddrFromFlat(pUVM, &AddrVal, aAddresses[i].GCPtrOrigSrc),
568	&Val.abVal[0], aAddresses[i].cb);
569	if (RT_SUCCESS(rc))
570	{
571	if (pThis->f64Bit && aAddresses[i].cb == sizeof(uint64_t))
572	{
573	if (LNX64_VALID_ADDRESS(Val.u64Val))
574	{
575	if (GCPtrLogBuf == 0)
576	GCPtrLogBuf = Val.u64Val;
577	else
578	{
579	rc = VERR_NOT_FOUND;
580	break;
581	}
582	}
583	}
584	else
585	{
586	AssertMsgBreakStmt(aAddresses[i].cb == sizeof(uint32_t),
587	("Invalid value size\n"), rc = VERR_INVALID_STATE);
588
589	/* Might be a kernel address or a size indicator. */
590	if (!pThis->f64Bit && LNX32_VALID_ADDRESS(Val.u32Val))
591	{
592	if (GCPtrLogBuf == 0)
593	GCPtrLogBuf = Val.u32Val;
594	else
595	{
596	rc = VERR_NOT_FOUND;
597	break;
598	}
599	}
600	else
601	{
602	/*
603	* The highest value will be the log buffer because the other
604	* accessed variables are indexes into the buffer and hence
605	* always smaller than the size.
606	*/
607	if (cbLogBuf < Val.u32Val)
608	cbLogBuf = Val.u32Val;
609	}
610	}
611	}
612	}
613
614	if ( RT_SUCCESS(rc)
615	&& GCPtrLogBuf != 0
616	&& cbLogBuf != 0)
617	{
618	*pGCPtrLogBuf = GCPtrLogBuf;
619	*pcbLogBuf = cbLogBuf;
620	}
621	else if (RT_SUCCESS(rc))
622	rc = VERR_NOT_FOUND;
623
624	fRet = true;
625	break;
626	}
627	else
628	{
629	/*
630	* In case of a memory to register move store the destination register index and the
631	* source address in the relation table for later processing.
632	*/
633	if ( (DisState.Param1.fUse & (DISUSE_BASE \| DISUSE_REG_GEN32 \| DISUSE_REG_GEN64))
634	&& (DisState.Param2.cb == sizeof(uint32_t) \|\| DisState.Param2.cb == sizeof(uint64_t))
635	&& (DisState.Param2.fUse & (DISUSE_RIPDISPLACEMENT32\|DISUSE_DISPLACEMENT32\|DISUSE_DISPLACEMENT64)))
636	{
637	RTGCPTR GCPtrVal = 0;
638
639	if (DisState.Param2.fUse & DISUSE_RIPDISPLACEMENT32)
640	GCPtrVal = GCPtrCur + DisState.Param2.uDisp.i32 + cbInstr;
641	else if (DisState.Param2.fUse & DISUSE_DISPLACEMENT32)
642	GCPtrVal = (RTGCPTR)DisState.Param2.uDisp.u32;
643	else if (DisState.Param2.fUse & DISUSE_DISPLACEMENT64)
644	GCPtrVal = (RTGCPTR)DisState.Param2.uDisp.u64;
645	else
646	AssertMsgFailedBreakStmt(("Invalid displacement\n"), rc = VERR_INVALID_STATE);
647
648	if (cAddressesUsed < RT_ELEMENTS(aAddresses))
649	{
650	/* movsxd reads always 32bits. */
651	if (DisState.pCurInstr->uOpcode == OP_MOVSXD)
652	aAddresses[cAddressesUsed].cb = sizeof(uint32_t);
653	else
654	aAddresses[cAddressesUsed].cb = DisState.Param2.cb;
655	aAddresses[cAddressesUsed].GCPtrOrigSrc = GCPtrVal;
656	cAddressesUsed++;
657	}
658	else
659	{
660	rc = VERR_INVALID_PARAMETER;
661	break;
662	}
663	}
664	}
665	break;
666	default:
667	/* All other instructions will cause an error for now (playing safe here). */
668	rc = VERR_INVALID_PARAMETER;
669	break;
670	}
671	cInstrDisassembled++;
672	offInstr += cbInstr;
673	}
674	}
675	} while ( RT_SUCCESS(rc)
676	&& cInstrDisassembled < 20
677	&& !fRet);
678	}
679
680	return rc;
681	}
682
683	/**
684	* Try to get at the log buffer starting address and size by disassembling some exposed helpers.
685	*
686	* @returns VBox status code.
687	* @param pThis The Linux digger data.
688	* @param pUVM The VM handle.
689	* @param hMod The module to use.
690	* @param pGCPtrLogBuf Where to store the log buffer pointer on success.
691	* @param pcbLogBuf Where to store the size of the log buffer on success.
692	*/
693	static int dbgDiggerLinuxQueryLogBufferPtrs(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
694	RTGCPTR pGCPtrLogBuf, uint32_t pcbLogBuf)
695	{
696	int rc = VINF_SUCCESS;
697
698	struct { void pvVar; uint32_t cbHost, cbGuest; const char pszSymbol; } aSymbols[] =
699	{
700	{ pGCPtrLogBuf, (uint32_t)sizeof(RTGCPTR), (uint32_t)(pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t)), "log_buf_addr_get" },
701	{ pcbLogBuf, (uint32_t)sizeof(uint32_t), (uint32_t)sizeof(uint32_t), "log_buf_len_get" }
702	};
703	for (uint32_t i = 0; i < RT_ELEMENTS(aSymbols) && RT_SUCCESS(rc); i++)
704	{
705	RT_BZERO(aSymbols[i].pvVar, aSymbols[i].cbHost);
706	Assert(aSymbols[i].cbHost >= aSymbols[i].cbGuest);
707	rc = dbgDiggerLinuxDisassembleSimpleGetter(pThis, pUVM, hMod, aSymbols[i].pszSymbol,
708	aSymbols[i].pvVar, aSymbols[i].cbGuest);
709	}
710
711	return rc;
712	}
713
714	/**
715	* Returns whether the log buffer is a simple ascii buffer or a record based implementation
716	* based on the kernel version found.
717	*
718	* @returns Flag whether the log buffer is the simple ascii buffer.
719	* @param pThis The Linux digger data.
720	* @param pUVM The user mode VM handle.
721	*/
722	static bool dbgDiggerLinuxLogBufferIsAsciiBuffer(PDBGDIGGERLINUX pThis, PUVM pUVM)
723	{
724	char szTmp[128];
725	char const *pszVer = &szTmp[sizeof(g_abLinuxVersion) - 1];
726
727	RT_ZERO(szTmp);
728	int rc = DBGFR3MemReadString(pUVM, 0, &pThis->AddrLinuxBanner, szTmp, sizeof(szTmp) - 1);
729	if ( RT_SUCCESS(rc)
730	&& RTStrVersionCompare(pszVer, "3.4") == -1)
731	return true;
732
733	return false;
734	}
735
736	/**
737	* Worker to get at the kernel log for pre 3.4 kernels where the log buffer was just a char buffer.
738	*
739	* @returns VBox status code.
740	* @param pThis The Linux digger data.
741	* @param pUVM The VM user mdoe handle.
742	* @param hMod The debug module handle.
743	* @param fFlags Flags reserved for future use, MBZ.
744	* @param cMessages The number of messages to retrieve, counting from the
745	* end of the log (i.e. like tail), use UINT32_MAX for all.
746	* @param pszBuf The output buffer.
747	* @param cbBuf The buffer size.
748	* @param pcbActual Where to store the number of bytes actually returned,
749	* including zero terminator. On VERR_BUFFER_OVERFLOW this
750	* holds the necessary buffer size. Optional.
751	*/
752	static int dbgDiggerLinuxLogBufferQueryAscii(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
753	uint32_t fFlags, uint32_t cMessages,
754	char pszBuf, size_t cbBuf, size_t pcbActual)
755	{
756	RT_NOREF2(fFlags, cMessages);
757	int rc = VINF_SUCCESS;
758	RTGCPTR GCPtrLogBuf;
759	uint32_t cbLogBuf;
760
761	struct { void pvVar; size_t cbHost, cbGuest; const char pszSymbol; } aSymbols[] =
762	{
763	{ &GCPtrLogBuf, sizeof(GCPtrLogBuf), pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t), "log_buf" },
764	{ &cbLogBuf, sizeof(cbLogBuf), sizeof(cbLogBuf), "log_buf_len" },
765	};
766	for (uint32_t i = 0; i < RT_ELEMENTS(aSymbols); i++)
767	{
768	RTDBGSYMBOL SymInfo;
769	rc = RTDbgModSymbolByName(hMod, aSymbols[i].pszSymbol, &SymInfo);
770	if (RT_SUCCESS(rc))
771	{
772	RT_BZERO(aSymbols[i].pvVar, aSymbols[i].cbHost);
773	Assert(aSymbols[i].cbHost >= aSymbols[i].cbGuest);
774	DBGFADDRESS Addr;
775	rc = DBGFR3MemRead(pUVM, 0 /idCpu/,
776	DBGFR3AddrFromFlat(pUVM, &Addr, (RTGCPTR)SymInfo.Value + pThis->AddrKernelBase.FlatPtr),
777	aSymbols[i].pvVar, aSymbols[i].cbGuest);
778	if (RT_SUCCESS(rc))
779	continue;
780	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Reading '%s' at %RGv: %Rrc\n", aSymbols[i].pszSymbol, Addr.FlatPtr, rc));
781	}
782	else
783	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Error looking up '%s': %Rrc\n", aSymbols[i].pszSymbol, rc));
784	rc = VERR_NOT_FOUND;
785	break;
786	}
787
788	/*
789	* Some kernels don't expose the variables in kallsyms so we have to try disassemble
790	* some public helpers to get at the addresses.
791	*
792	* @todo: Maybe cache those values so we don't have to do the heavy work every time?
793	*/
794	if (rc == VERR_NOT_FOUND)
795	{
796	rc = dbgDiggerLinuxQueryAsciiLogBufferPtrs(pThis, pUVM, hMod, &GCPtrLogBuf, &cbLogBuf);
797	if (RT_FAILURE(rc))
798	return rc;
799	}
800
801	/*
802	* Check if the values make sense.
803	*/
804	if (pThis->f64Bit ? !LNX64_VALID_ADDRESS(GCPtrLogBuf) : !LNX32_VALID_ADDRESS(GCPtrLogBuf))
805	{
806	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_buf' value %RGv is not valid.\n", GCPtrLogBuf));
807	return VERR_NOT_FOUND;
808	}
809	if ( cbLogBuf < 4096
810	\|\| !RT_IS_POWER_OF_TWO(cbLogBuf)
811	\|\| cbLogBuf > 16*_1M)
812	{
813	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_buf_len' value %#x is not valid.\n", cbLogBuf));
814	return VERR_NOT_FOUND;
815	}
816
817	/*
818	* Read the whole log buffer.
819	*/
820	uint8_t pbLogBuf = (uint8_t )RTMemAlloc(cbLogBuf);
821	if (!pbLogBuf)
822	{
823	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Failed to allocate %#x bytes for log buffer\n", cbLogBuf));
824	return VERR_NO_MEMORY;
825	}
826	DBGFADDRESS Addr;
827	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, DBGFR3AddrFromFlat(pUVM, &Addr, GCPtrLogBuf), pbLogBuf, cbLogBuf);
828	if (RT_FAILURE(rc))
829	{
830	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Error reading %#x bytes of log buffer at %RGv: %Rrc\n",
831	cbLogBuf, Addr.FlatPtr, rc));
832	RTMemFree(pbLogBuf);
833	return VERR_NOT_FOUND;
834	}
835
836	/** @todo Try to parse where the single messages start to make use of cMessages. */
837	size_t cchLength = RTStrNLen((const char *)pbLogBuf, cbLogBuf);
838	memcpy(&pszBuf[0], pbLogBuf, RT_MIN(cbBuf, cchLength));
839
840	/* Done with the buffer. */
841	RTMemFree(pbLogBuf);
842
843	/* Set return size value. */
844	if (pcbActual)
845	*pcbActual = RT_MIN(cbBuf, cchLength);
846
847	return cbBuf <= cchLength ? VERR_BUFFER_OVERFLOW : VINF_SUCCESS;
848	}
849
850
851	/**
852	* Worker to process a given record based kernel log.
853	*
854	* @returns VBox status code.
855	* @param pThis The Linux digger data.
856	* @param pUVM The VM user mode handle.
857	* @param GCPtrLogBuf Flat guest address of the start of the log buffer.
858	* @param cbLogBuf Power of two aligned size of the log buffer.
859	* @param idxFirst Index in the log bfufer of the first message.
860	* @param idxNext Index where to write hte next message in the log buffer.
861	* @param fFlags Flags reserved for future use, MBZ.
862	* @param cMessages The number of messages to retrieve, counting from the
863	* end of the log (i.e. like tail), use UINT32_MAX for all.
864	* @param pszBuf The output buffer.
865	* @param cbBuf The buffer size.
866	* @param pcbActual Where to store the number of bytes actually returned,
867	* including zero terminator. On VERR_BUFFER_OVERFLOW this
868	* holds the necessary buffer size. Optional.
869	*/
870	static int dbgDiggerLinuxKrnLogBufferProcess(PDBGDIGGERLINUX pThis, PUVM pUVM, RTGCPTR GCPtrLogBuf,
871	uint32_t cbLogBuf, uint32_t idxFirst, uint32_t idxNext,
872	uint32_t fFlags, uint32_t cMessages, char *pszBuf, size_t cbBuf,
873	size_t *pcbActual)
874	{
875	RT_NOREF(fFlags);
876
877	/*
878	* Check if the values make sense.
879	*/
880	if (pThis->f64Bit ? !LNX64_VALID_ADDRESS(GCPtrLogBuf) : !LNX32_VALID_ADDRESS(GCPtrLogBuf))
881	{
882	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_buf' value %RGv is not valid.\n", GCPtrLogBuf));
883	return VERR_NOT_FOUND;
884	}
885	if ( cbLogBuf < _4K
886	\|\| !RT_IS_POWER_OF_TWO(cbLogBuf)
887	\|\| cbLogBuf > LNX_MAX_KERNEL_LOG_SIZE)
888	{
889	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_buf_len' value %#x is not valid.\n", cbLogBuf));
890	return VERR_NOT_FOUND;
891	}
892	uint32_t const cbLogAlign = 4;
893	if ( idxFirst > cbLogBuf - sizeof(LNXPRINTKHDR)
894	\|\| (idxFirst & (cbLogAlign - 1)) != 0)
895	{
896	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_first_idx' value %#x is not valid.\n", idxFirst));
897	return VERR_NOT_FOUND;
898	}
899	if ( idxNext > cbLogBuf - sizeof(LNXPRINTKHDR)
900	\|\| (idxNext & (cbLogAlign - 1)) != 0)
901	{
902	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_next_idx' value %#x is not valid.\n", idxNext));
903	return VERR_NOT_FOUND;
904	}
905
906	/*
907	* Read the whole log buffer.
908	*/
909	uint8_t pbLogBuf = (uint8_t )RTMemAlloc(cbLogBuf);
910	if (!pbLogBuf)
911	{
912	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Failed to allocate %#x bytes for log buffer\n", cbLogBuf));
913	return VERR_NO_MEMORY;
914	}
915	DBGFADDRESS Addr;
916	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, DBGFR3AddrFromFlat(pUVM, &Addr, GCPtrLogBuf), pbLogBuf, cbLogBuf);
917	if (RT_FAILURE(rc))
918	{
919	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Error reading %#x bytes of log buffer at %RGv: %Rrc\n",
920	cbLogBuf, Addr.FlatPtr, rc));
921	RTMemFree(pbLogBuf);
922	return VERR_NOT_FOUND;
923	}
924
925	/*
926	* Count the messages in the buffer while doing some basic validation.
927	*/
928	uint32_t const cbUsed = idxFirst == idxNext ? cbLogBuf /* could be empty... */
929	: idxFirst < idxNext ? idxNext - idxFirst : cbLogBuf - idxFirst + idxNext;
930	uint32_t cbLeft = cbUsed;
931	uint32_t offCur = idxFirst;
932	uint32_t cLogMsgs = 0;
933
934	while (cbLeft > 0)
935	{
936	PCLNXPRINTKHDR pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
937	if (!pHdr->cbTotal)
938	{
939	/* Wrap around packet, most likely... */
940	if (cbLogBuf - offCur >= cbLeft)
941	break;
942	offCur = 0;
943	pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
944	}
945	if (RT_UNLIKELY( pHdr->cbTotal > cbLogBuf - sizeof(*pHdr) - offCur
946	\|\| pHdr->cbTotal > cbLeft
947	\|\| (pHdr->cbTotal & (cbLogAlign - 1)) != 0
948	\|\| pHdr->cbTotal < (uint32_t)pHdr->cbText + (uint32_t)pHdr->cbDict + sizeof(*pHdr) ))
949	{
950	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Invalid printk_log record at %#x: cbTotal=%#x cbText=%#x cbDict=%#x cbLogBuf=%#x cbLeft=%#x\n",
951	offCur, pHdr->cbTotal, pHdr->cbText, pHdr->cbDict, cbLogBuf, cbLeft));
952	break;
953	}
954
955	if (pHdr->cbText > 0)
956	cLogMsgs++;
957
958	/* next */
959	offCur += pHdr->cbTotal;
960	cbLeft -= pHdr->cbTotal;
961	}
962	if (!cLogMsgs)
963	{
964	RTMemFree(pbLogBuf);
965	return VERR_NOT_FOUND;
966	}
967
968	/*
969	* Copy the messages into the output buffer.
970	*/
971	offCur = idxFirst;
972	cbLeft = cbUsed - cbLeft;
973
974	/* Skip messages that the caller doesn't want. */
975	if (cMessages < cLogMsgs)
976	{
977	uint32_t cToSkip = cLogMsgs - cMessages;
978	cLogMsgs -= cToSkip;
979
980	while (cToSkip > 0)
981	{
982	PCLNXPRINTKHDR pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
983	if (!pHdr->cbTotal)
984	{
985	offCur = 0;
986	pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
987	}
988	if (pHdr->cbText > 0)
989	cToSkip--;
990
991	/* next */
992	offCur += pHdr->cbTotal;
993	cbLeft -= pHdr->cbTotal;
994	}
995	}
996
997	/* Now copy the messages. */
998	size_t offDst = 0;
999	while (cbLeft > 0)
1000	{
1001	PCLNXPRINTKHDR pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
1002	if ( !pHdr->cbTotal
1003	\|\| !cLogMsgs)
1004	{
1005	if (cbLogBuf - offCur >= cbLeft)
1006	break;
1007	offCur = 0;
1008	pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
1009	}
1010
1011	if (pHdr->cbText > 0)
1012	{
1013	char pchText = (char )(pHdr + 1);
1014	size_t cchText = RTStrNLen(pchText, pHdr->cbText);
1015	if (offDst + cchText < cbBuf)
1016	{
1017	memcpy(&pszBuf[offDst], pHdr + 1, cchText);
1018	pszBuf[offDst + cchText] = '\n';
1019	}
1020	else if (offDst < cbBuf)
1021	memcpy(&pszBuf[offDst], pHdr + 1, cbBuf - offDst);
1022	offDst += cchText + 1;
1023	}
1024
1025	/* next */
1026	offCur += pHdr->cbTotal;
1027	cbLeft -= pHdr->cbTotal;
1028	}
1029
1030	/* Done with the buffer. */
1031	RTMemFree(pbLogBuf);
1032
1033	/* Make sure we've reserved a char for the terminator. */
1034	if (!offDst)
1035	offDst = 1;
1036
1037	/* Set return size value. */
1038	if (pcbActual)
1039	*pcbActual = offDst;
1040
1041	if (offDst <= cbBuf)
1042	return VINF_SUCCESS;
1043	return VERR_BUFFER_OVERFLOW;
1044	}
1045
1046
1047	/**
1048	* Worker to get at the kernel log for post 3.4 kernels where the log buffer contains records.
1049	*
1050	* @returns VBox status code.
1051	* @param pThis The Linux digger data.
1052	* @param pUVM The VM user mdoe handle.
1053	* @param hMod The debug module handle.
1054	* @param fFlags Flags reserved for future use, MBZ.
1055	* @param cMessages The number of messages to retrieve, counting from the
1056	* end of the log (i.e. like tail), use UINT32_MAX for all.
1057	* @param pszBuf The output buffer.
1058	* @param cbBuf The buffer size.
1059	* @param pcbActual Where to store the number of bytes actually returned,
1060	* including zero terminator. On VERR_BUFFER_OVERFLOW this
1061	* holds the necessary buffer size. Optional.
1062	*/
1063	static int dbgDiggerLinuxLogBufferQueryRecords(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
1064	uint32_t fFlags, uint32_t cMessages,
1065	char pszBuf, size_t cbBuf, size_t pcbActual)
1066	{
1067	int rc = VINF_SUCCESS;
1068	RTGCPTR GCPtrLogBuf;
1069	uint32_t cbLogBuf;
1070	uint32_t idxFirst;
1071	uint32_t idxNext;
1072
1073	struct { void pvVar; size_t cbHost, cbGuest; const char pszSymbol; } aSymbols[] =
1074	{
1075	{ &GCPtrLogBuf, sizeof(GCPtrLogBuf), pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t), "log_buf" },
1076	{ &cbLogBuf, sizeof(cbLogBuf), sizeof(cbLogBuf), "log_buf_len" },
1077	{ &idxFirst, sizeof(idxFirst), sizeof(idxFirst), "log_first_idx" },
1078	{ &idxNext, sizeof(idxNext), sizeof(idxNext), "log_next_idx" },
1079	};
1080	for (uint32_t i = 0; i < RT_ELEMENTS(aSymbols); i++)
1081	{
1082	RTDBGSYMBOL SymInfo;
1083	rc = RTDbgModSymbolByName(hMod, aSymbols[i].pszSymbol, &SymInfo);
1084	if (RT_SUCCESS(rc))
1085	{
1086	RT_BZERO(aSymbols[i].pvVar, aSymbols[i].cbHost);
1087	Assert(aSymbols[i].cbHost >= aSymbols[i].cbGuest);
1088	DBGFADDRESS Addr;
1089	rc = DBGFR3MemRead(pUVM, 0 /idCpu/,
1090	DBGFR3AddrFromFlat(pUVM, &Addr, (RTGCPTR)SymInfo.Value + pThis->AddrKernelBase.FlatPtr),
1091	aSymbols[i].pvVar, aSymbols[i].cbGuest);
1092	if (RT_SUCCESS(rc))
1093	continue;
1094	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Reading '%s' at %RGv: %Rrc\n", aSymbols[i].pszSymbol, Addr.FlatPtr, rc));
1095	}
1096	else
1097	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Error looking up '%s': %Rrc\n", aSymbols[i].pszSymbol, rc));
1098	rc = VERR_NOT_FOUND;
1099	break;
1100	}
1101
1102	/*
1103	* Some kernels don't expose the variables in kallsyms so we have to try disassemble
1104	* some public helpers to get at the addresses.
1105	*
1106	* @todo: Maybe cache those values so we don't have to do the heavy work every time?
1107	*/
1108	if (rc == VERR_NOT_FOUND)
1109	{
1110	idxFirst = 0;
1111	idxNext = 0;
1112	rc = dbgDiggerLinuxQueryLogBufferPtrs(pThis, pUVM, hMod, &GCPtrLogBuf, &cbLogBuf);
1113	if (RT_FAILURE(rc))
1114	{
1115	/*
1116	* Last resort, scan for a known value which should appear only once in the kernel log buffer
1117	* and try to deduce the boundaries from there.
1118	*/
1119	rc = dbgDiggerLinuxKrnlLogBufFindByNeedle(pThis, pUVM, &GCPtrLogBuf, &cbLogBuf);
1120	return rc;
1121	}
1122	}
1123
1124	return dbgDiggerLinuxKrnLogBufferProcess(pThis, pUVM, GCPtrLogBuf, cbLogBuf, idxFirst, idxNext,
1125	fFlags, cMessages, pszBuf, cbBuf, pcbActual);
1126	}
1127
1128	/**
1129	* @interface_method_impl{DBGFOSIDMESG,pfnQueryKernelLog}
1130	*/
1131	static DECLCALLBACK(int) dbgDiggerLinuxIDmsg_QueryKernelLog(PDBGFOSIDMESG pThis, PUVM pUVM, uint32_t fFlags, uint32_t cMessages,
1132	char pszBuf, size_t cbBuf, size_t pcbActual)
1133	{
1134	PDBGDIGGERLINUX pData = RT_FROM_MEMBER(pThis, DBGDIGGERLINUX, IDmesg);
1135
1136	if (cMessages < 1)
1137	return VERR_INVALID_PARAMETER;
1138
1139	/*
1140	* Resolve the symbols we need and read their values.
1141	*/
1142	RTDBGAS hAs = DBGFR3AsResolveAndRetain(pUVM, DBGF_AS_KERNEL);
1143	RTDBGMOD hMod;
1144	int rc = RTDbgAsModuleByName(hAs, "vmlinux", 0, &hMod);
1145	RTDbgAsRelease(hAs);
1146
1147	size_t cbActual = 0;
1148	if (RT_SUCCESS(rc))
1149	{
1150	/*
1151	* Check whether the kernel log buffer is a simple char buffer or the newer
1152	* record based implementation.
1153	* The record based implementation was presumably introduced with kernel 3.4,
1154	* see: http://thread.gmane.org/gmane.linux.kernel/1284184
1155	*/
1156	if (dbgDiggerLinuxLogBufferIsAsciiBuffer(pData, pUVM))
1157	rc = dbgDiggerLinuxLogBufferQueryAscii(pData, pUVM, hMod, fFlags, cMessages, pszBuf, cbBuf, &cbActual);
1158	else
1159	rc = dbgDiggerLinuxLogBufferQueryRecords(pData, pUVM, hMod, fFlags, cMessages, pszBuf, cbBuf, &cbActual);
1160
1161	/* Release the module in any case. */
1162	RTDbgModRelease(hMod);
1163	}
1164	else
1165	{
1166	/*
1167	* For the record based kernel versions we have a last resort heuristic which doesn't
1168	* require any symbols, try that here.
1169	*/
1170	if (!dbgDiggerLinuxLogBufferIsAsciiBuffer(pData, pUVM))
1171	{
1172	RTGCPTR GCPtrLogBuf = 0;
1173	uint32_t cbLogBuf = 0;
1174
1175	rc = dbgDiggerLinuxKrnlLogBufFindByNeedle(pData, pUVM, &GCPtrLogBuf, &cbLogBuf);
1176	if (RT_SUCCESS(rc))
1177	rc = dbgDiggerLinuxKrnLogBufferProcess(pData, pUVM, GCPtrLogBuf, cbLogBuf, 0 /idxFirst/, 0 /idxNext/,
1178	fFlags, cMessages, pszBuf, cbBuf, &cbActual);
1179	}
1180	else
1181	rc = VERR_NOT_FOUND;
1182	}
1183
1184	if (RT_FAILURE(rc) && rc != VERR_BUFFER_OVERFLOW)
1185	return rc;
1186
1187	if (pcbActual)
1188	*pcbActual = cbActual;
1189
1190	/*
1191	* All VBox strings are UTF-8 and bad things may in theory happen if we
1192	* pass bad UTF-8 to code which assumes it's all valid. So, we enforce
1193	* UTF-8 upon the guest kernel messages here even if they (probably) have
1194	* no defined code set in reality.
1195	*/
1196	if ( RT_SUCCESS(rc)
1197	&& cbActual <= cbBuf)
1198	{
1199	pszBuf[cbActual - 1] = '\0';
1200	RTStrPurgeEncoding(pszBuf);
1201	return VINF_SUCCESS;
1202	}
1203
1204	if (cbBuf)
1205	{
1206	pszBuf[cbBuf - 1] = '\0';
1207	RTStrPurgeEncoding(pszBuf);
1208	}
1209	return VERR_BUFFER_OVERFLOW;
1210	}
1211
1212
1213	/**
1214	* Worker destroying the config database.
1215	*/
1216	static DECLCALLBACK(int) dbgDiggerLinuxCfgDbDestroyWorker(PRTSTRSPACECORE pStr, void *pvUser)
1217	{
1218	PDBGDIGGERLINUXCFGITEM pCfgItem = (PDBGDIGGERLINUXCFGITEM)pStr;
1219	RTStrFree((char *)pCfgItem->Core.pszString);
1220	RTMemFree(pCfgItem);
1221	NOREF(pvUser);
1222	return 0;
1223	}
1224
1225
1226	/**
1227	* Destroy the config database.
1228	*
1229	* @returns nothing.
1230	* @param pThis The Linux digger data.
1231	*/
1232	static void dbgDiggerLinuxCfgDbDestroy(PDBGDIGGERLINUX pThis)
1233	{
1234	RTStrSpaceDestroy(&pThis->hCfgDb, dbgDiggerLinuxCfgDbDestroyWorker, NULL);
1235	}
1236
1237
1238	/**
1239	* @copydoc DBGFOSREG::pfnStackUnwindAssist
1240	*/
1241	static DECLCALLBACK(int) dbgDiggerLinuxStackUnwindAssist(PUVM pUVM, void *pvData, VMCPUID idCpu, PDBGFSTACKFRAME pFrame,
1242	PRTDBGUNWINDSTATE pState, PCCPUMCTX pInitialCtx, RTDBGAS hAs,
1243	uint64_t *puScratch)
1244	{
1245	RT_NOREF(pUVM, pvData, idCpu, pFrame, pState, pInitialCtx, hAs, puScratch);
1246	return VINF_SUCCESS;
1247	}
1248
1249
1250	/**
1251	* @copydoc DBGFOSREG::pfnQueryInterface
1252	*/
1253	static DECLCALLBACK(void ) dbgDiggerLinuxQueryInterface(PUVM pUVM, void pvData, DBGFOSINTERFACE enmIf)
1254	{
1255	RT_NOREF1(pUVM);
1256	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
1257	switch (enmIf)
1258	{
1259	case DBGFOSINTERFACE_DMESG:
1260	return &pThis->IDmesg;
1261
1262	default:
1263	return NULL;
1264	}
1265	}
1266
1267
1268	/**
1269	* @copydoc DBGFOSREG::pfnQueryVersion
1270	*/
1271	static DECLCALLBACK(int) dbgDiggerLinuxQueryVersion(PUVM pUVM, void pvData, char pszVersion, size_t cchVersion)
1272	{
1273	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
1274	Assert(pThis->fValid);
1275
1276	/*
1277	* It's all in the linux banner.
1278	*/
1279	int rc = DBGFR3MemReadString(pUVM, 0, &pThis->AddrLinuxBanner, pszVersion, cchVersion);
1280	if (RT_SUCCESS(rc))
1281	{
1282	char *pszEnd = RTStrEnd(pszVersion, cchVersion);
1283	AssertReturn(pszEnd, VERR_BUFFER_OVERFLOW);
1284	while ( pszEnd > pszVersion
1285	&& RT_C_IS_SPACE(pszEnd[-1]))
1286	pszEnd--;
1287	*pszEnd = '\0';
1288	}
1289	else
1290	RTStrPrintf(pszVersion, cchVersion, "DBGFR3MemRead -> %Rrc", rc);
1291
1292	return rc;
1293	}
1294
1295
1296	/**
1297	* @copydoc DBGFOSREG::pfnTerm
1298	*/
1299	static DECLCALLBACK(void) dbgDiggerLinuxTerm(PUVM pUVM, void *pvData)
1300	{
1301	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
1302	Assert(pThis->fValid);
1303
1304	/*
1305	* Destroy configuration database.
1306	*/
1307	dbgDiggerLinuxCfgDbDestroy(pThis);
1308
1309	/*
1310	* Unlink and release our modules.
1311	*/
1312	RTDBGAS hDbgAs = DBGFR3AsResolveAndRetain(pUVM, DBGF_AS_KERNEL);
1313	if (hDbgAs != NIL_RTDBGAS)
1314	{
1315	uint32_t iMod = RTDbgAsModuleCount(hDbgAs);
1316	while (iMod-- > 0)
1317	{
1318	RTDBGMOD hMod = RTDbgAsModuleByIndex(hDbgAs, iMod);
1319	if (hMod != NIL_RTDBGMOD)
1320	{
1321	if (RTDbgModGetTag(hMod) == DIG_LNX_MOD_TAG)
1322	{
1323	int rc = RTDbgAsModuleUnlink(hDbgAs, hMod);
1324	AssertRC(rc);
1325	}
1326	RTDbgModRelease(hMod);
1327	}
1328	}
1329	RTDbgAsRelease(hDbgAs);
1330	}
1331
1332	pThis->fValid = false;
1333	}
1334
1335
1336	/**
1337	* @copydoc DBGFOSREG::pfnRefresh
1338	*/
1339	static DECLCALLBACK(int) dbgDiggerLinuxRefresh(PUVM pUVM, void *pvData)
1340	{
1341	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
1342	NOREF(pThis);
1343	Assert(pThis->fValid);
1344
1345	/*
1346	* For now we'll flush and reload everything.
1347	*/
1348	dbgDiggerLinuxTerm(pUVM, pvData);
1349	return dbgDiggerLinuxInit(pUVM, pvData);
1350	}
1351
1352
1353	/**
1354	* Worker for dbgDiggerLinuxFindStartOfNamesAndSymbolCount that update the
1355	* digger data.
1356	*
1357	* @returns VINF_SUCCESS.
1358	* @param pThis The Linux digger data to update.
1359	* @param pAddrKernelNames The kallsyms_names address.
1360	* @param cKernelSymbols The number of kernel symbol.
1361	* @param cbAddress The guest address size.
1362	*/
1363	static int dbgDiggerLinuxFoundStartOfNames(PDBGDIGGERLINUX pThis, PCDBGFADDRESS pAddrKernelNames,
1364	uint32_t cKernelSymbols, uint32_t cbAddress)
1365	{
1366	pThis->cKernelSymbols = cKernelSymbols;
1367	pThis->AddrKernelNames = *pAddrKernelNames;
1368	pThis->AddrKernelAddresses = *pAddrKernelNames;
1369	uint32_t cbSymbolsSkip = (pThis->fRelKrnlAddr ? 2 : 1) * cbAddress; /* Relative addressing introduces kallsyms_relative_base. */
1370	uint32_t cbOffsets = pThis->fRelKrnlAddr ? sizeof(int32_t) : cbAddress; /* Offsets are always 32bits wide for relative addressing. */
1371	uint32_t cbAlign = 0;
1372
1373	/*
1374	* If the number of symbols is odd there is padding to align the following guest pointer
1375	* sized data properly on 64bit systems with relative addressing.
1376	*/
1377	if ( pThis->fRelKrnlAddr
1378	&& pThis->f64Bit
1379	&& (pThis->cKernelSymbols & 1))
1380	cbAlign = sizeof(int32_t);
1381	DBGFR3AddrSub(&pThis->AddrKernelAddresses, cKernelSymbols * cbOffsets + cbSymbolsSkip + cbAlign);
1382
1383	Log(("dbgDiggerLinuxFoundStartOfNames: AddrKernelAddresses=%RGv\n"
1384	"dbgDiggerLinuxFoundStartOfNames: cKernelSymbols=%#x (at %RGv)\n"
1385	"dbgDiggerLinuxFoundStartOfNames: AddrKernelName=%RGv\n",
1386	pThis->AddrKernelAddresses.FlatPtr,
1387	pThis->cKernelSymbols, pThis->AddrKernelNames.FlatPtr - cbAddress,
1388	pThis->AddrKernelNames.FlatPtr));
1389	return VINF_SUCCESS;
1390	}
1391
1392
1393	/**
1394	* Tries to find the address of the kallsyms_names, kallsyms_num_syms and
1395	* kallsyms_addresses symbols.
1396	*
1397	* The kallsyms_num_syms is read and stored in pThis->cKernelSymbols, while the
1398	* addresses of the other two are stored as pThis->AddrKernelNames and
1399	* pThis->AddrKernelAddresses.
1400	*
1401	* @returns VBox status code, success indicating that all three variables have
1402	* been found and taken down.
1403	* @param pUVM The user mode VM handle.
1404	* @param pThis The Linux digger data.
1405	* @param pHitAddr An address we think is inside kallsyms_names.
1406	*/
1407	static int dbgDiggerLinuxFindStartOfNamesAndSymbolCount(PUVM pUVM, PDBGDIGGERLINUX pThis, PCDBGFADDRESS pHitAddr)
1408	{
1409	/*
1410	* Search backwards in chunks.
1411	*/
1412	union
1413	{
1414	uint8_t ab[0x1000];
1415	uint32_t au32[0x1000 / sizeof(uint32_t)];
1416	uint64_t au64[0x1000 / sizeof(uint64_t)];
1417	} uBuf;
1418	uint32_t cbLeft = LNX_MAX_KALLSYMS_NAMES_SIZE;
1419	uint32_t cbBuf = pHitAddr->FlatPtr & (sizeof(uBuf) - 1);
1420	DBGFADDRESS CurAddr = *pHitAddr;
1421	DBGFR3AddrSub(&CurAddr, cbBuf);
1422	cbBuf += sizeof(uint64_t) - 1; /* In case our kobj hit is in the first 4/8 bytes. */
1423	for (;;)
1424	{
1425	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &CurAddr, &uBuf, sizeof(uBuf));
1426	if (RT_FAILURE(rc))
1427	return rc;
1428
1429	/*
1430	* Since Linux 4.6 there are two different methods to store the kallsyms addresses
1431	* in the image.
1432	*
1433	* The first and longer existing method is to store the absolute addresses in an
1434	* array starting at kallsyms_addresses followed by a field which stores the number
1435	* of kernel symbols called kallsyms_num_syms.
1436	* The newer method is to use offsets stored in kallsyms_offsets and have a base pointer
1437	* to relate the offsets to called kallsyms_relative_base. One entry in kallsyms_offsets is
1438	* always 32bit wide regardless of the guest pointer size (this halves the table on 64bit
1439	* systems) but means more work for us for the 64bit case.
1440	*
1441	* When absolute addresses are used the following assumptions hold:
1442	*
1443	* We assume that the three symbols are aligned on guest pointer boundary.
1444	*
1445	* The boundary between the two tables should be noticable as the number
1446	* is unlikely to be more than 16 millions, there will be at least one zero
1447	* byte where it is, 64-bit will have 5 zero bytes. Zero bytes aren't all
1448	* that common in the kallsyms_names table.
1449	*
1450	* Also the kallsyms_names table starts with a length byte, which means
1451	* we're likely to see a byte in the range 1..31.
1452	*
1453	* The kallsyms_addresses are mostly sorted (except for the start where the
1454	* absolute symbols are), so we'll spot a bunch of kernel addresses
1455	* immediately preceeding the kallsyms_num_syms field.
1456	*
1457	* Lazy bird: If kallsyms_num_syms is on a buffer boundrary, we skip
1458	* the check for kernel addresses preceeding it.
1459	*
1460	* For relative offsets most of the assumptions from above are true too
1461	* except that we have to distinguish between the relative base address and the offsets.
1462	* Every observed kernel has a valid kernel address fo the relative base and kallsyms_relative_base
1463	* always comes before kallsyms_num_syms and is aligned on a guest pointer boundary.
1464	* Offsets are stored before kallsyms_relative_base and don't contain valid kernel addresses.
1465	*
1466	* To distinguish between absolute and relative offsetting we check the data before a candidate
1467	* for kallsyms_num_syms. If all entries before the kallsyms_num_syms candidate are valid kernel
1468	* addresses absolute addresses are assumed. If this is not the case but the first entry before
1469	* kallsyms_num_syms is a valid kernel address we check whether the data before and the possible
1470	* relative base form a valid kernel address and assume relative offsets.
1471	*/
1472	if (pThis->f64Bit)
1473	{
1474	uint32_t i = cbBuf / sizeof(uint64_t) - 1;
1475	while (i-- > 0)
1476	if ( uBuf.au64[i] <= LNX_MAX_KALLSYMS_SYMBOLS
1477	&& uBuf.au64[i] >= LNX_MIN_KALLSYMS_SYMBOLS)
1478	{
1479	uint8_t pb = (uint8_t )&uBuf.au64[i + 1];
1480	if ( pb[0] <= LNX_MAX_KALLSYMS_ENC_LENGTH
1481	&& pb[0] >= LNX_MIN_KALLSYMS_ENC_LENGTH)
1482	{
1483	/*
1484	* Check whether we have a valid kernel address and try to distinguish
1485	* whether the kernel uses relative offsetting or absolute addresses.
1486	*/
1487	if ( (i >= 1 && LNX64_VALID_ADDRESS(uBuf.au64[i - 1]))
1488	&& (i >= 2 && !LNX64_VALID_ADDRESS(uBuf.au64[i - 2]))
1489	&& (i >= 3 && !LNX64_VALID_ADDRESS(uBuf.au64[i - 3])))
1490	{
1491	RTGCUINTPTR uKrnlRelBase = uBuf.au64[i - 1];
1492	DBGFADDRESS RelAddr = CurAddr;
1493	int32_t aiRelOff[3];
1494	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, DBGFR3AddrAdd(&RelAddr, (i - 1) * sizeof(uint64_t) - sizeof(aiRelOff)),
1495	&aiRelOff[0], sizeof(aiRelOff));
1496	if ( RT_SUCCESS(rc)
1497	&& LNX64_VALID_ADDRESS(uKrnlRelBase + aiRelOff[0])
1498	&& LNX64_VALID_ADDRESS(uKrnlRelBase + aiRelOff[1])
1499	&& LNX64_VALID_ADDRESS(uKrnlRelBase + aiRelOff[2]))
1500	{
1501	Log(("dbgDiggerLinuxFindStartOfNamesAndSymbolCount: relative base %RGv (at %RGv)\n",
1502	uKrnlRelBase, CurAddr.FlatPtr + (i - 1) * sizeof(uint64_t)));
1503	pThis->fRelKrnlAddr = true;
1504	pThis->uKernelRelativeBase = uKrnlRelBase;
1505	return dbgDiggerLinuxFoundStartOfNames(pThis,
1506	DBGFR3AddrAdd(&CurAddr, (i + 1) * sizeof(uint64_t)),
1507	(uint32_t)uBuf.au64[i], sizeof(uint64_t));
1508	}
1509	}
1510
1511	if ( (i <= 0 \|\| LNX64_VALID_ADDRESS(uBuf.au64[i - 1]))
1512	&& (i <= 1 \|\| LNX64_VALID_ADDRESS(uBuf.au64[i - 2]))
1513	&& (i <= 2 \|\| LNX64_VALID_ADDRESS(uBuf.au64[i - 3])))
1514	return dbgDiggerLinuxFoundStartOfNames(pThis,
1515	DBGFR3AddrAdd(&CurAddr, (i + 1) * sizeof(uint64_t)),
1516	(uint32_t)uBuf.au64[i], sizeof(uint64_t));
1517	}
1518	}
1519	}
1520	else
1521	{
1522	uint32_t i = cbBuf / sizeof(uint32_t) - 1;
1523	while (i-- > 0)
1524	if ( uBuf.au32[i] <= LNX_MAX_KALLSYMS_SYMBOLS
1525	&& uBuf.au32[i] >= LNX_MIN_KALLSYMS_SYMBOLS)
1526	{
1527	uint8_t pb = (uint8_t )&uBuf.au32[i + 1];
1528	if ( pb[0] <= LNX_MAX_KALLSYMS_ENC_LENGTH
1529	&& pb[0] >= LNX_MIN_KALLSYMS_ENC_LENGTH)
1530	{
1531	/* Check for relative base addressing. */
1532	if (i >= 1 && LNX32_VALID_ADDRESS(uBuf.au32[i - 1]))
1533	{
1534	RTGCUINTPTR uKrnlRelBase = uBuf.au32[i - 1];
1535	if ( (i <= 1 \|\| LNX32_VALID_ADDRESS(uKrnlRelBase + uBuf.au32[i - 2]))
1536	&& (i <= 2 \|\| LNX32_VALID_ADDRESS(uKrnlRelBase + uBuf.au32[i - 3])))
1537	{
1538	Log(("dbgDiggerLinuxFindStartOfNamesAndSymbolCount: relative base %RGv (at %RGv)\n",
1539	uKrnlRelBase, CurAddr.FlatPtr + (i - 1) * sizeof(uint32_t)));
1540	pThis->fRelKrnlAddr = true;
1541	pThis->uKernelRelativeBase = uKrnlRelBase;
1542	return dbgDiggerLinuxFoundStartOfNames(pThis,
1543	DBGFR3AddrAdd(&CurAddr, (i + 1) * sizeof(uint32_t)),
1544	uBuf.au32[i], sizeof(uint32_t));
1545	}
1546	}
1547
1548	if ( (i <= 0 \|\| LNX32_VALID_ADDRESS(uBuf.au32[i - 1]))
1549	&& (i <= 1 \|\| LNX32_VALID_ADDRESS(uBuf.au32[i - 2]))
1550	&& (i <= 2 \|\| LNX32_VALID_ADDRESS(uBuf.au32[i - 3])))
1551	return dbgDiggerLinuxFoundStartOfNames(pThis,
1552	DBGFR3AddrAdd(&CurAddr, (i + 1) * sizeof(uint32_t)),
1553	uBuf.au32[i], sizeof(uint32_t));
1554	}
1555	}
1556	}
1557
1558	/*
1559	* Advance
1560	*/
1561	if (RT_UNLIKELY(cbLeft <= sizeof(uBuf)))
1562	{
1563	Log(("dbgDiggerLinuxFindStartOfNamesAndSymbolCount: failed (pHitAddr=%RGv)\n", pHitAddr->FlatPtr));
1564	return VERR_NOT_FOUND;
1565	}
1566	cbLeft -= sizeof(uBuf);
1567	DBGFR3AddrSub(&CurAddr, sizeof(uBuf));
1568	cbBuf = sizeof(uBuf);
1569	}
1570	}
1571
1572
1573	/**
1574	* Worker for dbgDiggerLinuxFindEndNames that records the findings.
1575	*
1576	* @returns VINF_SUCCESS
1577	* @param pThis The linux digger data to update.
1578	* @param pAddrMarkers The address of the marker (kallsyms_markers).
1579	* @param cbMarkerEntry The size of a marker entry (32-bit or 64-bit).
1580	*/
1581	static int dbgDiggerLinuxFoundMarkers(PDBGDIGGERLINUX pThis, PCDBGFADDRESS pAddrMarkers, uint32_t cbMarkerEntry)
1582	{
1583	pThis->cbKernelNames = pAddrMarkers->FlatPtr - pThis->AddrKernelNames.FlatPtr;
1584	pThis->AddrKernelNameMarkers = *pAddrMarkers;
1585	pThis->cKernelNameMarkers = RT_ALIGN_32(pThis->cKernelSymbols, 256) / 256;
1586	pThis->AddrKernelTokenTable = *pAddrMarkers;
1587	DBGFR3AddrAdd(&pThis->AddrKernelTokenTable, pThis->cKernelNameMarkers * cbMarkerEntry);
1588
1589	Log(("dbgDiggerLinuxFoundMarkers: AddrKernelNames=%RGv cbKernelNames=%#x\n"
1590	"dbgDiggerLinuxFoundMarkers: AddrKernelNameMarkers=%RGv cKernelNameMarkers=%#x\n"
1591	"dbgDiggerLinuxFoundMarkers: AddrKernelTokenTable=%RGv\n",
1592	pThis->AddrKernelNames.FlatPtr, pThis->cbKernelNames,
1593	pThis->AddrKernelNameMarkers.FlatPtr, pThis->cKernelNameMarkers,
1594	pThis->AddrKernelTokenTable.FlatPtr));
1595	return VINF_SUCCESS;
1596	}
1597
1598
1599	/**
1600	* Tries to find the end of kallsyms_names and thereby the start of
1601	* kallsyms_markers and kallsyms_token_table.
1602	*
1603	* The kallsyms_names size is stored in pThis->cbKernelNames, the addresses of
1604	* the two other symbols in pThis->AddrKernelNameMarkers and
1605	* pThis->AddrKernelTokenTable. The number of marker entries is stored in
1606	* pThis->cKernelNameMarkers.
1607	*
1608	* @returns VBox status code, success indicating that all three variables have
1609	* been found and taken down.
1610	* @param pUVM The user mode VM handle.
1611	* @param pThis The Linux digger data.
1612	* @param pHitAddr An address we think is inside kallsyms_names.
1613	*/
1614	static int dbgDiggerLinuxFindEndOfNamesAndMore(PUVM pUVM, PDBGDIGGERLINUX pThis, PCDBGFADDRESS pHitAddr)
1615	{
1616	/*
1617	* Search forward in chunks.
1618	*/
1619	union
1620	{
1621	uint8_t ab[0x1000];
1622	uint32_t au32[0x1000 / sizeof(uint32_t)];
1623	uint64_t au64[0x1000 / sizeof(uint64_t)];
1624	} uBuf;
1625	bool fPendingZeroHit = false;
1626	uint32_t cbLeft = LNX_MAX_KALLSYMS_NAMES_SIZE + sizeof(uBuf);
1627	uint32_t offBuf = pHitAddr->FlatPtr & (sizeof(uBuf) - 1);
1628	DBGFADDRESS CurAddr = *pHitAddr;
1629	DBGFR3AddrSub(&CurAddr, offBuf);
1630	for (;;)
1631	{
1632	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &CurAddr, &uBuf, sizeof(uBuf));
1633	if (RT_FAILURE(rc))
1634	return rc;
1635
1636	/*
1637	* The kallsyms_names table is followed by kallsyms_markers we assume,
1638	* using sizeof(unsigned long) alignment like the preceeding symbols.
1639	*
1640	* The kallsyms_markers table has entried sizeof(unsigned long) and
1641	* contains offsets into kallsyms_names. The kallsyms_markers used to
1642	* index kallsyms_names and reduce seek time when looking up the name
1643	* of an address/symbol. Each entry in kallsyms_markers covers 256
1644	* symbol names.
1645	*
1646	* Because of this, the first entry is always zero and all the entries
1647	* are ascending. It also follows that the size of the table can be
1648	* calculated from kallsyms_num_syms.
1649	*
1650	* Note! We could also have walked kallsyms_names by skipping
1651	* kallsyms_num_syms names, but this is faster and we will
1652	* validate the encoded names later.
1653	*/
1654	if (pThis->f64Bit)
1655	{
1656	if ( RT_UNLIKELY(fPendingZeroHit)
1657	&& uBuf.au64[0] >= (LNX_MIN_KALLSYMS_ENC_LENGTH + 1) * 256
1658	&& uBuf.au64[0] <= (LNX_MAX_KALLSYMS_ENC_LENGTH + 1) * 256)
1659	return dbgDiggerLinuxFoundMarkers(pThis, DBGFR3AddrSub(&CurAddr, sizeof(uint64_t)), sizeof(uint64_t));
1660
1661	uint32_t const cEntries = sizeof(uBuf) / sizeof(uint64_t);
1662	for (uint32_t i = offBuf / sizeof(uint64_t); i < cEntries; i++)
1663	if (uBuf.au64[i] == 0)
1664	{
1665	if (RT_UNLIKELY(i + 1 >= cEntries))
1666	{
1667	fPendingZeroHit = true;
1668	break;
1669	}
1670	if ( uBuf.au64[i + 1] >= (LNX_MIN_KALLSYMS_ENC_LENGTH + 1) * 256
1671	&& uBuf.au64[i + 1] <= (LNX_MAX_KALLSYMS_ENC_LENGTH + 1) * 256)
1672	return dbgDiggerLinuxFoundMarkers(pThis, DBGFR3AddrAdd(&CurAddr, i * sizeof(uint64_t)), sizeof(uint64_t));
1673	}
1674	}
1675	else
1676	{
1677	if ( RT_UNLIKELY(fPendingZeroHit)
1678	&& uBuf.au32[0] >= (LNX_MIN_KALLSYMS_ENC_LENGTH + 1) * 256
1679	&& uBuf.au32[0] <= (LNX_MAX_KALLSYMS_ENC_LENGTH + 1) * 256)
1680	return dbgDiggerLinuxFoundMarkers(pThis, DBGFR3AddrSub(&CurAddr, sizeof(uint32_t)), sizeof(uint32_t));
1681
1682	uint32_t const cEntries = sizeof(uBuf) / sizeof(uint32_t);
1683	for (uint32_t i = offBuf / sizeof(uint32_t); i < cEntries; i++)
1684	if (uBuf.au32[i] == 0)
1685	{
1686	if (RT_UNLIKELY(i + 1 >= cEntries))
1687	{
1688	fPendingZeroHit = true;
1689	break;
1690	}
1691	if ( uBuf.au32[i + 1] >= (LNX_MIN_KALLSYMS_ENC_LENGTH + 1) * 256
1692	&& uBuf.au32[i + 1] <= (LNX_MAX_KALLSYMS_ENC_LENGTH + 1) * 256)
1693	return dbgDiggerLinuxFoundMarkers(pThis, DBGFR3AddrAdd(&CurAddr, i * sizeof(uint32_t)), sizeof(uint32_t));
1694	}
1695	}
1696
1697	/*
1698	* Advance
1699	*/
1700	if (RT_UNLIKELY(cbLeft <= sizeof(uBuf)))
1701	{
1702	Log(("dbgDiggerLinuxFindEndOfNamesAndMore: failed (pHitAddr=%RGv)\n", pHitAddr->FlatPtr));
1703	return VERR_NOT_FOUND;
1704	}
1705	cbLeft -= sizeof(uBuf);
1706	DBGFR3AddrAdd(&CurAddr, sizeof(uBuf));
1707	offBuf = 0;
1708	}
1709	}
1710
1711
1712	/**
1713	* Locates the kallsyms_token_index table.
1714	*
1715	* Storing the address in pThis->AddrKernelTokenIndex and the size of the token
1716	* table in pThis->cbKernelTokenTable.
1717	*
1718	* @returns VBox status code.
1719	* @param pUVM The user mode VM handle.
1720	* @param pThis The Linux digger data.
1721	*/
1722	static int dbgDiggerLinuxFindTokenIndex(PUVM pUVM, PDBGDIGGERLINUX pThis)
1723	{
1724	/*
1725	* The kallsyms_token_table is very much like a string table. Due to the
1726	* nature of the compression algorithm it is reasonably short (one example
1727	* here is 853 bytes), so we'll not be reading it in chunks but in full.
1728	* To be on the safe side, we read 8KB, ASSUMING we won't run into unmapped
1729	* memory or any other nasty stuff...
1730	*/
1731	union
1732	{
1733	uint8_t ab[0x2000];
1734	uint16_t au16[0x2000 / sizeof(uint16_t)];
1735	} uBuf;
1736	DBGFADDRESS CurAddr = pThis->AddrKernelTokenTable;
1737	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &CurAddr, &uBuf, sizeof(uBuf));
1738	if (RT_FAILURE(rc))
1739	return rc;
1740
1741	/*
1742	* We've got two choices here, either walk the string table or look for
1743	* the next structure, kallsyms_token_index.
1744	*
1745	* The token index is a table of 256 uint16_t entries (index by bytes
1746	* from kallsyms_names) that gives offsets in kallsyms_token_table. It
1747	* starts with a zero entry and the following entries are sorted in
1748	* ascending order. The range of the entries are reasonably small since
1749	* kallsyms_token_table is small.
1750	*
1751	* The alignment seems to be sizeof(unsigned long), just like
1752	* kallsyms_token_table.
1753	*
1754	* So, we start by looking for a zero 16-bit entry.
1755	*/
1756	uint32_t cIncr = (pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t)) / sizeof(uint16_t);
1757
1758	for (uint32_t i = 0; i < sizeof(uBuf) / sizeof(uint16_t) - 16; i += cIncr)
1759	if ( uBuf.au16[i] == 0
1760	&& uBuf.au16[i + 1] > 0
1761	&& uBuf.au16[i + 1] <= LNX_MAX_KALLSYMS_TOKEN_LEN
1762	&& (uint16_t)(uBuf.au16[i + 2] - uBuf.au16[i + 1] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1763	&& (uint16_t)(uBuf.au16[i + 3] - uBuf.au16[i + 2] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1764	&& (uint16_t)(uBuf.au16[i + 4] - uBuf.au16[i + 3] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1765	&& (uint16_t)(uBuf.au16[i + 5] - uBuf.au16[i + 4] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1766	&& (uint16_t)(uBuf.au16[i + 6] - uBuf.au16[i + 5] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1767	)
1768	{
1769	pThis->AddrKernelTokenIndex = CurAddr;
1770	DBGFR3AddrAdd(&pThis->AddrKernelTokenIndex, i * sizeof(uint16_t));
1771	pThis->cbKernelTokenTable = i * sizeof(uint16_t);
1772	return VINF_SUCCESS;
1773	}
1774
1775	Log(("dbgDiggerLinuxFindTokenIndex: Failed (%RGv..%RGv)\n", CurAddr.FlatPtr, CurAddr.FlatPtr + (RTGCUINTPTR)sizeof(uBuf)));
1776	return VERR_NOT_FOUND;
1777	}
1778
1779
1780	/**
1781	* Loads the kernel symbols from the given kallsyms offset table decoding the symbol names
1782	* (worker common for dbgDiggerLinuxLoadKernelSymbolsAbsolute() and dbgDiggerLinuxLoadKernelSymbolsRelative()).
1783	*
1784	* @returns VBox status code.
1785	* @param pUVM The user mode VM handle.
1786	* @param pThis The Linux digger data.
1787	* @param uKernelStart Flat kernel start address.
1788	* @param cbKernel Size of the kernel in bytes.
1789	* @param pauSymOff Pointer to the array of symbol offsets in the kallsyms table
1790	* relative to the start of the kernel.
1791	*/
1792	static int dbgDiggerLinuxLoadKernelSymbolsWorker(PUVM pUVM, PDBGDIGGERLINUX pThis, RTGCUINTPTR uKernelStart,
1793	RTGCUINTPTR cbKernel, RTGCUINTPTR *pauSymOff)
1794	{
1795	uint8_t pbNames = (uint8_t )RTMemAllocZ(pThis->cbKernelNames);
1796	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &pThis->AddrKernelNames, pbNames, pThis->cbKernelNames);
1797	if (RT_SUCCESS(rc))
1798	{
1799	char pszzTokens = (char )RTMemAllocZ(pThis->cbKernelTokenTable);
1800	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &pThis->AddrKernelTokenTable, pszzTokens, pThis->cbKernelTokenTable);
1801	if (RT_SUCCESS(rc))
1802	{
1803	uint16_t paoffTokens = (uint16_t )RTMemAllocZ(256 * sizeof(uint16_t));
1804	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &pThis->AddrKernelTokenIndex, paoffTokens, 256 * sizeof(uint16_t));
1805	if (RT_SUCCESS(rc))
1806	{
1807	/*
1808	* Create a module for the kernel.
1809	*/
1810	RTDBGMOD hMod;
1811	rc = RTDbgModCreate(&hMod, "vmlinux", cbKernel, 0 /fFlags/);
1812	if (RT_SUCCESS(rc))
1813	{
1814	rc = RTDbgModSetTag(hMod, DIG_LNX_MOD_TAG); AssertRC(rc);
1815	rc = VINF_SUCCESS;
1816
1817	/*
1818	* Enumerate the symbols.
1819	*/
1820	uint32_t offName = 0;
1821	uint32_t cLeft = pThis->cKernelSymbols;
1822	while (cLeft-- > 0 && RT_SUCCESS(rc))
1823	{
1824	/* Decode the symbol name first. */
1825	if (RT_LIKELY(offName < pThis->cbKernelNames))
1826	{
1827	uint8_t cbName = pbNames[offName++];
1828	if (RT_LIKELY(offName + cbName <= pThis->cbKernelNames))
1829	{
1830	char szSymbol[4096];
1831	uint32_t offSymbol = 0;
1832	while (cbName-- > 0)
1833	{
1834	uint8_t bEnc = pbNames[offName++];
1835	uint16_t offToken = paoffTokens[bEnc];
1836	if (RT_LIKELY(offToken < pThis->cbKernelTokenTable))
1837	{
1838	const char *pszToken = &pszzTokens[offToken];
1839	char ch;
1840	while ((ch = *pszToken++) != '\0')
1841	if (offSymbol < sizeof(szSymbol) - 1)
1842	szSymbol[offSymbol++] = ch;
1843	}
1844	else
1845	{
1846	rc = VERR_INVALID_UTF8_ENCODING;
1847	break;
1848	}
1849	}
1850	szSymbol[offSymbol < sizeof(szSymbol) ? offSymbol : sizeof(szSymbol) - 1] = '\0';
1851
1852	/* The offset. */
1853	RTGCUINTPTR uSymOff = *pauSymOff;
1854	pauSymOff++;
1855
1856	/* Add it without the type char. */
1857	if (uSymOff <= cbKernel)
1858	{
1859	rc = RTDbgModSymbolAdd(hMod, &szSymbol[1], RTDBGSEGIDX_RVA, uSymOff,
1860	0 /cb/, 0 /fFlags/, NULL);
1861	if (RT_FAILURE(rc))
1862	{
1863	if ( rc == VERR_DBG_SYMBOL_NAME_OUT_OF_RANGE
1864	\|\| rc == VERR_DBG_INVALID_RVA
1865	\|\| rc == VERR_DBG_ADDRESS_CONFLICT
1866	\|\| rc == VERR_DBG_DUPLICATE_SYMBOL)
1867	{
1868	Log2(("dbgDiggerLinuxLoadKernelSymbols: RTDbgModSymbolAdd(,%s,) failed %Rrc (ignored)\n", szSymbol, rc));
1869	rc = VINF_SUCCESS;
1870	}
1871	else
1872	Log(("dbgDiggerLinuxLoadKernelSymbols: RTDbgModSymbolAdd(,%s,) failed %Rrc\n", szSymbol, rc));
1873	}
1874	}
1875	}
1876	else
1877	{
1878	rc = VERR_END_OF_STRING;
1879	Log(("dbgDiggerLinuxLoadKernelSymbols: offName=%#x cLeft=%#x cbName=%#x cbKernelNames=%#x\n",
1880	offName, cLeft, cbName, pThis->cbKernelNames));
1881	}
1882	}
1883	else
1884	{
1885	rc = VERR_END_OF_STRING;
1886	Log(("dbgDiggerLinuxLoadKernelSymbols: offName=%#x cLeft=%#x cbKernelNames=%#x\n",
1887	offName, cLeft, pThis->cbKernelNames));
1888	}
1889	}
1890
1891	/*
1892	* Link the module into the address space.
1893	*/
1894	if (RT_SUCCESS(rc))
1895	{
1896	RTDBGAS hAs = DBGFR3AsResolveAndRetain(pUVM, DBGF_AS_KERNEL);
1897	if (hAs != NIL_RTDBGAS)
1898	rc = RTDbgAsModuleLink(hAs, hMod, uKernelStart, RTDBGASLINK_FLAGS_REPLACE);
1899	else
1900	rc = VERR_INTERNAL_ERROR;
1901	RTDbgAsRelease(hAs);
1902	}
1903	else
1904	Log(("dbgDiggerLinuxLoadKernelSymbols: Failed: %Rrc\n", rc));
1905	RTDbgModRelease(hMod);
1906	}
1907	else
1908	Log(("dbgDiggerLinuxLoadKernelSymbols: RTDbgModCreate failed: %Rrc\n", rc));
1909	}
1910	else
1911	Log(("dbgDiggerLinuxLoadKernelSymbols: Reading token index at %RGv failed: %Rrc\n",
1912	pThis->AddrKernelTokenIndex.FlatPtr, rc));
1913	RTMemFree(paoffTokens);
1914	}
1915	else
1916	Log(("dbgDiggerLinuxLoadKernelSymbols: Reading token table at %RGv failed: %Rrc\n",
1917	pThis->AddrKernelTokenTable.FlatPtr, rc));
1918	RTMemFree(pszzTokens);
1919	}
1920	else
1921	Log(("dbgDiggerLinuxLoadKernelSymbols: Reading encoded names at %RGv failed: %Rrc\n",
1922	pThis->AddrKernelNames.FlatPtr, rc));
1923	RTMemFree(pbNames);
1924
1925	return rc;
1926	}
1927
1928	/**
1929	* Loads the kernel symbols from the kallsyms table if it contains absolute addresses
1930	*
1931	* @returns VBox status code.
1932	* @param pUVM The user mode VM handle.
1933	* @param pThis The Linux digger data.
1934	*/
1935	static int dbgDiggerLinuxLoadKernelSymbolsAbsolute(PUVM pUVM, PDBGDIGGERLINUX pThis)
1936	{
1937	/*
1938	* Allocate memory for temporary table copies, reading the tables as we go.
1939	*/
1940	uint32_t const cbGuestAddr = pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t);
1941	void pvAddresses = RTMemAllocZ(pThis->cKernelSymbols cbGuestAddr);
1942	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &pThis->AddrKernelAddresses, pvAddresses, pThis->cKernelSymbols * cbGuestAddr);
1943	if (RT_SUCCESS(rc))
1944	{
1945	/*
1946	* Figure out the kernel start and end and convert the absolute addresses to relative offsets.
1947	*/
1948	RTGCUINTPTR uKernelStart = pThis->AddrKernelAddresses.FlatPtr;
1949	RTGCUINTPTR uKernelEnd = pThis->AddrKernelTokenIndex.FlatPtr + 256 * sizeof(uint16_t);
1950	RTGCUINTPTR pauSymOff = (RTGCUINTPTR )RTMemTmpAllocZ(pThis->cKernelSymbols * sizeof(RTGCUINTPTR));
1951	uint32_t i;
1952	if (cbGuestAddr == sizeof(uint64_t))
1953	{
1954	uint64_t pauAddrs = (uint64_t )pvAddresses;
1955	for (i = 0; i < pThis->cKernelSymbols; i++)
1956	if ( pauAddrs[i] < uKernelStart
1957	&& LNX64_VALID_ADDRESS(pauAddrs[i])
1958	&& uKernelStart - pauAddrs[i] < LNX_MAX_KERNEL_SIZE)
1959	uKernelStart = pauAddrs[i];
1960
1961	for (i = pThis->cKernelSymbols - 1; i > 0; i--)
1962	if ( pauAddrs[i] > uKernelEnd
1963	&& LNX64_VALID_ADDRESS(pauAddrs[i])
1964	&& pauAddrs[i] - uKernelEnd < LNX_MAX_KERNEL_SIZE)
1965	uKernelEnd = pauAddrs[i];
1966
1967	for (i = 0; i < pThis->cKernelSymbols; i++)
1968	pauSymOff[i] = pauAddrs[i] - uKernelStart;
1969	}
1970	else
1971	{
1972	uint32_t pauAddrs = (uint32_t )pvAddresses;
1973	for (i = 0; i < pThis->cKernelSymbols; i++)
1974	if ( pauAddrs[i] < uKernelStart
1975	&& LNX32_VALID_ADDRESS(pauAddrs[i])
1976	&& uKernelStart - pauAddrs[i] < LNX_MAX_KERNEL_SIZE)
1977	uKernelStart = pauAddrs[i];
1978
1979	for (i = pThis->cKernelSymbols - 1; i > 0; i--)
1980	if ( pauAddrs[i] > uKernelEnd
1981	&& LNX32_VALID_ADDRESS(pauAddrs[i])
1982	&& pauAddrs[i] - uKernelEnd < LNX_MAX_KERNEL_SIZE)
1983	uKernelEnd = pauAddrs[i];
1984
1985	for (i = 0; i < pThis->cKernelSymbols; i++)
1986	pauSymOff[i] = pauAddrs[i] - uKernelStart;
1987	}
1988
1989	RTGCUINTPTR cbKernel = uKernelEnd - uKernelStart;
1990	pThis->cbKernel = (uint32_t)cbKernel;
1991	DBGFR3AddrFromFlat(pUVM, &pThis->AddrKernelBase, uKernelStart);
1992	Log(("dbgDiggerLinuxLoadKernelSymbolsAbsolute: uKernelStart=%RGv cbKernel=%#x\n", uKernelStart, cbKernel));
1993
1994	rc = dbgDiggerLinuxLoadKernelSymbolsWorker(pUVM, pThis, uKernelStart, cbKernel, pauSymOff);
1995	if (RT_FAILURE(rc))
1996	Log(("dbgDiggerLinuxLoadKernelSymbolsAbsolute: Loading symbols from given offset table failed: %Rrc\n", rc));
1997	RTMemTmpFree(pauSymOff);
1998	}
1999	else
2000	Log(("dbgDiggerLinuxLoadKernelSymbolsAbsolute: Reading symbol addresses at %RGv failed: %Rrc\n",
2001	pThis->AddrKernelAddresses.FlatPtr, rc));
2002	RTMemFree(pvAddresses);
2003
2004	return rc;
2005	}
2006
2007
2008	/**
2009	* Loads the kernel symbols from the kallsyms table if it contains absolute addresses
2010	*
2011	* @returns VBox status code.
2012	* @param pUVM The user mode VM handle.
2013	* @param pThis The Linux digger data.
2014	*/
2015	static int dbgDiggerLinuxLoadKernelSymbolsRelative(PUVM pUVM, PDBGDIGGERLINUX pThis)
2016	{
2017	/*
2018	* Allocate memory for temporary table copies, reading the tables as we go.
2019	*/
2020	int32_t pai32Offsets = (int32_t )RTMemAllocZ(pThis->cKernelSymbols * sizeof(int32_t));
2021	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &pThis->AddrKernelAddresses, pai32Offsets, pThis->cKernelSymbols * sizeof(int32_t));
2022	if (RT_SUCCESS(rc))
2023	{
2024	/*
2025	* Figure out the kernel start and end and convert the absolute addresses to relative offsets.
2026	*/
2027	RTGCUINTPTR uKernelStart = pThis->AddrKernelAddresses.FlatPtr;
2028	RTGCUINTPTR uKernelEnd = pThis->AddrKernelTokenIndex.FlatPtr + 256 * sizeof(uint16_t);
2029	RTGCUINTPTR pauSymOff = (RTGCUINTPTR )RTMemTmpAllocZ(pThis->cKernelSymbols * sizeof(RTGCUINTPTR));
2030	uint32_t i;
2031
2032	for (i = 0; i < pThis->cKernelSymbols; i++)
2033	{
2034	RTGCUINTPTR uSymAddr = dbgDiggerLinuxConvOffsetToAddr(pThis, pai32Offsets[i]);
2035
2036	if ( uSymAddr < uKernelStart
2037	&& (pThis->f64Bit ? LNX64_VALID_ADDRESS(uSymAddr) : LNX32_VALID_ADDRESS(uSymAddr))
2038	&& uKernelStart - uSymAddr < LNX_MAX_KERNEL_SIZE)
2039	uKernelStart = uSymAddr;
2040	}
2041
2042	for (i = pThis->cKernelSymbols - 1; i > 0; i--)
2043	{
2044	RTGCUINTPTR uSymAddr = dbgDiggerLinuxConvOffsetToAddr(pThis, pai32Offsets[i]);
2045
2046	if ( uSymAddr > uKernelEnd
2047	&& (pThis->f64Bit ? LNX64_VALID_ADDRESS(uSymAddr) : LNX32_VALID_ADDRESS(uSymAddr))
2048	&& uSymAddr - uKernelEnd < LNX_MAX_KERNEL_SIZE)
2049	uKernelEnd = uSymAddr;
2050
2051	/* Store the offset from the derived kernel start address. */
2052	pauSymOff[i] = uSymAddr - uKernelStart;
2053	}
2054
2055	RTGCUINTPTR cbKernel = uKernelEnd - uKernelStart;
2056	pThis->cbKernel = (uint32_t)cbKernel;
2057	DBGFR3AddrFromFlat(pUVM, &pThis->AddrKernelBase, uKernelStart);
2058	Log(("dbgDiggerLinuxLoadKernelSymbolsRelative: uKernelStart=%RGv cbKernel=%#x\n", uKernelStart, cbKernel));
2059
2060	rc = dbgDiggerLinuxLoadKernelSymbolsWorker(pUVM, pThis, uKernelStart, cbKernel, pauSymOff);
2061	if (RT_FAILURE(rc))
2062	Log(("dbgDiggerLinuxLoadKernelSymbolsRelative: Loading symbols from given offset table failed: %Rrc\n", rc));
2063	RTMemTmpFree(pauSymOff);
2064	}
2065	else
2066	Log(("dbgDiggerLinuxLoadKernelSymbolsRelative: Reading symbol addresses at %RGv failed: %Rrc\n",
2067	pThis->AddrKernelAddresses.FlatPtr, rc));
2068	RTMemFree(pai32Offsets);
2069
2070	return rc;
2071	}
2072
2073
2074	/**
2075	* Loads the kernel symbols.
2076	*
2077	* @returns VBox status code.
2078	* @param pUVM The user mode VM handle.
2079	* @param pThis The Linux digger data.
2080	*/
2081	static int dbgDiggerLinuxLoadKernelSymbols(PUVM pUVM, PDBGDIGGERLINUX pThis)
2082	{
2083	/*
2084	* First the kernel itself.
2085	*/
2086	if (pThis->fRelKrnlAddr)
2087	return dbgDiggerLinuxLoadKernelSymbolsRelative(pUVM, pThis);
2088	return dbgDiggerLinuxLoadKernelSymbolsAbsolute(pUVM, pThis);
2089	}
2090
2091
2092	/*
2093	* The module structure changed it was easier to produce different code for
2094	* each version of the structure. The C preprocessor rules!
2095	*/
2096	#define LNX_TEMPLATE_HEADER "DBGPlugInLinuxModuleCodeTmpl.cpp.h"
2097
2098	#define LNX_BIT_SUFFIX _amd64
2099	#define LNX_PTR_T uint64_t
2100	#define LNX_64BIT 1
2101	#include "DBGPlugInLinuxModuleVerTmpl.cpp.h"
2102
2103	#define LNX_BIT_SUFFIX _x86
2104	#define LNX_PTR_T uint32_t
2105	#define LNX_64BIT 0
2106	#include "DBGPlugInLinuxModuleVerTmpl.cpp.h"
2107
2108	#undef LNX_TEMPLATE_HEADER
2109
2110	static const struct
2111	{
2112	uint32_t uVersion;
2113	bool f64Bit;
2114	uint64_t (*pfnProcessModule)(PDBGDIGGERLINUX pThis, PUVM pUVM, PDBGFADDRESS pAddrModule);
2115	} g_aModVersions[] =
2116	{
2117	#define LNX_TEMPLATE_HEADER "DBGPlugInLinuxModuleTableEntryTmpl.cpp.h"
2118
2119	#define LNX_BIT_SUFFIX _amd64
2120	#define LNX_64BIT 1
2121	#include "DBGPlugInLinuxModuleVerTmpl.cpp.h"
2122
2123	#define LNX_BIT_SUFFIX _x86
2124	#define LNX_64BIT 0
2125	#include "DBGPlugInLinuxModuleVerTmpl.cpp.h"
2126
2127	#undef LNX_TEMPLATE_HEADER
2128	};
2129
2130
2131	/**
2132	* Tries to find and process the module list.
2133	*
2134	* @returns VBox status code.
2135	* @param pThis The Linux digger data.
2136	* @param pUVM The user mode VM handle.
2137	*/
2138	static int dbgDiggerLinuxLoadModules(PDBGDIGGERLINUX pThis, PUVM pUVM)
2139	{
2140	/*
2141	* Locate the list head.
2142	*/
2143	RTDBGAS hAs = DBGFR3AsResolveAndRetain(pUVM, DBGF_AS_KERNEL);
2144	RTDBGSYMBOL SymInfo;
2145	int rc = RTDbgAsSymbolByName(hAs, "vmlinux!modules", &SymInfo, NULL);
2146	RTDbgAsRelease(hAs);
2147	if (RT_FAILURE(rc))
2148	return VERR_NOT_FOUND;
2149
2150	if (RT_FAILURE(rc))
2151	{
2152	LogRel(("dbgDiggerLinuxLoadModules: Failed to locate the module list (%Rrc).\n", rc));
2153	return VERR_NOT_FOUND;
2154	}
2155
2156	/*
2157	* Read the list anchor.
2158	*/
2159	union
2160	{
2161	uint32_t volatile u32Pair[2];
2162	uint64_t u64Pair[2];
2163	} uListAnchor;
2164	DBGFADDRESS Addr;
2165	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, DBGFR3AddrFromFlat(pUVM, &Addr, SymInfo.Value),
2166	&uListAnchor, pThis->f64Bit ? sizeof(uListAnchor.u64Pair) : sizeof(uListAnchor.u32Pair));
2167	if (RT_FAILURE(rc))
2168	{
2169	LogRel(("dbgDiggerLinuxLoadModules: Error reading list anchor at %RX64: %Rrc\n", SymInfo.Value, rc));
2170	return VERR_NOT_FOUND;
2171	}
2172	if (!pThis->f64Bit)
2173	{
2174	uListAnchor.u64Pair[1] = uListAnchor.u32Pair[1];
2175	ASMCompilerBarrier();
2176	uListAnchor.u64Pair[0] = uListAnchor.u32Pair[0];
2177	}
2178
2179	/*
2180	* Get a numerical version number.
2181	*/
2182	char szVersion[256] = "Linux version 4.19.0";
2183	bool fValid = pThis->fValid;
2184	pThis->fValid = true;
2185	dbgDiggerLinuxQueryVersion(pUVM, pThis, szVersion, sizeof(szVersion));
2186	pThis->fValid = fValid;
2187
2188	const char *pszVersion = szVersion;
2189	while (pszVersion && !RT_C_IS_DIGIT(pszVersion))
2190	pszVersion++;
2191
2192	size_t offVersion = 0;
2193	uint32_t uMajor = 0;
2194	while (pszVersion[offVersion] && RT_C_IS_DIGIT(pszVersion[offVersion]))
2195	uMajor = uMajor * 10 + pszVersion[offVersion++] - '0';
2196
2197	if (pszVersion[offVersion] == '.')
2198	offVersion++;
2199
2200	uint32_t uMinor = 0;
2201	while (pszVersion[offVersion] && RT_C_IS_DIGIT(pszVersion[offVersion]))
2202	uMinor = uMinor * 10 + pszVersion[offVersion++] - '0';
2203
2204	if (pszVersion[offVersion] == '.')
2205	offVersion++;
2206
2207	uint32_t uBuild = 0;
2208	while (pszVersion[offVersion] && RT_C_IS_DIGIT(pszVersion[offVersion]))
2209	uBuild = uBuild * 10 + pszVersion[offVersion++] - '0';
2210
2211	uint32_t const uGuestVer = LNX_MK_VER(uMajor, uMinor, uBuild);
2212	if (uGuestVer == 0)
2213	{
2214	LogRel(("dbgDiggerLinuxLoadModules: Failed to parse version string: %s\n", pszVersion));
2215	return VERR_NOT_FOUND;
2216	}
2217
2218	/*
2219	* Find the g_aModVersion entry that fits the best.
2220	* ASSUMES strict descending order by bitcount and version.
2221	*/
2222	Assert(g_aModVersions[0].f64Bit == true);
2223	unsigned i = 0;
2224	if (!pThis->f64Bit)
2225	while (i < RT_ELEMENTS(g_aModVersions) && g_aModVersions[i].f64Bit)
2226	i++;
2227	while ( i < RT_ELEMENTS(g_aModVersions)
2228	&& g_aModVersions[i].f64Bit == pThis->f64Bit
2229	&& uGuestVer < g_aModVersions[i].uVersion)
2230	i++;
2231	if (i >= RT_ELEMENTS(g_aModVersions))
2232	{
2233	LogRel(("dbgDiggerLinuxLoadModules: Failed to find anything matching version: %u.%u.%u (%s)\n",
2234	uMajor, uMinor, uBuild, pszVersion));
2235	return VERR_NOT_FOUND;
2236	}
2237
2238	/*
2239	* Walk the list.
2240	*/
2241	uint64_t uModAddr = uListAnchor.u64Pair[0];
2242	for (size_t iModule = 0; iModule < 4096 && uModAddr != SymInfo.Value && uModAddr != 0; iModule++)
2243	uModAddr = g_aModVersions[i].pfnProcessModule(pThis, pUVM, DBGFR3AddrFromFlat(pUVM, &Addr, uModAddr));
2244
2245	return VINF_SUCCESS;
2246	}
2247
2248
2249	/**
2250	* Checks if there is a likely kallsyms_names fragment at pHitAddr.
2251	*
2252	* @returns true if it's a likely fragment, false if not.
2253	* @param pUVM The user mode VM handle.
2254	* @param pHitAddr The address where paNeedle was found.
2255	* @param pabNeedle The fragment we've been searching for.
2256	* @param cbNeedle The length of the fragment.
2257	*/
2258	static bool dbgDiggerLinuxIsLikelyNameFragment(PUVM pUVM, PCDBGFADDRESS pHitAddr, uint8_t const *pabNeedle, uint8_t cbNeedle)
2259	{
2260	/*
2261	* Examples of lead and tail bytes of our choosen needle in a randomly
2262	* picked kernel:
2263	* k o b j
2264	* 22 6b 6f 62 6a aa
2265	* fc 6b 6f 62 6a aa
2266	* 82 6b 6f 62 6a 5f - ascii trail byte (_).
2267	* ee 6b 6f 62 6a aa
2268	* fc 6b 6f 62 6a 5f - ascii trail byte (_).
2269	* 0a 74 6b 6f 62 6a 5f ea - ascii lead (t) and trail (_) bytes.
2270	* 0b 54 6b 6f 62 6a aa - ascii lead byte (T).
2271	* ... omitting 29 samples similar to the last two ...
2272	* d8 6b 6f 62 6a aa
2273	* d8 6b 6f 62 6a aa
2274	* d8 6b 6f 62 6a aa
2275	* d8 6b 6f 62 6a aa
2276	* f9 5f 6b 6f 62 6a 5f 94 - ascii lead and trail bytes (_)
2277	* f9 5f 6b 6f 62 6a 0c - ascii lead byte (_).
2278	* fd 6b 6f 62 6a 0f
2279	* ... enough.
2280	*/
2281	uint8_t abBuf[32];
2282	DBGFADDRESS ReadAddr = *pHitAddr;
2283	DBGFR3AddrSub(&ReadAddr, 2);
2284	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &ReadAddr, abBuf, 2 + cbNeedle + 2);
2285	if (RT_SUCCESS(rc))
2286	{
2287	if (memcmp(&abBuf[2], pabNeedle, cbNeedle) == 0) /* paranoia */
2288	{
2289	uint8_t const bLead = abBuf[1] == '_' \|\| abBuf[1] == 'T' \|\| abBuf[1] == 't' ? abBuf[0] : abBuf[1];
2290	uint8_t const offTail = 2 + cbNeedle;
2291	uint8_t const bTail = abBuf[offTail] == '_' ? abBuf[offTail] : abBuf[offTail + 1];
2292	if ( bLead >= 1 && (bLead < 0x20 \|\| bLead >= 0x80)
2293	&& bTail >= 1 && (bTail < 0x20 \|\| bTail >= 0x80))
2294	return true;
2295	Log(("dbgDiggerLinuxIsLikelyNameFragment: failed at %RGv: bLead=%#x bTail=%#x (offTail=%#x)\n",
2296	pHitAddr->FlatPtr, bLead, bTail, offTail));
2297	}
2298	else
2299	Log(("dbgDiggerLinuxIsLikelyNameFragment: failed at %RGv: Needle changed!\n", pHitAddr->FlatPtr));
2300	}
2301	else
2302	Log(("dbgDiggerLinuxIsLikelyNameFragment: failed at %RGv: %Rrc\n", pHitAddr->FlatPtr, rc));
2303
2304	return false;
2305	}
2306
2307	/**
2308	* Tries to find and load the kernel symbol table with the given needle.
2309	*
2310	* @returns VBox status code.
2311	* @param pThis The Linux digger data.
2312	* @param pUVM The user mode VM handle.
2313	* @param pabNeedle The needle to use for searching.
2314	* @param cbNeedle Size of the needle in bytes.
2315	*/
2316	static int dbgDiggerLinuxFindSymbolTableFromNeedle(PDBGDIGGERLINUX pThis, PUVM pUVM, uint8_t const *pabNeedle, uint8_t cbNeedle)
2317	{
2318	int rc = VINF_SUCCESS;
2319
2320	/*
2321	* Go looking for the kallsyms table. If it's there, it will be somewhere
2322	* after the linux_banner symbol, so use it for starting the search.
2323	*/
2324	DBGFADDRESS CurAddr = pThis->AddrLinuxBanner;
2325	uint32_t cbLeft = LNX_MAX_KERNEL_SIZE;
2326	while (cbLeft > 4096)
2327	{
2328	DBGFADDRESS HitAddr;
2329	rc = DBGFR3MemScan(pUVM, 0 /idCpu/, &CurAddr, cbLeft, 1 /uAlign/,
2330	pabNeedle, cbNeedle, &HitAddr);
2331	if (RT_FAILURE(rc))
2332	break;
2333	if (dbgDiggerLinuxIsLikelyNameFragment(pUVM, &HitAddr, pabNeedle, cbNeedle))
2334	{
2335	/* There will be another hit near by. */
2336	DBGFR3AddrAdd(&HitAddr, 1);
2337	rc = DBGFR3MemScan(pUVM, 0 /idCpu/, &HitAddr, LNX_MAX_KALLSYMS_NAMES_SIZE, 1 /uAlign/,
2338	pabNeedle, cbNeedle, &HitAddr);
2339	if ( RT_SUCCESS(rc)
2340	&& dbgDiggerLinuxIsLikelyNameFragment(pUVM, &HitAddr, pabNeedle, cbNeedle))
2341	{
2342	/*
2343	* We've got a very likely candidate for a location inside kallsyms_names.
2344	* Try find the start of it, that is to say, try find kallsyms_num_syms.
2345	* kallsyms_num_syms is aligned on sizeof(unsigned long) boundrary
2346	*/
2347	rc = dbgDiggerLinuxFindStartOfNamesAndSymbolCount(pUVM, pThis, &HitAddr);
2348	if (RT_SUCCESS(rc))
2349	rc = dbgDiggerLinuxFindEndOfNamesAndMore(pUVM, pThis, &HitAddr);
2350	if (RT_SUCCESS(rc))
2351	rc = dbgDiggerLinuxFindTokenIndex(pUVM, pThis);
2352	if (RT_SUCCESS(rc))
2353	rc = dbgDiggerLinuxLoadKernelSymbols(pUVM, pThis);
2354	if (RT_SUCCESS(rc))
2355	{
2356	rc = dbgDiggerLinuxLoadModules(pThis, pUVM);
2357	break;
2358	}
2359	}
2360	}
2361
2362	/*
2363	* Advance.
2364	*/
2365	RTGCUINTPTR cbDistance = HitAddr.FlatPtr - CurAddr.FlatPtr + cbNeedle;
2366	if (RT_UNLIKELY(cbDistance >= cbLeft))
2367	{
2368	Log(("dbgDiggerLinuxInit: Failed to find kallsyms\n"));
2369	break;
2370	}
2371	cbLeft -= cbDistance;
2372	DBGFR3AddrAdd(&CurAddr, cbDistance);
2373
2374	}
2375
2376	return rc;
2377	}
2378
2379	/**
2380	* Skips whitespace and comments in the given config returning the pointer
2381	* to the first non whitespace character.
2382	*
2383	* @returns Pointer to the first non whitespace character or NULL if the end
2384	* of the string was reached.
2385	* @param pszCfg The config string.
2386	*/
2387	static const char dbgDiggerLinuxCfgSkipWhitespace(const char pszCfg)
2388	{
2389	do
2390	{
2391	while ( *pszCfg != '\0'
2392	&& ( RT_C_IS_SPACE(*pszCfg)
2393	\|\| *pszCfg == '\n'))
2394	pszCfg++;
2395
2396	/* Do we have a comment? Skip it. */
2397	if (*pszCfg == '#')
2398	{
2399	while ( *pszCfg != '\n'
2400	&& *pszCfg != '\0')
2401	pszCfg++;
2402	}
2403	} while ( *pszCfg != '\0'
2404	&& ( RT_C_IS_SPACE(*pszCfg)
2405	\|\| *pszCfg == '\n'
2406	\|\| *pszCfg == '#'));
2407
2408	return pszCfg;
2409	}
2410
2411	/**
2412	* Parses an identifier at the given position.
2413	*
2414	* @returns VBox status code.
2415	* @param pszCfg The config data.
2416	* @param ppszCfgNext Where to store the pointer to the data following the identifier.
2417	* @param ppszIde Where to store the pointer to the identifier on success.
2418	* Free with RTStrFree().
2419	*/
2420	static int dbgDiggerLinuxCfgParseIde(const char pszCfg, const char ppszCfgNext, char *ppszIde)
2421	{
2422	int rc = VINF_SUCCESS;
2423	size_t cchIde = 0;
2424
2425	while ( *pszCfg != '\0'
2426	&& ( RT_C_IS_ALNUM(*pszCfg)
2427	\|\| *pszCfg == '_'))
2428	{
2429	cchIde++;
2430	pszCfg++;
2431	}
2432
2433	if (cchIde)
2434	{
2435	*ppszIde = RTStrDupN(pszCfg - cchIde, cchIde);
2436	if (!*ppszIde)
2437	rc = VERR_NO_STR_MEMORY;
2438	}
2439
2440	*ppszCfgNext = pszCfg;
2441	return rc;
2442	}
2443
2444	/**
2445	* Parses a value for a config item.
2446	*
2447	* @returns VBox status code.
2448	* @param pszCfg The config data.
2449	* @param ppszCfgNext Where to store the pointer to the data following the identifier.
2450	* @param ppCfgItem Where to store the created config item on success.
2451	*/
2452	static int dbgDiggerLinuxCfgParseVal(const char pszCfg, const char *ppszCfgNext,
2453	PDBGDIGGERLINUXCFGITEM *ppCfgItem)
2454	{
2455	int rc = VINF_SUCCESS;
2456	PDBGDIGGERLINUXCFGITEM pCfgItem = NULL;
2457
2458	if (RT_C_IS_DIGIT(pszCfg) \|\| pszCfg == '-')
2459	{
2460	/* Parse the number. */
2461	int64_t i64Num;
2462	rc = RTStrToInt64Ex(pszCfg, (char **)ppszCfgNext, 0, &i64Num);
2463	if ( RT_SUCCESS(rc)
2464	\|\| rc == VWRN_TRAILING_CHARS
2465	\|\| rc == VWRN_TRAILING_SPACES)
2466	{
2467	pCfgItem = (PDBGDIGGERLINUXCFGITEM)RTMemAllocZ(sizeof(DBGDIGGERLINUXCFGITEM));
2468	if (pCfgItem)
2469	{
2470	pCfgItem->enmType = DBGDIGGERLINUXCFGITEMTYPE_NUMBER;
2471	pCfgItem->u.i64Num = i64Num;
2472	}
2473	else
2474	rc = VERR_NO_MEMORY;
2475	}
2476	}
2477	else if (*pszCfg == '\"')
2478	{
2479	/* Parse a string. */
2480	const char *pszCfgCur = pszCfg + 1;
2481	while ( *pszCfgCur != '\0'
2482	&& *pszCfgCur != '\"')
2483	pszCfgCur++;
2484
2485	if (*pszCfgCur == '\"')
2486	{
2487	pCfgItem = (PDBGDIGGERLINUXCFGITEM)RTMemAllocZ(RT_UOFFSETOF_DYN(DBGDIGGERLINUXCFGITEM,
2488	u.aszString[pszCfgCur - pszCfg + 1]));
2489	if (pCfgItem)
2490	{
2491	pCfgItem->enmType = DBGDIGGERLINUXCFGITEMTYPE_STRING;
2492	RTStrCopyEx(&pCfgItem->u.aszString[0], pszCfgCur - pszCfg + 1, pszCfg, pszCfgCur - pszCfg);
2493	*ppszCfgNext = pszCfgCur + 1;
2494	}
2495	else
2496	rc = VERR_NO_MEMORY;
2497	}
2498	else
2499	rc = VERR_INVALID_STATE;
2500	}
2501	else if ( *pszCfg == 'y'
2502	\|\| *pszCfg == 'm')
2503	{
2504	/* Included or module. */
2505	pCfgItem = (PDBGDIGGERLINUXCFGITEM)RTMemAllocZ(sizeof(DBGDIGGERLINUXCFGITEM));
2506	if (pCfgItem)
2507	{
2508	pCfgItem->enmType = DBGDIGGERLINUXCFGITEMTYPE_FLAG;
2509	pCfgItem->u.fModule = *pszCfg == 'm';
2510	}
2511	else
2512	rc = VERR_NO_MEMORY;
2513	pszCfg++;
2514	*ppszCfgNext = pszCfg;
2515	}
2516	else
2517	rc = VERR_INVALID_STATE;
2518
2519	if (RT_SUCCESS(rc))
2520	*ppCfgItem = pCfgItem;
2521	else if (pCfgItem)
2522	RTMemFree(pCfgItem);
2523
2524	return rc;
2525	}
2526
2527	/**
2528	* Parses the given kernel config and creates the config database.
2529	*
2530	* @returns VBox status code
2531	* @param pThis The Linux digger data.
2532	* @param pszCfg The config string.
2533	*/
2534	static int dbgDiggerLinuxCfgParse(PDBGDIGGERLINUX pThis, const char *pszCfg)
2535	{
2536	int rc = VINF_SUCCESS;
2537
2538	/*
2539	* The config is a text file with the following elements:
2540	* # starts a comment which goes till the end of the line
2541	* <Ide>=<val> where <Ide> is an identifier consisting of
2542	* alphanumerical characters (including _)
2543	* <val> denotes the value for the identifier and can have the following
2544	* formats:
2545	* (-)[0-9]* for numbers
2546	* "..." for a string value
2547	* m when a feature is enabled as a module
2548	* y when a feature is enabled
2549	* Newlines are used as a separator between values and mark the end
2550	* of a comment
2551	*/
2552	const char *pszCfgCur = pszCfg;
2553	while ( RT_SUCCESS(rc)
2554	&& *pszCfgCur != '\0')
2555	{
2556	/* Start skipping the whitespace. */
2557	pszCfgCur = dbgDiggerLinuxCfgSkipWhitespace(pszCfgCur);
2558	if ( pszCfgCur
2559	&& *pszCfgCur != '\0')
2560	{
2561	char *pszIde = NULL;
2562	/* Must be an identifier, parse it. */
2563	rc = dbgDiggerLinuxCfgParseIde(pszCfgCur, &pszCfgCur, &pszIde);
2564	if (RT_SUCCESS(rc))
2565	{
2566	/*
2567	* Skip whitespace again (shouldn't be required because = follows immediately
2568	* in the observed configs).
2569	*/
2570	pszCfgCur = dbgDiggerLinuxCfgSkipWhitespace(pszCfgCur);
2571	if ( pszCfgCur
2572	&& *pszCfgCur == '=')
2573	{
2574	pszCfgCur++;
2575	pszCfgCur = dbgDiggerLinuxCfgSkipWhitespace(pszCfgCur);
2576	if ( pszCfgCur
2577	&& *pszCfgCur != '\0')
2578	{
2579	/* Get the value. */
2580	PDBGDIGGERLINUXCFGITEM pCfgItem = NULL;
2581	rc = dbgDiggerLinuxCfgParseVal(pszCfgCur, &pszCfgCur, &pCfgItem);
2582	if (RT_SUCCESS(rc))
2583	{
2584	pCfgItem->Core.pszString = pszIde;
2585	bool fRc = RTStrSpaceInsert(&pThis->hCfgDb, &pCfgItem->Core);
2586	if (!fRc)
2587	{
2588	RTStrFree(pszIde);
2589	RTMemFree(pCfgItem);
2590	rc = VERR_INVALID_STATE;
2591	}
2592	}
2593	}
2594	else
2595	rc = VERR_EOF;
2596	}
2597	else
2598	rc = VERR_INVALID_STATE;
2599	}
2600
2601	if (RT_FAILURE(rc))
2602	RTStrFree(pszIde);
2603	}
2604	else
2605	break; /* Reached the end of the config. */
2606	}
2607
2608	if (RT_FAILURE(rc))
2609	dbgDiggerLinuxCfgDbDestroy(pThis);
2610
2611	return rc;
2612	}
2613
2614	/**
2615	* Decompresses the given config and validates the UTF-8 encoding.
2616	*
2617	* @returns VBox status code.
2618	* @param pbCfgComp The compressed config.
2619	* @param cbCfgComp Size of the compressed config.
2620	* @param ppszCfg Where to store the pointer to the decompressed config
2621	* on success.
2622	*/
2623	static int dbgDiggerLinuxCfgDecompress(const uint8_t pbCfgComp, size_t cbCfgComp, char *ppszCfg)
2624	{
2625	int rc = VINF_SUCCESS;
2626	RTVFSIOSTREAM hVfsIos = NIL_RTVFSIOSTREAM;
2627
2628	rc = RTVfsIoStrmFromBuffer(RTFILE_O_READ, pbCfgComp, cbCfgComp, &hVfsIos);
2629	if (RT_SUCCESS(rc))
2630	{
2631	RTVFSIOSTREAM hVfsIosDecomp = NIL_RTVFSIOSTREAM;
2632	rc = RTZipGzipDecompressIoStream(hVfsIos, RTZIPGZIPDECOMP_F_ALLOW_ZLIB_HDR, &hVfsIosDecomp);
2633	if (RT_SUCCESS(rc))
2634	{
2635	char *pszCfg = NULL;
2636	size_t cchCfg = 0;
2637	size_t cbRead = 0;
2638
2639	do
2640	{
2641	uint8_t abBuf[_64K];
2642	rc = RTVfsIoStrmRead(hVfsIosDecomp, abBuf, sizeof(abBuf), true /fBlocking/, &cbRead);
2643	if (rc == VINF_EOF && cbRead == 0)
2644	rc = VINF_SUCCESS;
2645	if ( RT_SUCCESS(rc)
2646	&& cbRead > 0)
2647	{
2648	/* Append data. */
2649	char *pszCfgNew = pszCfg;
2650	rc = RTStrRealloc(&pszCfgNew, cchCfg + cbRead + 1);
2651	if (RT_SUCCESS(rc))
2652	{
2653	pszCfg = pszCfgNew;
2654	memcpy(pszCfg + cchCfg, &abBuf[0], cbRead);
2655	cchCfg += cbRead;
2656	pszCfg[cchCfg] = '\0'; /* Enforce string termination. */
2657	}
2658	}
2659	} while (RT_SUCCESS(rc) && cbRead > 0);
2660
2661	if (RT_SUCCESS(rc))
2662	*ppszCfg = pszCfg;
2663	else if (RT_FAILURE(rc) && pszCfg)
2664	RTStrFree(pszCfg);
2665
2666	RTVfsIoStrmRelease(hVfsIosDecomp);
2667	}
2668	RTVfsIoStrmRelease(hVfsIos);
2669	}
2670
2671	return rc;
2672	}
2673
2674	/**
2675	* Reads and decodes the compressed kernel config.
2676	*
2677	* @returns VBox status code.
2678	* @param pThis The Linux digger data.
2679	* @param pUVM The user mode VM handle.
2680	* @param pAddrStart The start address of the compressed config.
2681	* @param cbCfgComp The size of the compressed config.
2682	*/
2683	static int dbgDiggerLinuxCfgDecode(PDBGDIGGERLINUX pThis, PUVM pUVM,
2684	PCDBGFADDRESS pAddrStart, size_t cbCfgComp)
2685	{
2686	int rc = VINF_SUCCESS;
2687	uint8_t pbCfgComp = (uint8_t )RTMemTmpAlloc(cbCfgComp);
2688	if (!pbCfgComp)
2689	return VERR_NO_MEMORY;
2690
2691	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, pAddrStart, pbCfgComp, cbCfgComp);
2692	if (RT_SUCCESS(rc))
2693	{
2694	char *pszCfg = NULL;
2695	rc = dbgDiggerLinuxCfgDecompress(pbCfgComp, cbCfgComp, &pszCfg);
2696	if (RT_SUCCESS(rc))
2697	{
2698	if (RTStrIsValidEncoding(pszCfg))
2699	rc = dbgDiggerLinuxCfgParse(pThis, pszCfg);
2700	else
2701	rc = VERR_INVALID_UTF8_ENCODING;
2702	RTStrFree(pszCfg);
2703	}
2704	}
2705
2706	RTMemFree(pbCfgComp);
2707	return rc;
2708	}
2709
2710	/**
2711	* Tries to find the compressed kernel config in the kernel address space
2712	* and sets up the config database.
2713	*
2714	* @returns VBox status code.
2715	* @param pThis The Linux digger data.
2716	* @param pUVM The user mode VM handle.
2717	*/
2718	static int dbgDiggerLinuxCfgFind(PDBGDIGGERLINUX pThis, PUVM pUVM)
2719	{
2720	int rc = VINF_SUCCESS;
2721
2722	/*
2723	* Go looking for the IKCFG_ST string which indicates the start
2724	* of the compressed config file.
2725	*/
2726	static const uint8_t s_abCfgNeedleStart[] = "IKCFG_ST";
2727	static const uint8_t s_abCfgNeedleEnd[] = "IKCFG_ED";
2728	DBGFADDRESS CurAddr = pThis->AddrLinuxBanner;
2729	uint32_t cbLeft = LNX_MAX_KERNEL_SIZE;
2730	while (cbLeft > 4096)
2731	{
2732	DBGFADDRESS HitAddrStart;
2733	rc = DBGFR3MemScan(pUVM, 0 /idCpu/, &CurAddr, cbLeft, 1 /uAlign/,
2734	s_abCfgNeedleStart, sizeof(s_abCfgNeedleStart) - 1, &HitAddrStart);
2735	if (RT_FAILURE(rc))
2736	break;
2737
2738	/* Check for the end marker which shouldn't be that far away. */
2739	DBGFR3AddrAdd(&HitAddrStart, sizeof(s_abCfgNeedleStart) - 1);
2740	DBGFADDRESS HitAddrEnd;
2741	rc = DBGFR3MemScan(pUVM, 0 /* idCpu */, &HitAddrStart, LNX_MAX_COMPRESSED_CFG_SIZE,
2742	1 /* uAlign */, s_abCfgNeedleEnd, sizeof(s_abCfgNeedleEnd) - 1, &HitAddrEnd);
2743	if (RT_SUCCESS(rc))
2744	{
2745	/* Allocate a buffer to hold the compressed data between the markers and fetch it. */
2746	RTGCUINTPTR cbCfg = HitAddrEnd.FlatPtr - HitAddrStart.FlatPtr;
2747	Assert(cbCfg == (size_t)cbCfg);
2748	rc = dbgDiggerLinuxCfgDecode(pThis, pUVM, &HitAddrStart, cbCfg);
2749	if (RT_SUCCESS(rc))
2750	break;
2751	}
2752
2753	/*
2754	* Advance.
2755	*/
2756	RTGCUINTPTR cbDistance = HitAddrStart.FlatPtr - CurAddr.FlatPtr + sizeof(s_abCfgNeedleStart) - 1;
2757	if (RT_UNLIKELY(cbDistance >= cbLeft))
2758	{
2759	LogFunc(("Failed to find compressed kernel config\n"));
2760	break;
2761	}
2762	cbLeft -= cbDistance;
2763	DBGFR3AddrAdd(&CurAddr, cbDistance);
2764
2765	}
2766
2767	return rc;
2768	}
2769
2770	/**
2771	* Probes for a Linux kernel starting at the given address.
2772	*
2773	* @returns Flag whether something which looks like a valid Linux kernel was found.
2774	* @param pThis The Linux digger data.
2775	* @param pUVM The user mode VM handle.
2776	* @param uAddrStart The address to start scanning at.
2777	* @param cbScan How much to scan.
2778	*/
2779	static bool dbgDiggerLinuxProbeWithAddr(PDBGDIGGERLINUX pThis, PUVM pUVM, RTGCUINTPTR uAddrStart, size_t cbScan)
2780	{
2781	/*
2782	* Look for "Linux version " at the start of the rodata segment.
2783	* Hope that this comes before any message buffer or other similar string.
2784	*/
2785	DBGFADDRESS KernelAddr;
2786	DBGFR3AddrFromFlat(pUVM, &KernelAddr, uAddrStart);
2787	DBGFADDRESS HitAddr;
2788	int rc = DBGFR3MemScan(pUVM, 0, &KernelAddr, cbScan, 1,
2789	g_abLinuxVersion, sizeof(g_abLinuxVersion) - 1, &HitAddr);
2790	if (RT_SUCCESS(rc))
2791	{
2792	char szTmp[128];
2793	char const *pszX = &szTmp[sizeof(g_abLinuxVersion) - 1];
2794	rc = DBGFR3MemReadString(pUVM, 0, &HitAddr, szTmp, sizeof(szTmp));
2795	if ( RT_SUCCESS(rc)
2796	&& ( ( pszX[0] == '2' /* 2.x.y with x in {0..6} */
2797	&& pszX[1] == '.'
2798	&& pszX[2] >= '0'
2799	&& pszX[2] <= '6')
2800	\|\| ( pszX[0] >= '3' /* 3.x, 4.x, ... 9.x */
2801	&& pszX[0] <= '9'
2802	&& pszX[1] == '.'
2803	&& pszX[2] >= '0'
2804	&& pszX[2] <= '9')
2805	)
2806	)
2807	{
2808	pThis->AddrKernelBase = KernelAddr;
2809	pThis->AddrLinuxBanner = HitAddr;
2810	return true;
2811	}
2812	}
2813
2814	return false;
2815	}
2816
2817	/**
2818	* Probes for a Linux kernel which has KASLR enabled.
2819	*
2820	* @returns Flag whether a possible candidate location was found.
2821	* @param pThis The Linux digger data.
2822	* @param pUVM The user mode VM handle.
2823	*/
2824	static bool dbgDiggerLinuxProbeKaslr(PDBGDIGGERLINUX pThis, PUVM pUVM)
2825	{
2826	/**
2827	* With KASLR the kernel is loaded at a different address at each boot making detection
2828	* more difficult for us.
2829	*
2830	* The randomization is done in arch/x86/boot/compressed/kaslr.c:choose_random_location() (as of Nov 2017).
2831	* At the end of the method a random offset is chosen using find_random_virt_addr() which is added to the
2832	* kernel map start in the caller (the start of the kernel depends on the bit size, see LNX32_KERNEL_ADDRESS_START
2833	* and LNX64_KERNEL_ADDRESS_START for 32bit and 64bit kernels respectively).
2834	* The lowest offset possible is LOAD_PHYSICAL_ADDR which is defined in arch/x86/include/asm/boot.h
2835	* using CONFIG_PHYSICAL_START aligned to CONFIG_PHYSICAL_ALIGN.
2836	* The default CONFIG_PHYSICAL_START and CONFIG_PHYSICAL_ALIGN are both 0x1000000 no matter whether a 32bit
2837	* or a 64bit kernel is used. So the lowest offset to the kernel start address is 0x1000000.
2838	* The find_random_virt_addr() the number of possible slots where the kernel can be placed based on the image size
2839	* is calculated using the following formula:
2840	* cSlots = ((KERNEL_IMAGE_SIZE - 0x1000000 (minimum) - image_size) / 0x1000000 (CONFIG_PHYSICAL_ALIGN)) + 1
2841	*
2842	* KERNEL_IMAGE_SIZE is 1GB for 64bit kernels and 512MB for 32bit kernels, so the maximum number of slots (resulting
2843	* in the largest possible offset) can be achieved when image_size (which contains the real size of the kernel image
2844	* which is unknown for us) goes to 0 and a 1GB KERNEL_IMAGE_SIZE is assumed. With that the biggest cSlots which can be
2845	* achieved is 64. The chosen random offset is taken from a random long integer using kaslr_get_random_long() modulo the
2846	* number of slots which selects a slot between 0 and 63. The final offset is calculated using:
2847	* offAddr = random_addr * 0x1000000 (CONFIG_PHYSICAL_ALIGN) + 0x1000000 (minimum)
2848	*
2849	* So the highest offset the kernel can start is 0x40000000 which is 1GB (plus the maximum kernel size we defined).
2850	*/
2851	if (dbgDiggerLinuxProbeWithAddr(pThis, pUVM, LNX64_KERNEL_ADDRESS_START, _1G + LNX_MAX_KERNEL_SIZE))
2852	return true;
2853
2854	/*
2855	* 32bit variant, makes sure we don't exceed the 4GB address space or DBGFR3MemScan() returns VERR_DBGF_MEM_NOT_FOUND immediately
2856	* without searching the remainder of the address space.
2857	*
2858	* The default split is 3GB userspace and 1GB kernel, so we just search the entire upper 1GB kernel space.
2859	*/
2860	if (dbgDiggerLinuxProbeWithAddr(pThis, pUVM, LNX32_KERNEL_ADDRESS_START, _4G - LNX32_KERNEL_ADDRESS_START))
2861	return true;
2862
2863	return false;
2864	}
2865
2866	/**
2867	* @copydoc DBGFOSREG::pfnInit
2868	*/
2869	static DECLCALLBACK(int) dbgDiggerLinuxInit(PUVM pUVM, void *pvData)
2870	{
2871	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
2872	Assert(!pThis->fValid);
2873
2874	/*
2875	* Assume 64-bit kernels all live way beyond 32-bit address space.
2876	*/
2877	pThis->f64Bit = pThis->AddrLinuxBanner.FlatPtr > UINT32_MAX;
2878	pThis->fRelKrnlAddr = false;
2879
2880	pThis->hCfgDb = NULL;
2881
2882	/*
2883	* Try to find the compressed kernel config and parse it before we try
2884	* to get the symbol table, the config database is required to select
2885	* the method to use.
2886	*/
2887	int rc = dbgDiggerLinuxCfgFind(pThis, pUVM);
2888	if (RT_FAILURE(rc))
2889	LogFlowFunc(("Failed to find kernel config (%Rrc), no config database available\n", rc));
2890
2891	static const uint8_t s_abNeedle[] = "kobj";
2892	rc = dbgDiggerLinuxFindSymbolTableFromNeedle(pThis, pUVM, s_abNeedle, sizeof(s_abNeedle) - 1);
2893	if (RT_FAILURE(rc))
2894	{
2895	/* Try alternate needle (seen on older x86 Linux kernels). */
2896	static const uint8_t s_abNeedleAlt[] = "kobjec";
2897	rc = dbgDiggerLinuxFindSymbolTableFromNeedle(pThis, pUVM, s_abNeedleAlt, sizeof(s_abNeedleAlt) - 1);
2898	if (RT_FAILURE(rc))
2899	{
2900	static const uint8_t s_abNeedleOSuseX86[] = "nmi"; /* OpenSuSe 10.2 x86 */
2901	rc = dbgDiggerLinuxFindSymbolTableFromNeedle(pThis, pUVM, s_abNeedleOSuseX86, sizeof(s_abNeedleOSuseX86) - 1);
2902	}
2903	}
2904
2905	pThis->fValid = true;
2906	return VINF_SUCCESS;
2907	}
2908
2909
2910	/**
2911	* @copydoc DBGFOSREG::pfnProbe
2912	*/
2913	static DECLCALLBACK(bool) dbgDiggerLinuxProbe(PUVM pUVM, void *pvData)
2914	{
2915	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
2916
2917	for (unsigned i = 0; i < RT_ELEMENTS(g_au64LnxKernelAddresses); i++)
2918	{
2919	if (dbgDiggerLinuxProbeWithAddr(pThis, pUVM, g_au64LnxKernelAddresses[i], LNX_MAX_KERNEL_SIZE))
2920	return true;
2921	}
2922
2923	/* Maybe the kernel uses KASLR. */
2924	if (dbgDiggerLinuxProbeKaslr(pThis, pUVM))
2925	return true;
2926
2927	return false;
2928	}
2929
2930
2931	/**
2932	* @copydoc DBGFOSREG::pfnDestruct
2933	*/
2934	static DECLCALLBACK(void) dbgDiggerLinuxDestruct(PUVM pUVM, void *pvData)
2935	{
2936	RT_NOREF2(pUVM, pvData);
2937	}
2938
2939
2940	/**
2941	* @copydoc DBGFOSREG::pfnConstruct
2942	*/
2943	static DECLCALLBACK(int) dbgDiggerLinuxConstruct(PUVM pUVM, void *pvData)
2944	{
2945	RT_NOREF1(pUVM);
2946	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
2947	pThis->IDmesg.u32Magic = DBGFOSIDMESG_MAGIC;
2948	pThis->IDmesg.pfnQueryKernelLog = dbgDiggerLinuxIDmsg_QueryKernelLog;
2949	pThis->IDmesg.u32EndMagic = DBGFOSIDMESG_MAGIC;
2950
2951	return VINF_SUCCESS;
2952	}
2953
2954
2955	const DBGFOSREG g_DBGDiggerLinux =
2956	{
2957	/* .u32Magic = */ DBGFOSREG_MAGIC,
2958	/* .fFlags = */ 0,
2959	/* .cbData = */ sizeof(DBGDIGGERLINUX),
2960	/* .szName = */ "Linux",
2961	/* .pfnConstruct = */ dbgDiggerLinuxConstruct,
2962	/* .pfnDestruct = */ dbgDiggerLinuxDestruct,
2963	/* .pfnProbe = */ dbgDiggerLinuxProbe,
2964	/* .pfnInit = */ dbgDiggerLinuxInit,
2965	/* .pfnRefresh = */ dbgDiggerLinuxRefresh,
2966	/* .pfnTerm = */ dbgDiggerLinuxTerm,
2967	/* .pfnQueryVersion = */ dbgDiggerLinuxQueryVersion,
2968	/* .pfnQueryInterface = */ dbgDiggerLinuxQueryInterface,
2969	/* .pfnStackUnwindAssist = */ dbgDiggerLinuxStackUnwindAssist,
2970	/* .u32EndMagic = */ DBGFOSREG_MAGIC
2971	};
2972

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

source: vbox/trunk/src/VBox/Debugger/DBGPlugInLinux.cpp@ 86719

Download in other formats: