DBGPlugInLinux.cpp@ 77874

Last change on this file since 77874 was 77874, checked in by vboxsync, 6 years ago
DBGPlugInLinux: Started looking for kernel modules. Works to some extent for 2.6.24.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 111.7 KB

Line
1	/* $Id: DBGPlugInLinux.cpp 77874 2019-03-26 01:37:19Z vboxsync $ */
2	/** @file
3	* DBGPlugInLinux - Debugger and Guest OS Digger Plugin For Linux.
4	*/
5
6	/*
7	* Copyright (C) 2008-2019 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
185	/** The maximum size we expect for kallsyms_names. */
186	#define LNX_MAX_KALLSYMS_NAMES_SIZE UINT32_C(0x200000)
187	/** The maximum size we expect for kallsyms_token_table. */
188	#define LNX_MAX_KALLSYMS_TOKEN_TABLE_SIZE UINT32_C(0x10000)
189	/** The minimum number of symbols we expect in kallsyms_num_syms. */
190	#define LNX_MIN_KALLSYMS_SYMBOLS UINT32_C(2048)
191	/** The maximum number of symbols we expect in kallsyms_num_syms. */
192	#define LNX_MAX_KALLSYMS_SYMBOLS UINT32_C(1048576)
193	/** The min length an encoded symbol in kallsyms_names is expected to have. */
194	#define LNX_MIN_KALLSYMS_ENC_LENGTH UINT8_C(1)
195	/** The max length an encoded symbol in kallsyms_names is expected to have.
196	* @todo check real life here. */
197	#define LNX_MAX_KALLSYMS_ENC_LENGTH UINT8_C(28)
198	/** The approximate maximum length of a string token. */
199	#define LNX_MAX_KALLSYMS_TOKEN_LEN UINT16_C(32)
200	/** Maximum compressed config size expected. */
201	#define LNX_MAX_COMPRESSED_CFG_SIZE _1M
202
203	/** Module tag for linux ('linuxmod' on little endian ASCII systems). */
204	#define DIG_LNX_MOD_TAG UINT64_C(0x545f5d78758e898c)
205
206
207	/*********************************************************************************************************************************
208	* Internal Functions *
209	*********************************************************************************************************************************/
210	static DECLCALLBACK(int) dbgDiggerLinuxInit(PUVM pUVM, void *pvData);
211
212
213	/*********************************************************************************************************************************
214	* Global Variables *
215	*********************************************************************************************************************************/
216	/** Table of common linux kernel addresses. */
217	static uint64_t g_au64LnxKernelAddresses[] =
218	{
219	UINT64_C(0xc0100000),
220	UINT64_C(0x90100000),
221	UINT64_C(0xffffffff80200000)
222	};
223
224	static const uint8_t g_abLinuxVersion[] = "Linux version ";
225
226	/**
227	* Converts a given offset into an absolute address if relative kernel offsets are used for
228	* kallsyms.
229	*
230	* @returns The absolute kernel address.
231	* @param pThis The Linux digger data.
232	* @param uOffset The offset to convert.
233	*/
234	DECLINLINE(RTGCUINTPTR) dbgDiggerLinuxConvOffsetToAddr(PDBGDIGGERLINUX pThis, int32_t uOffset)
235	{
236	RTGCUINTPTR uAddr;
237
238	/*
239	* How the absolute address is calculated from the offset depends on the
240	* CONFIG_KALLSYMS_ABSOLUTE_PERCPU config which is only set for 64bit
241	* SMP kernels (we assume that all 64bit kernels always have SMP enabled too).
242	*/
243	if (pThis->f64Bit)
244	{
245	if (uOffset >= 0)
246	uAddr = uOffset;
247	else
248	uAddr = pThis->uKernelRelativeBase - 1 - uOffset;
249	}
250	else
251	uAddr = pThis->uKernelRelativeBase + (uint32_t)uOffset;
252
253	return uAddr;
254	}
255
256	/**
257	* Disassembles a simple getter returning the value for it.
258	*
259	* @returns VBox status code.
260	* @param pThis The Linux digger data.
261	* @param pUVM The VM handle.
262	* @param hMod The module to use.
263	* @param pszSymbol The symbol of the getter.
264	* @param pvVal Where to store the value on success.
265	* @param cbVal Size of the value in bytes.
266	*/
267	static int dbgDiggerLinuxDisassembleSimpleGetter(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
268	const char pszSymbol, void pvVal, uint32_t cbVal)
269	{
270	int rc = VINF_SUCCESS;
271
272	RTDBGSYMBOL SymInfo;
273	rc = RTDbgModSymbolByName(hMod, pszSymbol, &SymInfo);
274	if (RT_SUCCESS(rc))
275	{
276	/*
277	* Do the diassembling. Disassemble until a ret instruction is encountered
278	* or a limit is reached (don't want to disassemble for too long as the getter
279	* should be short).
280	* push and pop instructions are skipped as well as any mov instructions not
281	* touching the rax or eax register (depending on the size of the value).
282	*/
283	unsigned cInstrDisassembled = 0;
284	uint32_t offInstr = 0;
285	bool fRet = false;
286	DISSTATE DisState;
287	RT_ZERO(DisState);
288
289	do
290	{
291	DBGFADDRESS Addr;
292	RTGCPTR GCPtrCur = (RTGCPTR)SymInfo.Value + pThis->AddrKernelBase.FlatPtr + offInstr;
293	DBGFR3AddrFromFlat(pUVM, &Addr, GCPtrCur);
294
295	/* Prefetch the instruction. */
296	uint8_t abInstr[32];
297	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &Addr, &abInstr[0], sizeof(abInstr));
298	if (RT_SUCCESS(rc))
299	{
300	uint32_t cbInstr = 0;
301
302	rc = DISInstr(&abInstr[0], pThis->f64Bit ? DISCPUMODE_64BIT : DISCPUMODE_32BIT, &DisState, &cbInstr);
303	if (RT_SUCCESS(rc))
304	{
305	switch (DisState.pCurInstr->uOpcode)
306	{
307	case OP_PUSH:
308	case OP_POP:
309	case OP_NOP:
310	case OP_LEA:
311	break;
312	case OP_RETN:
313	/* Getter returned, abort disassembling. */
314	fRet = true;
315	break;
316	case OP_MOV:
317	/*
318	* Check that the destination is either rax or eax depending on the
319	* value size.
320	*
321	* Param1 is the destination and Param2 the source.
322	*/
323	if ( ( ( (DisState.Param1.fUse & (DISUSE_BASE \| DISUSE_REG_GEN32))
324	&& cbVal == sizeof(uint32_t))
325	\|\| ( (DisState.Param1.fUse & (DISUSE_BASE \| DISUSE_REG_GEN64))
326	&& cbVal == sizeof(uint64_t)))
327	&& DisState.Param1.Base.idxGenReg == DISGREG_RAX)
328	{
329	/* Parse the source. */
330	if (DisState.Param2.fUse & (DISUSE_IMMEDIATE32 \| DISUSE_IMMEDIATE64))
331	memcpy(pvVal, &DisState.Param2.uValue, cbVal);
332	else if (DisState.Param2.fUse & (DISUSE_RIPDISPLACEMENT32\|DISUSE_DISPLACEMENT32\|DISUSE_DISPLACEMENT64))
333	{
334	RTGCPTR GCPtrVal = 0;
335
336	if (DisState.Param2.fUse & DISUSE_RIPDISPLACEMENT32)
337	GCPtrVal = GCPtrCur + DisState.Param2.uDisp.i32 + cbInstr;
338	else if (DisState.Param2.fUse & DISUSE_DISPLACEMENT32)
339	GCPtrVal = (RTGCPTR)DisState.Param2.uDisp.u32;
340	else if (DisState.Param2.fUse & DISUSE_DISPLACEMENT64)
341	GCPtrVal = (RTGCPTR)DisState.Param2.uDisp.u64;
342	else
343	AssertMsgFailedBreakStmt(("Invalid displacement\n"), rc = VERR_INVALID_STATE);
344
345	DBGFADDRESS AddrVal;
346	rc = DBGFR3MemRead(pUVM, 0 /idCpu/,
347	DBGFR3AddrFromFlat(pUVM, &AddrVal, GCPtrVal),
348	pvVal, cbVal);
349	}
350	}
351	break;
352	default:
353	/* All other instructions will cause an error for now (playing safe here). */
354	rc = VERR_INVALID_PARAMETER;
355	break;
356	}
357	cInstrDisassembled++;
358	offInstr += cbInstr;
359	}
360	}
361	} while ( RT_SUCCESS(rc)
362	&& cInstrDisassembled < 20
363	&& !fRet);
364	}
365
366	return rc;
367	}
368
369	/**
370	* Try to get at the log buffer starting address and size by disassembling emit_log_char.
371	*
372	* @returns VBox status code.
373	* @param pThis The Linux digger data.
374	* @param pUVM The VM handle.
375	* @param hMod The module to use.
376	* @param pGCPtrLogBuf Where to store the log buffer pointer on success.
377	* @param pcbLogBuf Where to store the size of the log buffer on success.
378	*/
379	static int dbgDiggerLinuxQueryAsciiLogBufferPtrs(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
380	RTGCPTR pGCPtrLogBuf, uint32_t pcbLogBuf)
381	{
382	int rc = VINF_SUCCESS;
383
384	/**
385	* We disassemble emit_log_char to get at the log buffer address and size.
386	* This is used in case the symbols are not exported in kallsyms.
387	*
388	* This is what it typically looks like:
389	* vmlinux!emit_log_char:
390	* %00000000c01204a1 56 push esi
391	* %00000000c01204a2 8b 35 d0 1c 34 c0 mov esi, dword [0c0341cd0h]
392	* %00000000c01204a8 53 push ebx
393	* %00000000c01204a9 8b 1d 74 3b 3e c0 mov ebx, dword [0c03e3b74h]
394	* %00000000c01204af 8b 0d d8 1c 34 c0 mov ecx, dword [0c0341cd8h]
395	* %00000000c01204b5 8d 56 ff lea edx, [esi-001h]
396	* %00000000c01204b8 21 da and edx, ebx
397	* %00000000c01204ba 88 04 11 mov byte [ecx+edx], al
398	* %00000000c01204bd 8d 53 01 lea edx, [ebx+001h]
399	* %00000000c01204c0 89 d0 mov eax, edx
400	* [...]
401	*/
402	RTDBGSYMBOL SymInfo;
403	rc = RTDbgModSymbolByName(hMod, "emit_log_char", &SymInfo);
404	if (RT_SUCCESS(rc))
405	{
406	/*
407	* Do the diassembling. Disassemble until a ret instruction is encountered
408	* or a limit is reached (don't want to disassemble for too long as the getter
409	* should be short). Certain instructions found are ignored (push, nop, etc.).
410	*/
411	unsigned cInstrDisassembled = 0;
412	uint32_t offInstr = 0;
413	bool fRet = false;
414	DISSTATE DisState;
415	unsigned cAddressesUsed = 0;
416	struct { size_t cb; RTGCPTR GCPtrOrigSrc; } aAddresses[5];
417	RT_ZERO(DisState);
418	RT_ZERO(aAddresses);
419
420	do
421	{
422	DBGFADDRESS Addr;
423	RTGCPTR GCPtrCur = (RTGCPTR)SymInfo.Value + pThis->AddrKernelBase.FlatPtr + offInstr;
424	DBGFR3AddrFromFlat(pUVM, &Addr, GCPtrCur);
425
426	/* Prefetch the instruction. */
427	uint8_t abInstr[32];
428	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &Addr, &abInstr[0], sizeof(abInstr));
429	if (RT_SUCCESS(rc))
430	{
431	uint32_t cbInstr = 0;
432
433	rc = DISInstr(&abInstr[0], pThis->f64Bit ? DISCPUMODE_64BIT : DISCPUMODE_32BIT, &DisState, &cbInstr);
434	if (RT_SUCCESS(rc))
435	{
436	switch (DisState.pCurInstr->uOpcode)
437	{
438	case OP_PUSH:
439	case OP_POP:
440	case OP_NOP:
441	case OP_LEA:
442	case OP_AND:
443	case OP_CBW:
444	break;
445	case OP_RETN:
446	/* emit_log_char returned, abort disassembling. */
447	rc = VERR_NOT_FOUND;
448	fRet = true;
449	break;
450	case OP_MOV:
451	case OP_MOVSXD:
452	/*
453	* If a mov is encountered writing to memory with al (or dil for amd64) being the source the
454	* character is stored and we can infer the base address and size of the log buffer from
455	* the source addresses.
456	*/
457	if ( (DisState.Param2.fUse & DISUSE_REG_GEN8)
458	&& ( (DisState.Param2.Base.idxGenReg == DISGREG_AL && !pThis->f64Bit)
459	\|\| (DisState.Param2.Base.idxGenReg == DISGREG_DIL && pThis->f64Bit))
460	&& DISUSE_IS_EFFECTIVE_ADDR(DisState.Param1.fUse))
461	{
462	RTGCPTR GCPtrLogBuf = 0;
463	uint32_t cbLogBuf = 0;
464
465	/*
466	* We can stop disassembling now and inspect all registers, look for a valid kernel address first.
467	* Only one of the accessed registers should hold a valid kernel address.
468	* For the log size look for the biggest non kernel address.
469	*/
470	for (unsigned i = 0; i < cAddressesUsed; i++)
471	{
472	DBGFADDRESS AddrVal;
473	union { uint8_t abVal[8]; uint32_t u32Val; uint64_t u64Val; } Val;
474
475	rc = DBGFR3MemRead(pUVM, 0 /idCpu/,
476	DBGFR3AddrFromFlat(pUVM, &AddrVal, aAddresses[i].GCPtrOrigSrc),
477	&Val.abVal[0], aAddresses[i].cb);
478	if (RT_SUCCESS(rc))
479	{
480	if (pThis->f64Bit && aAddresses[i].cb == sizeof(uint64_t))
481	{
482	if (LNX64_VALID_ADDRESS(Val.u64Val))
483	{
484	if (GCPtrLogBuf == 0)
485	GCPtrLogBuf = Val.u64Val;
486	else
487	{
488	rc = VERR_NOT_FOUND;
489	break;
490	}
491	}
492	}
493	else
494	{
495	AssertMsgBreakStmt(aAddresses[i].cb == sizeof(uint32_t),
496	("Invalid value size\n"), rc = VERR_INVALID_STATE);
497
498	/* Might be a kernel address or a size indicator. */
499	if (!pThis->f64Bit && LNX32_VALID_ADDRESS(Val.u32Val))
500	{
501	if (GCPtrLogBuf == 0)
502	GCPtrLogBuf = Val.u32Val;
503	else
504	{
505	rc = VERR_NOT_FOUND;
506	break;
507	}
508	}
509	else
510	{
511	/*
512	* The highest value will be the log buffer because the other
513	* accessed variables are indexes into the buffer and hence
514	* always smaller than the size.
515	*/
516	if (cbLogBuf < Val.u32Val)
517	cbLogBuf = Val.u32Val;
518	}
519	}
520	}
521	}
522
523	if ( RT_SUCCESS(rc)
524	&& GCPtrLogBuf != 0
525	&& cbLogBuf != 0)
526	{
527	*pGCPtrLogBuf = GCPtrLogBuf;
528	*pcbLogBuf = cbLogBuf;
529	}
530	else if (RT_SUCCESS(rc))
531	rc = VERR_NOT_FOUND;
532
533	fRet = true;
534	break;
535	}
536	else
537	{
538	/*
539	* In case of a memory to register move store the destination register index and the
540	* source address in the relation table for later processing.
541	*/
542	if ( (DisState.Param1.fUse & (DISUSE_BASE \| DISUSE_REG_GEN32 \| DISUSE_REG_GEN64))
543	&& (DisState.Param2.cb == sizeof(uint32_t) \|\| DisState.Param2.cb == sizeof(uint64_t))
544	&& (DisState.Param2.fUse & (DISUSE_RIPDISPLACEMENT32\|DISUSE_DISPLACEMENT32\|DISUSE_DISPLACEMENT64)))
545	{
546	RTGCPTR GCPtrVal = 0;
547
548	if (DisState.Param2.fUse & DISUSE_RIPDISPLACEMENT32)
549	GCPtrVal = GCPtrCur + DisState.Param2.uDisp.i32 + cbInstr;
550	else if (DisState.Param2.fUse & DISUSE_DISPLACEMENT32)
551	GCPtrVal = (RTGCPTR)DisState.Param2.uDisp.u32;
552	else if (DisState.Param2.fUse & DISUSE_DISPLACEMENT64)
553	GCPtrVal = (RTGCPTR)DisState.Param2.uDisp.u64;
554	else
555	AssertMsgFailedBreakStmt(("Invalid displacement\n"), rc = VERR_INVALID_STATE);
556
557	if (cAddressesUsed < RT_ELEMENTS(aAddresses))
558	{
559	/* movsxd reads always 32bits. */
560	if (DisState.pCurInstr->uOpcode == OP_MOVSXD)
561	aAddresses[cAddressesUsed].cb = sizeof(uint32_t);
562	else
563	aAddresses[cAddressesUsed].cb = DisState.Param2.cb;
564	aAddresses[cAddressesUsed].GCPtrOrigSrc = GCPtrVal;
565	cAddressesUsed++;
566	}
567	else
568	{
569	rc = VERR_INVALID_PARAMETER;
570	break;
571	}
572	}
573	}
574	break;
575	default:
576	/* All other instructions will cause an error for now (playing safe here). */
577	rc = VERR_INVALID_PARAMETER;
578	break;
579	}
580	cInstrDisassembled++;
581	offInstr += cbInstr;
582	}
583	}
584	} while ( RT_SUCCESS(rc)
585	&& cInstrDisassembled < 20
586	&& !fRet);
587	}
588
589	return rc;
590	}
591
592	/**
593	* Try to get at the log buffer starting address and size by disassembling some exposed helpers.
594	*
595	* @returns VBox status code.
596	* @param pThis The Linux digger data.
597	* @param pUVM The VM handle.
598	* @param hMod The module to use.
599	* @param pGCPtrLogBuf Where to store the log buffer pointer on success.
600	* @param pcbLogBuf Where to store the size of the log buffer on success.
601	*/
602	static int dbgDiggerLinuxQueryLogBufferPtrs(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
603	RTGCPTR pGCPtrLogBuf, uint32_t pcbLogBuf)
604	{
605	int rc = VINF_SUCCESS;
606
607	struct { void pvVar; uint32_t cbHost, cbGuest; const char pszSymbol; } aSymbols[] =
608	{
609	{ pGCPtrLogBuf, (uint32_t)sizeof(RTGCPTR), (uint32_t)(pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t)), "log_buf_addr_get" },
610	{ pcbLogBuf, (uint32_t)sizeof(uint32_t), (uint32_t)sizeof(uint32_t), "log_buf_len_get" }
611	};
612	for (uint32_t i = 0; i < RT_ELEMENTS(aSymbols) && RT_SUCCESS(rc); i++)
613	{
614	RT_BZERO(aSymbols[i].pvVar, aSymbols[i].cbHost);
615	Assert(aSymbols[i].cbHost >= aSymbols[i].cbGuest);
616	rc = dbgDiggerLinuxDisassembleSimpleGetter(pThis, pUVM, hMod, aSymbols[i].pszSymbol,
617	aSymbols[i].pvVar, aSymbols[i].cbGuest);
618	}
619
620	return rc;
621	}
622
623	/**
624	* Returns whether the log buffer is a simple ascii buffer or a record based implementation
625	* based on the kernel version found.
626	*
627	* @returns Flag whether the log buffer is the simple ascii buffer.
628	* @param pThis The Linux digger data.
629	* @param pUVM The user mode VM handle.
630	*/
631	static bool dbgDiggerLinuxLogBufferIsAsciiBuffer(PDBGDIGGERLINUX pThis, PUVM pUVM)
632	{
633	char szTmp[128];
634	char const *pszVer = &szTmp[sizeof(g_abLinuxVersion) - 1];
635
636	RT_ZERO(szTmp);
637	int rc = DBGFR3MemReadString(pUVM, 0, &pThis->AddrLinuxBanner, szTmp, sizeof(szTmp) - 1);
638	if ( RT_SUCCESS(rc)
639	&& RTStrVersionCompare(pszVer, "3.4") == -1)
640	return true;
641
642	return false;
643	}
644
645	/**
646	* Worker to get at the kernel log for pre 3.4 kernels where the log buffer was just a char buffer.
647	*
648	* @returns VBox status code.
649	* @param pThis The Linux digger data.
650	* @param pUVM The VM user mdoe handle.
651	* @param hMod The debug module handle.
652	* @param fFlags Flags reserved for future use, MBZ.
653	* @param cMessages The number of messages to retrieve, counting from the
654	* end of the log (i.e. like tail), use UINT32_MAX for all.
655	* @param pszBuf The output buffer.
656	* @param cbBuf The buffer size.
657	* @param pcbActual Where to store the number of bytes actually returned,
658	* including zero terminator. On VERR_BUFFER_OVERFLOW this
659	* holds the necessary buffer size. Optional.
660	*/
661	static int dbgDiggerLinuxLogBufferQueryAscii(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
662	uint32_t fFlags, uint32_t cMessages,
663	char pszBuf, size_t cbBuf, size_t pcbActual)
664	{
665	RT_NOREF2(fFlags, cMessages);
666	int rc = VINF_SUCCESS;
667	RTGCPTR GCPtrLogBuf;
668	uint32_t cbLogBuf;
669
670	struct { void pvVar; size_t cbHost, cbGuest; const char pszSymbol; } aSymbols[] =
671	{
672	{ &GCPtrLogBuf, sizeof(GCPtrLogBuf), pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t), "log_buf" },
673	{ &cbLogBuf, sizeof(cbLogBuf), sizeof(cbLogBuf), "log_buf_len" },
674	};
675	for (uint32_t i = 0; i < RT_ELEMENTS(aSymbols); i++)
676	{
677	RTDBGSYMBOL SymInfo;
678	rc = RTDbgModSymbolByName(hMod, aSymbols[i].pszSymbol, &SymInfo);
679	if (RT_SUCCESS(rc))
680	{
681	RT_BZERO(aSymbols[i].pvVar, aSymbols[i].cbHost);
682	Assert(aSymbols[i].cbHost >= aSymbols[i].cbGuest);
683	DBGFADDRESS Addr;
684	rc = DBGFR3MemRead(pUVM, 0 /idCpu/,
685	DBGFR3AddrFromFlat(pUVM, &Addr, (RTGCPTR)SymInfo.Value + pThis->AddrKernelBase.FlatPtr),
686	aSymbols[i].pvVar, aSymbols[i].cbGuest);
687	if (RT_SUCCESS(rc))
688	continue;
689	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Reading '%s' at %RGv: %Rrc\n", aSymbols[i].pszSymbol, Addr.FlatPtr, rc));
690	}
691	else
692	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Error looking up '%s': %Rrc\n", aSymbols[i].pszSymbol, rc));
693	rc = VERR_NOT_FOUND;
694	break;
695	}
696
697	/*
698	* Some kernels don't expose the variables in kallsyms so we have to try disassemble
699	* some public helpers to get at the addresses.
700	*
701	* @todo: Maybe cache those values so we don't have to do the heavy work every time?
702	*/
703	if (rc == VERR_NOT_FOUND)
704	{
705	rc = dbgDiggerLinuxQueryAsciiLogBufferPtrs(pThis, pUVM, hMod, &GCPtrLogBuf, &cbLogBuf);
706	if (RT_FAILURE(rc))
707	return rc;
708	}
709
710	/*
711	* Check if the values make sense.
712	*/
713	if (pThis->f64Bit ? !LNX64_VALID_ADDRESS(GCPtrLogBuf) : !LNX32_VALID_ADDRESS(GCPtrLogBuf))
714	{
715	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_buf' value %RGv is not valid.\n", GCPtrLogBuf));
716	return VERR_NOT_FOUND;
717	}
718	if ( cbLogBuf < 4096
719	\|\| !RT_IS_POWER_OF_TWO(cbLogBuf)
720	\|\| cbLogBuf > 16*_1M)
721	{
722	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_buf_len' value %#x is not valid.\n", cbLogBuf));
723	return VERR_NOT_FOUND;
724	}
725
726	/*
727	* Read the whole log buffer.
728	*/
729	uint8_t pbLogBuf = (uint8_t )RTMemAlloc(cbLogBuf);
730	if (!pbLogBuf)
731	{
732	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Failed to allocate %#x bytes for log buffer\n", cbLogBuf));
733	return VERR_NO_MEMORY;
734	}
735	DBGFADDRESS Addr;
736	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, DBGFR3AddrFromFlat(pUVM, &Addr, GCPtrLogBuf), pbLogBuf, cbLogBuf);
737	if (RT_FAILURE(rc))
738	{
739	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Error reading %#x bytes of log buffer at %RGv: %Rrc\n",
740	cbLogBuf, Addr.FlatPtr, rc));
741	RTMemFree(pbLogBuf);
742	return VERR_NOT_FOUND;
743	}
744
745	/** @todo Try to parse where the single messages start to make use of cMessages. */
746	size_t cchLength = RTStrNLen((const char *)pbLogBuf, cbLogBuf);
747	memcpy(&pszBuf[0], pbLogBuf, RT_MIN(cbBuf, cchLength));
748
749	/* Done with the buffer. */
750	RTMemFree(pbLogBuf);
751
752	/* Set return size value. */
753	if (pcbActual)
754	*pcbActual = RT_MIN(cbBuf, cchLength);
755
756	return cbBuf <= cchLength ? VERR_BUFFER_OVERFLOW : VINF_SUCCESS;
757	}
758
759	/**
760	* Worker to get at the kernel log for post 3.4 kernels where the log buffer contains records.
761	*
762	* @returns VBox status code.
763	* @param pThis The Linux digger data.
764	* @param pUVM The VM user mdoe handle.
765	* @param hMod The debug module handle.
766	* @param fFlags Flags reserved for future use, MBZ.
767	* @param cMessages The number of messages to retrieve, counting from the
768	* end of the log (i.e. like tail), use UINT32_MAX for all.
769	* @param pszBuf The output buffer.
770	* @param cbBuf The buffer size.
771	* @param pcbActual Where to store the number of bytes actually returned,
772	* including zero terminator. On VERR_BUFFER_OVERFLOW this
773	* holds the necessary buffer size. Optional.
774	*/
775	static int dbgDiggerLinuxLogBufferQueryRecords(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
776	uint32_t fFlags, uint32_t cMessages,
777	char pszBuf, size_t cbBuf, size_t pcbActual)
778	{
779	RT_NOREF1(fFlags);
780	int rc = VINF_SUCCESS;
781	RTGCPTR GCPtrLogBuf;
782	uint32_t cbLogBuf;
783	uint32_t idxFirst;
784	uint32_t idxNext;
785
786	struct { void pvVar; size_t cbHost, cbGuest; const char pszSymbol; } aSymbols[] =
787	{
788	{ &GCPtrLogBuf, sizeof(GCPtrLogBuf), pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t), "log_buf" },
789	{ &cbLogBuf, sizeof(cbLogBuf), sizeof(cbLogBuf), "log_buf_len" },
790	{ &idxFirst, sizeof(idxFirst), sizeof(idxFirst), "log_first_idx" },
791	{ &idxNext, sizeof(idxNext), sizeof(idxNext), "log_next_idx" },
792	};
793	for (uint32_t i = 0; i < RT_ELEMENTS(aSymbols); i++)
794	{
795	RTDBGSYMBOL SymInfo;
796	rc = RTDbgModSymbolByName(hMod, aSymbols[i].pszSymbol, &SymInfo);
797	if (RT_SUCCESS(rc))
798	{
799	RT_BZERO(aSymbols[i].pvVar, aSymbols[i].cbHost);
800	Assert(aSymbols[i].cbHost >= aSymbols[i].cbGuest);
801	DBGFADDRESS Addr;
802	rc = DBGFR3MemRead(pUVM, 0 /idCpu/,
803	DBGFR3AddrFromFlat(pUVM, &Addr, (RTGCPTR)SymInfo.Value + pThis->AddrKernelBase.FlatPtr),
804	aSymbols[i].pvVar, aSymbols[i].cbGuest);
805	if (RT_SUCCESS(rc))
806	continue;
807	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Reading '%s' at %RGv: %Rrc\n", aSymbols[i].pszSymbol, Addr.FlatPtr, rc));
808	}
809	else
810	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Error looking up '%s': %Rrc\n", aSymbols[i].pszSymbol, rc));
811	rc = VERR_NOT_FOUND;
812	break;
813	}
814
815	/*
816	* Some kernels don't expose the variables in kallsyms so we have to try disassemble
817	* some public helpers to get at the addresses.
818	*
819	* @todo: Maybe cache those values so we don't have to do the heavy work every time?
820	*/
821	if (rc == VERR_NOT_FOUND)
822	{
823	idxFirst = 0;
824	idxNext = 0;
825	rc = dbgDiggerLinuxQueryLogBufferPtrs(pThis, pUVM, hMod, &GCPtrLogBuf, &cbLogBuf);
826	if (RT_FAILURE(rc))
827	return rc;
828	}
829
830	/*
831	* Check if the values make sense.
832	*/
833	if (pThis->f64Bit ? !LNX64_VALID_ADDRESS(GCPtrLogBuf) : !LNX32_VALID_ADDRESS(GCPtrLogBuf))
834	{
835	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_buf' value %RGv is not valid.\n", GCPtrLogBuf));
836	return VERR_NOT_FOUND;
837	}
838	if ( cbLogBuf < 4096
839	\|\| !RT_IS_POWER_OF_TWO(cbLogBuf)
840	\|\| cbLogBuf > 16*_1M)
841	{
842	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_buf_len' value %#x is not valid.\n", cbLogBuf));
843	return VERR_NOT_FOUND;
844	}
845	uint32_t const cbLogAlign = 4;
846	if ( idxFirst > cbLogBuf - sizeof(LNXPRINTKHDR)
847	\|\| (idxFirst & (cbLogAlign - 1)) != 0)
848	{
849	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_first_idx' value %#x is not valid.\n", idxFirst));
850	return VERR_NOT_FOUND;
851	}
852	if ( idxNext > cbLogBuf - sizeof(LNXPRINTKHDR)
853	\|\| (idxNext & (cbLogAlign - 1)) != 0)
854	{
855	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_next_idx' value %#x is not valid.\n", idxNext));
856	return VERR_NOT_FOUND;
857	}
858
859	/*
860	* Read the whole log buffer.
861	*/
862	uint8_t pbLogBuf = (uint8_t )RTMemAlloc(cbLogBuf);
863	if (!pbLogBuf)
864	{
865	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Failed to allocate %#x bytes for log buffer\n", cbLogBuf));
866	return VERR_NO_MEMORY;
867	}
868	DBGFADDRESS Addr;
869	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, DBGFR3AddrFromFlat(pUVM, &Addr, GCPtrLogBuf), pbLogBuf, cbLogBuf);
870	if (RT_FAILURE(rc))
871	{
872	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Error reading %#x bytes of log buffer at %RGv: %Rrc\n",
873	cbLogBuf, Addr.FlatPtr, rc));
874	RTMemFree(pbLogBuf);
875	return VERR_NOT_FOUND;
876	}
877
878	/*
879	* Count the messages in the buffer while doing some basic validation.
880	*/
881	uint32_t const cbUsed = idxFirst == idxNext ? cbLogBuf /* could be empty... */
882	: idxFirst < idxNext ? idxNext - idxFirst : cbLogBuf - idxFirst + idxNext;
883	uint32_t cbLeft = cbUsed;
884	uint32_t offCur = idxFirst;
885	uint32_t cLogMsgs = 0;
886
887	while (cbLeft > 0)
888	{
889	PCLNXPRINTKHDR pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
890	if (!pHdr->cbTotal)
891	{
892	/* Wrap around packet, most likely... */
893	if (cbLogBuf - offCur >= cbLeft)
894	break;
895	offCur = 0;
896	pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
897	}
898	if (RT_UNLIKELY( pHdr->cbTotal > cbLogBuf - sizeof(*pHdr) - offCur
899	\|\| pHdr->cbTotal > cbLeft
900	\|\| (pHdr->cbTotal & (cbLogAlign - 1)) != 0
901	\|\| pHdr->cbTotal < (uint32_t)pHdr->cbText + (uint32_t)pHdr->cbDict + sizeof(*pHdr) ))
902	{
903	LogRel(("dbgDiggerLinuxIDmsg_QueryKernelLog: Invalid printk_log record at %#x: cbTotal=%#x cbText=%#x cbDict=%#x cbLogBuf=%#x cbLeft=%#x\n",
904	offCur, pHdr->cbTotal, pHdr->cbText, pHdr->cbDict, cbLogBuf, cbLeft));
905	rc = VERR_INVALID_STATE;
906	break;
907	}
908
909	if (pHdr->cbText > 0)
910	cLogMsgs++;
911
912	/* next */
913	offCur += pHdr->cbTotal;
914	cbLeft -= pHdr->cbTotal;
915	}
916	if (RT_FAILURE(rc))
917	{
918	RTMemFree(pbLogBuf);
919	return rc;
920	}
921
922	/*
923	* Copy the messages into the output buffer.
924	*/
925	offCur = idxFirst;
926	cbLeft = cbUsed;
927
928	/* Skip messages that the caller doesn't want. */
929	if (cMessages < cLogMsgs)
930	{
931	uint32_t cToSkip = cLogMsgs - cMessages;
932	while (cToSkip > 0)
933	{
934	PCLNXPRINTKHDR pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
935	if (!pHdr->cbTotal)
936	{
937	offCur = 0;
938	pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
939	}
940	if (pHdr->cbText > 0)
941	cToSkip--;
942
943	/* next */
944	offCur += pHdr->cbTotal;
945	cbLeft -= pHdr->cbTotal;
946	}
947	}
948
949	/* Now copy the messages. */
950	size_t offDst = 0;
951	while (cbLeft > 0)
952	{
953	PCLNXPRINTKHDR pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
954	if (!pHdr->cbTotal)
955	{
956	if (cbLogBuf - offCur >= cbLeft)
957	break;
958	offCur = 0;
959	pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
960	}
961
962	if (pHdr->cbText > 0)
963	{
964	char pchText = (char )(pHdr + 1);
965	size_t cchText = RTStrNLen(pchText, pHdr->cbText);
966	if (offDst + cchText < cbBuf)
967	{
968	memcpy(&pszBuf[offDst], pHdr + 1, cchText);
969	pszBuf[offDst + cchText] = '\n';
970	}
971	else if (offDst < cbBuf)
972	memcpy(&pszBuf[offDst], pHdr + 1, cbBuf - offDst);
973	offDst += cchText + 1;
974	}
975
976	/* next */
977	offCur += pHdr->cbTotal;
978	cbLeft -= pHdr->cbTotal;
979	}
980
981	/* Done with the buffer. */
982	RTMemFree(pbLogBuf);
983
984	/* Make sure we've reserved a char for the terminator. */
985	if (!offDst)
986	offDst = 1;
987
988	/* Set return size value. */
989	if (pcbActual)
990	*pcbActual = offDst;
991
992	if (offDst <= cbBuf)
993	return VINF_SUCCESS;
994	return VERR_BUFFER_OVERFLOW;
995	}
996
997	/**
998	* @interface_method_impl{DBGFOSIDMESG,pfnQueryKernelLog}
999	*/
1000	static DECLCALLBACK(int) dbgDiggerLinuxIDmsg_QueryKernelLog(PDBGFOSIDMESG pThis, PUVM pUVM, uint32_t fFlags, uint32_t cMessages,
1001	char pszBuf, size_t cbBuf, size_t pcbActual)
1002	{
1003	PDBGDIGGERLINUX pData = RT_FROM_MEMBER(pThis, DBGDIGGERLINUX, IDmesg);
1004
1005	if (cMessages < 1)
1006	return VERR_INVALID_PARAMETER;
1007
1008	/*
1009	* Resolve the symbols we need and read their values.
1010	*/
1011	RTDBGAS hAs = DBGFR3AsResolveAndRetain(pUVM, DBGF_AS_KERNEL);
1012	RTDBGMOD hMod;
1013	int rc = RTDbgAsModuleByName(hAs, "vmlinux", 0, &hMod);
1014	RTDbgAsRelease(hAs);
1015	if (RT_FAILURE(rc))
1016	return VERR_NOT_FOUND;
1017
1018	/*
1019	* Check whether the kernel log buffer is a simple char buffer or the newer
1020	* record based implementation.
1021	* The record based implementation was presumably introduced with kernel 3.4,
1022	* see: http://thread.gmane.org/gmane.linux.kernel/1284184
1023	*/
1024	size_t cbActual;
1025	if (dbgDiggerLinuxLogBufferIsAsciiBuffer(pData, pUVM))
1026	rc = dbgDiggerLinuxLogBufferQueryAscii(pData, pUVM, hMod, fFlags, cMessages, pszBuf, cbBuf, &cbActual);
1027	else
1028	rc = dbgDiggerLinuxLogBufferQueryRecords(pData, pUVM, hMod, fFlags, cMessages, pszBuf, cbBuf, &cbActual);
1029
1030	/* Release the module in any case. */
1031	RTDbgModRelease(hMod);
1032
1033	if (RT_FAILURE(rc) && rc != VERR_BUFFER_OVERFLOW)
1034	return rc;
1035
1036	if (pcbActual)
1037	*pcbActual = cbActual;
1038
1039	/*
1040	* All VBox strings are UTF-8 and bad things may in theory happen if we
1041	* pass bad UTF-8 to code which assumes it's all valid. So, we enforce
1042	* UTF-8 upon the guest kernel messages here even if they (probably) have
1043	* no defined code set in reality.
1044	*/
1045	if ( RT_SUCCESS(rc)
1046	&& cbActual <= cbBuf)
1047	{
1048	pszBuf[cbActual - 1] = '\0';
1049	RTStrPurgeEncoding(pszBuf);
1050	return VINF_SUCCESS;
1051	}
1052
1053	if (cbBuf)
1054	{
1055	pszBuf[cbBuf - 1] = '\0';
1056	RTStrPurgeEncoding(pszBuf);
1057	}
1058	return VERR_BUFFER_OVERFLOW;
1059	}
1060
1061
1062	/**
1063	* Worker destroying the config database.
1064	*/
1065	static DECLCALLBACK(int) dbgDiggerLinuxCfgDbDestroyWorker(PRTSTRSPACECORE pStr, void *pvUser)
1066	{
1067	PDBGDIGGERLINUXCFGITEM pCfgItem = (PDBGDIGGERLINUXCFGITEM)pStr;
1068	RTStrFree((char *)pCfgItem->Core.pszString);
1069	RTMemFree(pCfgItem);
1070	NOREF(pvUser);
1071	return 0;
1072	}
1073
1074
1075	/**
1076	* Destroy the config database.
1077	*
1078	* @returns nothing.
1079	* @param pThis The Linux digger data.
1080	*/
1081	static void dbgDiggerLinuxCfgDbDestroy(PDBGDIGGERLINUX pThis)
1082	{
1083	RTStrSpaceDestroy(&pThis->hCfgDb, dbgDiggerLinuxCfgDbDestroyWorker, NULL);
1084	}
1085
1086
1087	/**
1088	* @copydoc DBGFOSREG::pfnStackUnwindAssist
1089	*/
1090	static DECLCALLBACK(int) dbgDiggerLinuxStackUnwindAssist(PUVM pUVM, void *pvData, VMCPUID idCpu, PDBGFSTACKFRAME pFrame,
1091	PRTDBGUNWINDSTATE pState, PCCPUMCTX pInitialCtx, RTDBGAS hAs,
1092	uint64_t *puScratch)
1093	{
1094	RT_NOREF(pUVM, pvData, idCpu, pFrame, pState, pInitialCtx, hAs, puScratch);
1095	return VINF_SUCCESS;
1096	}
1097
1098
1099	/**
1100	* @copydoc DBGFOSREG::pfnQueryInterface
1101	*/
1102	static DECLCALLBACK(void ) dbgDiggerLinuxQueryInterface(PUVM pUVM, void pvData, DBGFOSINTERFACE enmIf)
1103	{
1104	RT_NOREF1(pUVM);
1105	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
1106	switch (enmIf)
1107	{
1108	case DBGFOSINTERFACE_DMESG:
1109	return &pThis->IDmesg;
1110
1111	default:
1112	return NULL;
1113	}
1114	}
1115
1116
1117	/**
1118	* @copydoc DBGFOSREG::pfnQueryVersion
1119	*/
1120	static DECLCALLBACK(int) dbgDiggerLinuxQueryVersion(PUVM pUVM, void pvData, char pszVersion, size_t cchVersion)
1121	{
1122	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
1123	Assert(pThis->fValid);
1124
1125	/*
1126	* It's all in the linux banner.
1127	*/
1128	int rc = DBGFR3MemReadString(pUVM, 0, &pThis->AddrLinuxBanner, pszVersion, cchVersion);
1129	if (RT_SUCCESS(rc))
1130	{
1131	char *pszEnd = RTStrEnd(pszVersion, cchVersion);
1132	AssertReturn(pszEnd, VERR_BUFFER_OVERFLOW);
1133	while ( pszEnd > pszVersion
1134	&& RT_C_IS_SPACE(pszEnd[-1]))
1135	pszEnd--;
1136	*pszEnd = '\0';
1137	}
1138	else
1139	RTStrPrintf(pszVersion, cchVersion, "DBGFR3MemRead -> %Rrc", rc);
1140
1141	return rc;
1142	}
1143
1144
1145	/**
1146	* @copydoc DBGFOSREG::pfnTerm
1147	*/
1148	static DECLCALLBACK(void) dbgDiggerLinuxTerm(PUVM pUVM, void *pvData)
1149	{
1150	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
1151	Assert(pThis->fValid);
1152
1153	/*
1154	* Destroy configuration database.
1155	*/
1156	dbgDiggerLinuxCfgDbDestroy(pThis);
1157
1158	/*
1159	* Unlink and release our modules.
1160	*/
1161	RTDBGAS hDbgAs = DBGFR3AsResolveAndRetain(pUVM, DBGF_AS_KERNEL);
1162	if (hDbgAs != NIL_RTDBGAS)
1163	{
1164	uint32_t iMod = RTDbgAsModuleCount(hDbgAs);
1165	while (iMod-- > 0)
1166	{
1167	RTDBGMOD hMod = RTDbgAsModuleByIndex(hDbgAs, iMod);
1168	if (hMod != NIL_RTDBGMOD)
1169	{
1170	if (RTDbgModGetTag(hMod) == DIG_LNX_MOD_TAG)
1171	{
1172	int rc = RTDbgAsModuleUnlink(hDbgAs, hMod);
1173	AssertRC(rc);
1174	}
1175	RTDbgModRelease(hMod);
1176	}
1177	}
1178	RTDbgAsRelease(hDbgAs);
1179	}
1180
1181	pThis->fValid = false;
1182	}
1183
1184
1185	/**
1186	* @copydoc DBGFOSREG::pfnRefresh
1187	*/
1188	static DECLCALLBACK(int) dbgDiggerLinuxRefresh(PUVM pUVM, void *pvData)
1189	{
1190	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
1191	NOREF(pThis);
1192	Assert(pThis->fValid);
1193
1194	/*
1195	* For now we'll flush and reload everything.
1196	*/
1197	dbgDiggerLinuxTerm(pUVM, pvData);
1198	return dbgDiggerLinuxInit(pUVM, pvData);
1199	}
1200
1201
1202	/**
1203	* Worker for dbgDiggerLinuxFindStartOfNamesAndSymbolCount that update the
1204	* digger data.
1205	*
1206	* @returns VINF_SUCCESS.
1207	* @param pThis The Linux digger data to update.
1208	* @param pAddrKernelNames The kallsyms_names address.
1209	* @param cKernelSymbols The number of kernel symbol.
1210	* @param cbAddress The guest address size.
1211	*/
1212	static int dbgDiggerLinuxFoundStartOfNames(PDBGDIGGERLINUX pThis, PCDBGFADDRESS pAddrKernelNames,
1213	uint32_t cKernelSymbols, uint32_t cbAddress)
1214	{
1215	pThis->cKernelSymbols = cKernelSymbols;
1216	pThis->AddrKernelNames = *pAddrKernelNames;
1217	pThis->AddrKernelAddresses = *pAddrKernelNames;
1218	uint32_t cbSymbolsSkip = (pThis->fRelKrnlAddr ? 2 : 1) * cbAddress; /* Relative addressing introduces kallsyms_relative_base. */
1219	uint32_t cbOffsets = pThis->fRelKrnlAddr ? sizeof(int32_t) : cbAddress; /* Offsets are always 32bits wide for relative addressing. */
1220	uint32_t cbAlign = 0;
1221
1222	/*
1223	* If the number of symbols is odd there is padding to align the following guest pointer
1224	* sized data properly on 64bit systems with relative addressing.
1225	*/
1226	if ( pThis->fRelKrnlAddr
1227	&& pThis->f64Bit
1228	&& (pThis->cKernelSymbols & 1))
1229	cbAlign = sizeof(int32_t);
1230	DBGFR3AddrSub(&pThis->AddrKernelAddresses, cKernelSymbols * cbOffsets + cbSymbolsSkip + cbAlign);
1231
1232	Log(("dbgDiggerLinuxFoundStartOfNames: AddrKernelAddresses=%RGv\n"
1233	"dbgDiggerLinuxFoundStartOfNames: cKernelSymbols=%#x (at %RGv)\n"
1234	"dbgDiggerLinuxFoundStartOfNames: AddrKernelName=%RGv\n",
1235	pThis->AddrKernelAddresses.FlatPtr,
1236	pThis->cKernelSymbols, pThis->AddrKernelNames.FlatPtr - cbAddress,
1237	pThis->AddrKernelNames.FlatPtr));
1238	return VINF_SUCCESS;
1239	}
1240
1241
1242	/**
1243	* Tries to find the address of the kallsyms_names, kallsyms_num_syms and
1244	* kallsyms_addresses symbols.
1245	*
1246	* The kallsyms_num_syms is read and stored in pThis->cKernelSymbols, while the
1247	* addresses of the other two are stored as pThis->AddrKernelNames and
1248	* pThis->AddrKernelAddresses.
1249	*
1250	* @returns VBox status code, success indicating that all three variables have
1251	* been found and taken down.
1252	* @param pUVM The user mode VM handle.
1253	* @param pThis The Linux digger data.
1254	* @param pHitAddr An address we think is inside kallsyms_names.
1255	*/
1256	static int dbgDiggerLinuxFindStartOfNamesAndSymbolCount(PUVM pUVM, PDBGDIGGERLINUX pThis, PCDBGFADDRESS pHitAddr)
1257	{
1258	/*
1259	* Search backwards in chunks.
1260	*/
1261	union
1262	{
1263	uint8_t ab[0x1000];
1264	uint32_t au32[0x1000 / sizeof(uint32_t)];
1265	uint64_t au64[0x1000 / sizeof(uint64_t)];
1266	} uBuf;
1267	uint32_t cbLeft = LNX_MAX_KALLSYMS_NAMES_SIZE;
1268	uint32_t cbBuf = pHitAddr->FlatPtr & (sizeof(uBuf) - 1);
1269	DBGFADDRESS CurAddr = *pHitAddr;
1270	DBGFR3AddrSub(&CurAddr, cbBuf);
1271	cbBuf += sizeof(uint64_t) - 1; /* In case our kobj hit is in the first 4/8 bytes. */
1272	for (;;)
1273	{
1274	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &CurAddr, &uBuf, sizeof(uBuf));
1275	if (RT_FAILURE(rc))
1276	return rc;
1277
1278	/*
1279	* Since Linux 4.6 there are two different methods to store the kallsyms addresses
1280	* in the image.
1281	*
1282	* The first and longer existing method is to store the absolute addresses in an
1283	* array starting at kallsyms_addresses followed by a field which stores the number
1284	* of kernel symbols called kallsyms_num_syms.
1285	* The newer method is to use offsets stored in kallsyms_offsets and have a base pointer
1286	* to relate the offsets to called kallsyms_relative_base. One entry in kallsyms_offsets is
1287	* always 32bit wide regardless of the guest pointer size (this halves the table on 64bit
1288	* systems) but means more work for us for the 64bit case.
1289	*
1290	* When absolute addresses are used the following assumptions hold:
1291	*
1292	* We assume that the three symbols are aligned on guest pointer boundary.
1293	*
1294	* The boundary between the two tables should be noticable as the number
1295	* is unlikely to be more than 16 millions, there will be at least one zero
1296	* byte where it is, 64-bit will have 5 zero bytes. Zero bytes aren't all
1297	* that common in the kallsyms_names table.
1298	*
1299	* Also the kallsyms_names table starts with a length byte, which means
1300	* we're likely to see a byte in the range 1..31.
1301	*
1302	* The kallsyms_addresses are mostly sorted (except for the start where the
1303	* absolute symbols are), so we'll spot a bunch of kernel addresses
1304	* immediately preceeding the kallsyms_num_syms field.
1305	*
1306	* Lazy bird: If kallsyms_num_syms is on a buffer boundrary, we skip
1307	* the check for kernel addresses preceeding it.
1308	*
1309	* For relative offsets most of the assumptions from above are true too
1310	* except that we have to distinguish between the relative base address and the offsets.
1311	* Every observed kernel has a valid kernel address fo the relative base and kallsyms_relative_base
1312	* always comes before kallsyms_num_syms and is aligned on a guest pointer boundary.
1313	* Offsets are stored before kallsyms_relative_base and don't contain valid kernel addresses.
1314	*
1315	* To distinguish between absolute and relative offsetting we check the data before a candidate
1316	* for kallsyms_num_syms. If all entries before the kallsyms_num_syms candidate are valid kernel
1317	* addresses absolute addresses are assumed. If this is not the case but the first entry before
1318	* kallsyms_num_syms is a valid kernel address we check whether the data before and the possible
1319	* relative base form a valid kernel address and assume relative offsets.
1320	*/
1321	if (pThis->f64Bit)
1322	{
1323	uint32_t i = cbBuf / sizeof(uint64_t);
1324	while (i-- > 0)
1325	if ( uBuf.au64[i] <= LNX_MAX_KALLSYMS_SYMBOLS
1326	&& uBuf.au64[i] >= LNX_MIN_KALLSYMS_SYMBOLS)
1327	{
1328	uint8_t pb = (uint8_t )&uBuf.au64[i + 1];
1329	if ( pb[0] <= LNX_MAX_KALLSYMS_ENC_LENGTH
1330	&& pb[0] >= LNX_MIN_KALLSYMS_ENC_LENGTH)
1331	{
1332	/*
1333	* Check whether we have a valid kernel address and try to distinguish
1334	* whether the kernel uses relative offsetting or absolute addresses.
1335	*/
1336	if ( (i >= 1 && LNX64_VALID_ADDRESS(uBuf.au64[i - 1]))
1337	&& (i >= 2 && !LNX64_VALID_ADDRESS(uBuf.au64[i - 2]))
1338	&& (i >= 3 && !LNX64_VALID_ADDRESS(uBuf.au64[i - 3])))
1339	{
1340	RTGCUINTPTR uKrnlRelBase = uBuf.au64[i - 1];
1341	DBGFADDRESS RelAddr = CurAddr;
1342	int32_t aiRelOff[3];
1343	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, DBGFR3AddrAdd(&RelAddr, (i - 1) * sizeof(uint64_t) - sizeof(aiRelOff)),
1344	&aiRelOff[0], sizeof(aiRelOff));
1345	if ( RT_SUCCESS(rc)
1346	&& LNX64_VALID_ADDRESS(uKrnlRelBase + aiRelOff[0])
1347	&& LNX64_VALID_ADDRESS(uKrnlRelBase + aiRelOff[1])
1348	&& LNX64_VALID_ADDRESS(uKrnlRelBase + aiRelOff[2]))
1349	{
1350	Log(("dbgDiggerLinuxFindStartOfNamesAndSymbolCount: relative base %RGv (at %RGv)\n",
1351	uKrnlRelBase, CurAddr.FlatPtr + (i - 1) * sizeof(uint64_t)));
1352	pThis->fRelKrnlAddr = true;
1353	pThis->uKernelRelativeBase = uKrnlRelBase;
1354	return dbgDiggerLinuxFoundStartOfNames(pThis,
1355	DBGFR3AddrAdd(&CurAddr, (i + 1) * sizeof(uint64_t)),
1356	(uint32_t)uBuf.au64[i], sizeof(uint64_t));
1357	}
1358	}
1359
1360	if ( (i <= 0 \|\| LNX64_VALID_ADDRESS(uBuf.au64[i - 1]))
1361	&& (i <= 1 \|\| LNX64_VALID_ADDRESS(uBuf.au64[i - 2]))
1362	&& (i <= 2 \|\| LNX64_VALID_ADDRESS(uBuf.au64[i - 3])))
1363	return dbgDiggerLinuxFoundStartOfNames(pThis,
1364	DBGFR3AddrAdd(&CurAddr, (i + 1) * sizeof(uint64_t)),
1365	(uint32_t)uBuf.au64[i], sizeof(uint64_t));
1366	}
1367	}
1368	}
1369	else
1370	{
1371	uint32_t i = cbBuf / sizeof(uint32_t);
1372	while (i-- > 0)
1373	if ( uBuf.au32[i] <= LNX_MAX_KALLSYMS_SYMBOLS
1374	&& uBuf.au32[i] >= LNX_MIN_KALLSYMS_SYMBOLS)
1375	{
1376	uint8_t pb = (uint8_t )&uBuf.au32[i + 1];
1377	if ( pb[0] <= LNX_MAX_KALLSYMS_ENC_LENGTH
1378	&& pb[0] >= LNX_MIN_KALLSYMS_ENC_LENGTH)
1379	{
1380	/* Check for relative base addressing. */
1381	if (i >= 1 && LNX32_VALID_ADDRESS(uBuf.au32[i - 1]))
1382	{
1383	RTGCUINTPTR uKrnlRelBase = uBuf.au32[i - 1];
1384	if ( (i <= 1 \|\| LNX32_VALID_ADDRESS(uKrnlRelBase + uBuf.au32[i - 2]))
1385	&& (i <= 2 \|\| LNX32_VALID_ADDRESS(uKrnlRelBase + uBuf.au32[i - 3])))
1386	{
1387	Log(("dbgDiggerLinuxFindStartOfNamesAndSymbolCount: relative base %RGv (at %RGv)\n",
1388	uKrnlRelBase, CurAddr.FlatPtr + (i - 1) * sizeof(uint32_t)));
1389	pThis->fRelKrnlAddr = true;
1390	pThis->uKernelRelativeBase = uKrnlRelBase;
1391	return dbgDiggerLinuxFoundStartOfNames(pThis,
1392	DBGFR3AddrAdd(&CurAddr, (i + 1) * sizeof(uint32_t)),
1393	uBuf.au32[i], sizeof(uint32_t));
1394	}
1395	}
1396
1397	if ( (i <= 0 \|\| LNX32_VALID_ADDRESS(uBuf.au32[i - 1]))
1398	&& (i <= 1 \|\| LNX32_VALID_ADDRESS(uBuf.au32[i - 2]))
1399	&& (i <= 2 \|\| LNX32_VALID_ADDRESS(uBuf.au32[i - 3])))
1400	return dbgDiggerLinuxFoundStartOfNames(pThis,
1401	DBGFR3AddrAdd(&CurAddr, (i + 1) * sizeof(uint32_t)),
1402	uBuf.au32[i], sizeof(uint32_t));
1403	}
1404	}
1405	}
1406
1407	/*
1408	* Advance
1409	*/
1410	if (RT_UNLIKELY(cbLeft <= sizeof(uBuf)))
1411	{
1412	Log(("dbgDiggerLinuxFindStartOfNamesAndSymbolCount: failed (pHitAddr=%RGv)\n", pHitAddr->FlatPtr));
1413	return VERR_NOT_FOUND;
1414	}
1415	cbLeft -= sizeof(uBuf);
1416	DBGFR3AddrSub(&CurAddr, sizeof(uBuf));
1417	cbBuf = sizeof(uBuf);
1418	}
1419	}
1420
1421
1422	/**
1423	* Worker for dbgDiggerLinuxFindEndNames that records the findings.
1424	*
1425	* @returns VINF_SUCCESS
1426	* @param pThis The linux digger data to update.
1427	* @param pAddrMarkers The address of the marker (kallsyms_markers).
1428	* @param cbMarkerEntry The size of a marker entry (32-bit or 64-bit).
1429	*/
1430	static int dbgDiggerLinuxFoundMarkers(PDBGDIGGERLINUX pThis, PCDBGFADDRESS pAddrMarkers, uint32_t cbMarkerEntry)
1431	{
1432	pThis->cbKernelNames = pAddrMarkers->FlatPtr - pThis->AddrKernelNames.FlatPtr;
1433	pThis->AddrKernelNameMarkers = *pAddrMarkers;
1434	pThis->cKernelNameMarkers = RT_ALIGN_32(pThis->cKernelSymbols, 256) / 256;
1435	pThis->AddrKernelTokenTable = *pAddrMarkers;
1436	DBGFR3AddrAdd(&pThis->AddrKernelTokenTable, pThis->cKernelNameMarkers * cbMarkerEntry);
1437
1438	Log(("dbgDiggerLinuxFoundMarkers: AddrKernelNames=%RGv cbKernelNames=%#x\n"
1439	"dbgDiggerLinuxFoundMarkers: AddrKernelNameMarkers=%RGv cKernelNameMarkers=%#x\n"
1440	"dbgDiggerLinuxFoundMarkers: AddrKernelTokenTable=%RGv\n",
1441	pThis->AddrKernelNames.FlatPtr, pThis->cbKernelNames,
1442	pThis->AddrKernelNameMarkers.FlatPtr, pThis->cKernelNameMarkers,
1443	pThis->AddrKernelTokenTable.FlatPtr));
1444	return VINF_SUCCESS;
1445	}
1446
1447
1448	/**
1449	* Tries to find the end of kallsyms_names and thereby the start of
1450	* kallsyms_markers and kallsyms_token_table.
1451	*
1452	* The kallsyms_names size is stored in pThis->cbKernelNames, the addresses of
1453	* the two other symbols in pThis->AddrKernelNameMarkers and
1454	* pThis->AddrKernelTokenTable. The number of marker entries is stored in
1455	* pThis->cKernelNameMarkers.
1456	*
1457	* @returns VBox status code, success indicating that all three variables have
1458	* been found and taken down.
1459	* @param pUVM The user mode VM handle.
1460	* @param pThis The Linux digger data.
1461	* @param pHitAddr An address we think is inside kallsyms_names.
1462	*/
1463	static int dbgDiggerLinuxFindEndOfNamesAndMore(PUVM pUVM, PDBGDIGGERLINUX pThis, PCDBGFADDRESS pHitAddr)
1464	{
1465	/*
1466	* Search forward in chunks.
1467	*/
1468	union
1469	{
1470	uint8_t ab[0x1000];
1471	uint32_t au32[0x1000 / sizeof(uint32_t)];
1472	uint64_t au64[0x1000 / sizeof(uint64_t)];
1473	} uBuf;
1474	bool fPendingZeroHit = false;
1475	uint32_t cbLeft = LNX_MAX_KALLSYMS_NAMES_SIZE + sizeof(uBuf);
1476	uint32_t offBuf = pHitAddr->FlatPtr & (sizeof(uBuf) - 1);
1477	DBGFADDRESS CurAddr = *pHitAddr;
1478	DBGFR3AddrSub(&CurAddr, offBuf);
1479	for (;;)
1480	{
1481	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &CurAddr, &uBuf, sizeof(uBuf));
1482	if (RT_FAILURE(rc))
1483	return rc;
1484
1485	/*
1486	* The kallsyms_names table is followed by kallsyms_markers we assume,
1487	* using sizeof(unsigned long) alignment like the preceeding symbols.
1488	*
1489	* The kallsyms_markers table has entried sizeof(unsigned long) and
1490	* contains offsets into kallsyms_names. The kallsyms_markers used to
1491	* index kallsyms_names and reduce seek time when looking up the name
1492	* of an address/symbol. Each entry in kallsyms_markers covers 256
1493	* symbol names.
1494	*
1495	* Because of this, the first entry is always zero and all the entries
1496	* are ascending. It also follows that the size of the table can be
1497	* calculated from kallsyms_num_syms.
1498	*
1499	* Note! We could also have walked kallsyms_names by skipping
1500	* kallsyms_num_syms names, but this is faster and we will
1501	* validate the encoded names later.
1502	*/
1503	if (pThis->f64Bit)
1504	{
1505	if ( RT_UNLIKELY(fPendingZeroHit)
1506	&& uBuf.au64[0] >= (LNX_MIN_KALLSYMS_ENC_LENGTH + 1) * 256
1507	&& uBuf.au64[0] <= (LNX_MAX_KALLSYMS_ENC_LENGTH + 1) * 256)
1508	return dbgDiggerLinuxFoundMarkers(pThis, DBGFR3AddrSub(&CurAddr, sizeof(uint64_t)), sizeof(uint64_t));
1509
1510	uint32_t const cEntries = sizeof(uBuf) / sizeof(uint64_t);
1511	for (uint32_t i = offBuf / sizeof(uint64_t); i < cEntries; i++)
1512	if (uBuf.au64[i] == 0)
1513	{
1514	if (RT_UNLIKELY(i + 1 >= cEntries))
1515	{
1516	fPendingZeroHit = true;
1517	break;
1518	}
1519	if ( uBuf.au64[i + 1] >= (LNX_MIN_KALLSYMS_ENC_LENGTH + 1) * 256
1520	&& uBuf.au64[i + 1] <= (LNX_MAX_KALLSYMS_ENC_LENGTH + 1) * 256)
1521	return dbgDiggerLinuxFoundMarkers(pThis, DBGFR3AddrAdd(&CurAddr, i * sizeof(uint64_t)), sizeof(uint64_t));
1522	}
1523	}
1524	else
1525	{
1526	if ( RT_UNLIKELY(fPendingZeroHit)
1527	&& uBuf.au32[0] >= (LNX_MIN_KALLSYMS_ENC_LENGTH + 1) * 256
1528	&& uBuf.au32[0] <= (LNX_MAX_KALLSYMS_ENC_LENGTH + 1) * 256)
1529	return dbgDiggerLinuxFoundMarkers(pThis, DBGFR3AddrSub(&CurAddr, sizeof(uint32_t)), sizeof(uint32_t));
1530
1531	uint32_t const cEntries = sizeof(uBuf) / sizeof(uint32_t);
1532	for (uint32_t i = offBuf / sizeof(uint32_t); i < cEntries; i++)
1533	if (uBuf.au32[i] == 0)
1534	{
1535	if (RT_UNLIKELY(i + 1 >= cEntries))
1536	{
1537	fPendingZeroHit = true;
1538	break;
1539	}
1540	if ( uBuf.au32[i + 1] >= (LNX_MIN_KALLSYMS_ENC_LENGTH + 1) * 256
1541	&& uBuf.au32[i + 1] <= (LNX_MAX_KALLSYMS_ENC_LENGTH + 1) * 256)
1542	return dbgDiggerLinuxFoundMarkers(pThis, DBGFR3AddrAdd(&CurAddr, i * sizeof(uint32_t)), sizeof(uint32_t));
1543	}
1544	}
1545
1546	/*
1547	* Advance
1548	*/
1549	if (RT_UNLIKELY(cbLeft <= sizeof(uBuf)))
1550	{
1551	Log(("dbgDiggerLinuxFindEndOfNamesAndMore: failed (pHitAddr=%RGv)\n", pHitAddr->FlatPtr));
1552	return VERR_NOT_FOUND;
1553	}
1554	cbLeft -= sizeof(uBuf);
1555	DBGFR3AddrAdd(&CurAddr, sizeof(uBuf));
1556	offBuf = 0;
1557	}
1558	}
1559
1560
1561	/**
1562	* Locates the kallsyms_token_index table.
1563	*
1564	* Storing the address in pThis->AddrKernelTokenIndex and the size of the token
1565	* table in pThis->cbKernelTokenTable.
1566	*
1567	* @returns VBox status code.
1568	* @param pUVM The user mode VM handle.
1569	* @param pThis The Linux digger data.
1570	*/
1571	static int dbgDiggerLinuxFindTokenIndex(PUVM pUVM, PDBGDIGGERLINUX pThis)
1572	{
1573	/*
1574	* The kallsyms_token_table is very much like a string table. Due to the
1575	* nature of the compression algorithm it is reasonably short (one example
1576	* here is 853 bytes), so we'll not be reading it in chunks but in full.
1577	* To be on the safe side, we read 8KB, ASSUMING we won't run into unmapped
1578	* memory or any other nasty stuff...
1579	*/
1580	union
1581	{
1582	uint8_t ab[0x2000];
1583	uint16_t au16[0x2000 / sizeof(uint16_t)];
1584	} uBuf;
1585	DBGFADDRESS CurAddr = pThis->AddrKernelTokenTable;
1586	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &CurAddr, &uBuf, sizeof(uBuf));
1587	if (RT_FAILURE(rc))
1588	return rc;
1589
1590	/*
1591	* We've got two choices here, either walk the string table or look for
1592	* the next structure, kallsyms_token_index.
1593	*
1594	* The token index is a table of 256 uint16_t entries (index by bytes
1595	* from kallsyms_names) that gives offsets in kallsyms_token_table. It
1596	* starts with a zero entry and the following entries are sorted in
1597	* ascending order. The range of the entries are reasonably small since
1598	* kallsyms_token_table is small.
1599	*
1600	* The alignment seems to be sizeof(unsigned long), just like
1601	* kallsyms_token_table.
1602	*
1603	* So, we start by looking for a zero 16-bit entry.
1604	*/
1605	uint32_t cIncr = (pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t)) / sizeof(uint16_t);
1606
1607	for (uint32_t i = 0; i < sizeof(uBuf) / sizeof(uint16_t) - 16; i += cIncr)
1608	if ( uBuf.au16[i] == 0
1609	&& uBuf.au16[i + 1] > 0
1610	&& uBuf.au16[i + 1] <= LNX_MAX_KALLSYMS_TOKEN_LEN
1611	&& (uint16_t)(uBuf.au16[i + 2] - uBuf.au16[i + 1] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1612	&& (uint16_t)(uBuf.au16[i + 3] - uBuf.au16[i + 2] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1613	&& (uint16_t)(uBuf.au16[i + 4] - uBuf.au16[i + 3] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1614	&& (uint16_t)(uBuf.au16[i + 5] - uBuf.au16[i + 4] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1615	&& (uint16_t)(uBuf.au16[i + 6] - uBuf.au16[i + 5] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1616	)
1617	{
1618	pThis->AddrKernelTokenIndex = CurAddr;
1619	DBGFR3AddrAdd(&pThis->AddrKernelTokenIndex, i * sizeof(uint16_t));
1620	pThis->cbKernelTokenTable = i * sizeof(uint16_t);
1621	return VINF_SUCCESS;
1622	}
1623
1624	Log(("dbgDiggerLinuxFindTokenIndex: Failed (%RGv..%RGv)\n", CurAddr.FlatPtr, CurAddr.FlatPtr + (RTGCUINTPTR)sizeof(uBuf)));
1625	return VERR_NOT_FOUND;
1626	}
1627
1628
1629	/**
1630	* Loads the kernel symbols from the given kallsyms offset table decoding the symbol names
1631	* (worker common for dbgDiggerLinuxLoadKernelSymbolsAbsolute() and dbgDiggerLinuxLoadKernelSymbolsRelative()).
1632	*
1633	* @returns VBox status code.
1634	* @param pUVM The user mode VM handle.
1635	* @param pThis The Linux digger data.
1636	* @param uKernelStart Flat kernel start address.
1637	* @param cbKernel Size of the kernel in bytes.
1638	* @param pauSymOff Pointer to the array of symbol offsets in the kallsyms table
1639	* relative to the start of the kernel.
1640	*/
1641	static int dbgDiggerLinuxLoadKernelSymbolsWorker(PUVM pUVM, PDBGDIGGERLINUX pThis, RTGCUINTPTR uKernelStart,
1642	RTGCUINTPTR cbKernel, RTGCUINTPTR *pauSymOff)
1643	{
1644	uint8_t pbNames = (uint8_t )RTMemAllocZ(pThis->cbKernelNames);
1645	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &pThis->AddrKernelNames, pbNames, pThis->cbKernelNames);
1646	if (RT_SUCCESS(rc))
1647	{
1648	char pszzTokens = (char )RTMemAllocZ(pThis->cbKernelTokenTable);
1649	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &pThis->AddrKernelTokenTable, pszzTokens, pThis->cbKernelTokenTable);
1650	if (RT_SUCCESS(rc))
1651	{
1652	uint16_t paoffTokens = (uint16_t )RTMemAllocZ(256 * sizeof(uint16_t));
1653	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &pThis->AddrKernelTokenIndex, paoffTokens, 256 * sizeof(uint16_t));
1654	if (RT_SUCCESS(rc))
1655	{
1656	/*
1657	* Create a module for the kernel.
1658	*/
1659	RTDBGMOD hMod;
1660	rc = RTDbgModCreate(&hMod, "vmlinux", cbKernel, 0 /fFlags/);
1661	if (RT_SUCCESS(rc))
1662	{
1663	rc = RTDbgModSetTag(hMod, DIG_LNX_MOD_TAG); AssertRC(rc);
1664	rc = VINF_SUCCESS;
1665
1666	/*
1667	* Enumerate the symbols.
1668	*/
1669	uint32_t offName = 0;
1670	uint32_t cLeft = pThis->cKernelSymbols;
1671	while (cLeft-- > 0 && RT_SUCCESS(rc))
1672	{
1673	/* Decode the symbol name first. */
1674	if (RT_LIKELY(offName < pThis->cbKernelNames))
1675	{
1676	uint8_t cbName = pbNames[offName++];
1677	if (RT_LIKELY(offName + cbName <= pThis->cbKernelNames))
1678	{
1679	char szSymbol[4096];
1680	uint32_t offSymbol = 0;
1681	while (cbName-- > 0)
1682	{
1683	uint8_t bEnc = pbNames[offName++];
1684	uint16_t offToken = paoffTokens[bEnc];
1685	if (RT_LIKELY(offToken < pThis->cbKernelTokenTable))
1686	{
1687	const char *pszToken = &pszzTokens[offToken];
1688	char ch;
1689	while ((ch = *pszToken++) != '\0')
1690	if (offSymbol < sizeof(szSymbol) - 1)
1691	szSymbol[offSymbol++] = ch;
1692	}
1693	else
1694	{
1695	rc = VERR_INVALID_UTF8_ENCODING;
1696	break;
1697	}
1698	}
1699	szSymbol[offSymbol < sizeof(szSymbol) ? offSymbol : sizeof(szSymbol) - 1] = '\0';
1700
1701	/* The offset. */
1702	RTGCUINTPTR uSymOff = *pauSymOff;
1703	pauSymOff++;
1704
1705	/* Add it without the type char. */
1706	if (uSymOff <= cbKernel)
1707	{
1708	rc = RTDbgModSymbolAdd(hMod, &szSymbol[1], RTDBGSEGIDX_RVA, uSymOff,
1709	0 /cb/, 0 /fFlags/, NULL);
1710	if (RT_FAILURE(rc))
1711	{
1712	if ( rc == VERR_DBG_SYMBOL_NAME_OUT_OF_RANGE
1713	\|\| rc == VERR_DBG_INVALID_RVA
1714	\|\| rc == VERR_DBG_ADDRESS_CONFLICT
1715	\|\| rc == VERR_DBG_DUPLICATE_SYMBOL)
1716	{
1717	Log2(("dbgDiggerLinuxLoadKernelSymbols: RTDbgModSymbolAdd(,%s,) failed %Rrc (ignored)\n", szSymbol, rc));
1718	rc = VINF_SUCCESS;
1719	}
1720	else
1721	Log(("dbgDiggerLinuxLoadKernelSymbols: RTDbgModSymbolAdd(,%s,) failed %Rrc\n", szSymbol, rc));
1722	}
1723	}
1724	}
1725	else
1726	{
1727	rc = VERR_END_OF_STRING;
1728	Log(("dbgDiggerLinuxLoadKernelSymbols: offName=%#x cLeft=%#x cbName=%#x cbKernelNames=%#x\n",
1729	offName, cLeft, cbName, pThis->cbKernelNames));
1730	}
1731	}
1732	else
1733	{
1734	rc = VERR_END_OF_STRING;
1735	Log(("dbgDiggerLinuxLoadKernelSymbols: offName=%#x cLeft=%#x cbKernelNames=%#x\n",
1736	offName, cLeft, pThis->cbKernelNames));
1737	}
1738	}
1739
1740	/*
1741	* Link the module into the address space.
1742	*/
1743	if (RT_SUCCESS(rc))
1744	{
1745	RTDBGAS hAs = DBGFR3AsResolveAndRetain(pUVM, DBGF_AS_KERNEL);
1746	if (hAs != NIL_RTDBGAS)
1747	rc = RTDbgAsModuleLink(hAs, hMod, uKernelStart, RTDBGASLINK_FLAGS_REPLACE);
1748	else
1749	rc = VERR_INTERNAL_ERROR;
1750	RTDbgAsRelease(hAs);
1751	}
1752	else
1753	Log(("dbgDiggerLinuxLoadKernelSymbols: Failed: %Rrc\n", rc));
1754	RTDbgModRelease(hMod);
1755	}
1756	else
1757	Log(("dbgDiggerLinuxLoadKernelSymbols: RTDbgModCreate failed: %Rrc\n", rc));
1758	}
1759	else
1760	Log(("dbgDiggerLinuxLoadKernelSymbols: Reading token index at %RGv failed: %Rrc\n",
1761	pThis->AddrKernelTokenIndex.FlatPtr, rc));
1762	RTMemFree(paoffTokens);
1763	}
1764	else
1765	Log(("dbgDiggerLinuxLoadKernelSymbols: Reading token table at %RGv failed: %Rrc\n",
1766	pThis->AddrKernelTokenTable.FlatPtr, rc));
1767	RTMemFree(pszzTokens);
1768	}
1769	else
1770	Log(("dbgDiggerLinuxLoadKernelSymbols: Reading encoded names at %RGv failed: %Rrc\n",
1771	pThis->AddrKernelNames.FlatPtr, rc));
1772	RTMemFree(pbNames);
1773
1774	return rc;
1775	}
1776
1777	/**
1778	* Loads the kernel symbols from the kallsyms table if it contains absolute addresses
1779	*
1780	* @returns VBox status code.
1781	* @param pUVM The user mode VM handle.
1782	* @param pThis The Linux digger data.
1783	*/
1784	static int dbgDiggerLinuxLoadKernelSymbolsAbsolute(PUVM pUVM, PDBGDIGGERLINUX pThis)
1785	{
1786	/*
1787	* Allocate memory for temporary table copies, reading the tables as we go.
1788	*/
1789	uint32_t const cbGuestAddr = pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t);
1790	void pvAddresses = RTMemAllocZ(pThis->cKernelSymbols cbGuestAddr);
1791	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &pThis->AddrKernelAddresses, pvAddresses, pThis->cKernelSymbols * cbGuestAddr);
1792	if (RT_SUCCESS(rc))
1793	{
1794	/*
1795	* Figure out the kernel start and end and convert the absolute addresses to relative offsets.
1796	*/
1797	RTGCUINTPTR uKernelStart = pThis->AddrKernelAddresses.FlatPtr;
1798	RTGCUINTPTR uKernelEnd = pThis->AddrKernelTokenIndex.FlatPtr + 256 * sizeof(uint16_t);
1799	RTGCUINTPTR pauSymOff = (RTGCUINTPTR )RTMemTmpAllocZ(pThis->cKernelSymbols * sizeof(RTGCUINTPTR));
1800	uint32_t i;
1801	if (cbGuestAddr == sizeof(uint64_t))
1802	{
1803	uint64_t pauAddrs = (uint64_t )pvAddresses;
1804	for (i = 0; i < pThis->cKernelSymbols; i++)
1805	if ( pauAddrs[i] < uKernelStart
1806	&& LNX64_VALID_ADDRESS(pauAddrs[i])
1807	&& uKernelStart - pauAddrs[i] < LNX_MAX_KERNEL_SIZE)
1808	uKernelStart = pauAddrs[i];
1809
1810	for (i = pThis->cKernelSymbols - 1; i > 0; i--)
1811	if ( pauAddrs[i] > uKernelEnd
1812	&& LNX64_VALID_ADDRESS(pauAddrs[i])
1813	&& pauAddrs[i] - uKernelEnd < LNX_MAX_KERNEL_SIZE)
1814	uKernelEnd = pauAddrs[i];
1815
1816	for (i = 0; i < pThis->cKernelSymbols; i++)
1817	pauSymOff[i] = pauAddrs[i] - uKernelStart;
1818	}
1819	else
1820	{
1821	uint32_t pauAddrs = (uint32_t )pvAddresses;
1822	for (i = 0; i < pThis->cKernelSymbols; i++)
1823	if ( pauAddrs[i] < uKernelStart
1824	&& LNX32_VALID_ADDRESS(pauAddrs[i])
1825	&& uKernelStart - pauAddrs[i] < LNX_MAX_KERNEL_SIZE)
1826	uKernelStart = pauAddrs[i];
1827
1828	for (i = pThis->cKernelSymbols - 1; i > 0; i--)
1829	if ( pauAddrs[i] > uKernelEnd
1830	&& LNX32_VALID_ADDRESS(pauAddrs[i])
1831	&& pauAddrs[i] - uKernelEnd < LNX_MAX_KERNEL_SIZE)
1832	uKernelEnd = pauAddrs[i];
1833
1834	for (i = 0; i < pThis->cKernelSymbols; i++)
1835	pauSymOff[i] = pauAddrs[i] - uKernelStart;
1836	}
1837
1838	RTGCUINTPTR cbKernel = uKernelEnd - uKernelStart;
1839	pThis->cbKernel = (uint32_t)cbKernel;
1840	DBGFR3AddrFromFlat(pUVM, &pThis->AddrKernelBase, uKernelStart);
1841	Log(("dbgDiggerLinuxLoadKernelSymbolsAbsolute: uKernelStart=%RGv cbKernel=%#x\n", uKernelStart, cbKernel));
1842
1843	rc = dbgDiggerLinuxLoadKernelSymbolsWorker(pUVM, pThis, uKernelStart, cbKernel, pauSymOff);
1844	if (RT_FAILURE(rc))
1845	Log(("dbgDiggerLinuxLoadKernelSymbolsAbsolute: Loading symbols from given offset table failed: %Rrc\n", rc));
1846	RTMemTmpFree(pauSymOff);
1847	}
1848	else
1849	Log(("dbgDiggerLinuxLoadKernelSymbolsAbsolute: Reading symbol addresses at %RGv failed: %Rrc\n",
1850	pThis->AddrKernelAddresses.FlatPtr, rc));
1851	RTMemFree(pvAddresses);
1852
1853	return rc;
1854	}
1855
1856
1857	/**
1858	* Loads the kernel symbols from the kallsyms table if it contains absolute addresses
1859	*
1860	* @returns VBox status code.
1861	* @param pUVM The user mode VM handle.
1862	* @param pThis The Linux digger data.
1863	*/
1864	static int dbgDiggerLinuxLoadKernelSymbolsRelative(PUVM pUVM, PDBGDIGGERLINUX pThis)
1865	{
1866	/*
1867	* Allocate memory for temporary table copies, reading the tables as we go.
1868	*/
1869	int32_t pai32Offsets = (int32_t )RTMemAllocZ(pThis->cKernelSymbols * sizeof(int32_t));
1870	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &pThis->AddrKernelAddresses, pai32Offsets, pThis->cKernelSymbols * sizeof(int32_t));
1871	if (RT_SUCCESS(rc))
1872	{
1873	/*
1874	* Figure out the kernel start and end and convert the absolute addresses to relative offsets.
1875	*/
1876	RTGCUINTPTR uKernelStart = pThis->AddrKernelAddresses.FlatPtr;
1877	RTGCUINTPTR uKernelEnd = pThis->AddrKernelTokenIndex.FlatPtr + 256 * sizeof(uint16_t);
1878	RTGCUINTPTR pauSymOff = (RTGCUINTPTR )RTMemTmpAllocZ(pThis->cKernelSymbols * sizeof(RTGCUINTPTR));
1879	uint32_t i;
1880
1881	for (i = 0; i < pThis->cKernelSymbols; i++)
1882	{
1883	RTGCUINTPTR uSymAddr = dbgDiggerLinuxConvOffsetToAddr(pThis, pai32Offsets[i]);
1884
1885	if ( uSymAddr < uKernelStart
1886	&& (pThis->f64Bit ? LNX64_VALID_ADDRESS(uSymAddr) : LNX32_VALID_ADDRESS(uSymAddr))
1887	&& uKernelStart - uSymAddr < LNX_MAX_KERNEL_SIZE)
1888	uKernelStart = uSymAddr;
1889	}
1890
1891	for (i = pThis->cKernelSymbols - 1; i > 0; i--)
1892	{
1893	RTGCUINTPTR uSymAddr = dbgDiggerLinuxConvOffsetToAddr(pThis, pai32Offsets[i]);
1894
1895	if ( uSymAddr > uKernelEnd
1896	&& (pThis->f64Bit ? LNX64_VALID_ADDRESS(uSymAddr) : LNX32_VALID_ADDRESS(uSymAddr))
1897	&& uSymAddr - uKernelEnd < LNX_MAX_KERNEL_SIZE)
1898	uKernelEnd = uSymAddr;
1899
1900	/* Store the offset from the derived kernel start address. */
1901	pauSymOff[i] = uSymAddr - uKernelStart;
1902	}
1903
1904	RTGCUINTPTR cbKernel = uKernelEnd - uKernelStart;
1905	pThis->cbKernel = (uint32_t)cbKernel;
1906	DBGFR3AddrFromFlat(pUVM, &pThis->AddrKernelBase, uKernelStart);
1907	Log(("dbgDiggerLinuxLoadKernelSymbolsRelative: uKernelStart=%RGv cbKernel=%#x\n", uKernelStart, cbKernel));
1908
1909	rc = dbgDiggerLinuxLoadKernelSymbolsWorker(pUVM, pThis, uKernelStart, cbKernel, pauSymOff);
1910	if (RT_FAILURE(rc))
1911	Log(("dbgDiggerLinuxLoadKernelSymbolsRelative: Loading symbols from given offset table failed: %Rrc\n", rc));
1912	RTMemTmpFree(pauSymOff);
1913	}
1914	else
1915	Log(("dbgDiggerLinuxLoadKernelSymbolsRelative: Reading symbol addresses at %RGv failed: %Rrc\n",
1916	pThis->AddrKernelAddresses.FlatPtr, rc));
1917	RTMemFree(pai32Offsets);
1918
1919	return rc;
1920	}
1921
1922
1923	/**
1924	* Loads the kernel symbols.
1925	*
1926	* @returns VBox status code.
1927	* @param pUVM The user mode VM handle.
1928	* @param pThis The Linux digger data.
1929	*/
1930	static int dbgDiggerLinuxLoadKernelSymbols(PUVM pUVM, PDBGDIGGERLINUX pThis)
1931	{
1932	/*
1933	* First the kernel itself.
1934	*/
1935	if (pThis->fRelKrnlAddr)
1936	return dbgDiggerLinuxLoadKernelSymbolsRelative(pUVM, pThis);
1937	return dbgDiggerLinuxLoadKernelSymbolsAbsolute(pUVM, pThis);
1938	}
1939
1940
1941	/*
1942	* The module structure changed it was easier to produce different code for
1943	* each version of the structure. The C preprocessor rules!
1944	*/
1945	#define LNX_TEMPLATE_HEADER "DBGPlugInLinuxModuleCodeTmpl.cpp.h"
1946
1947	#define LNX_BIT_SUFFIX _amd64
1948	#define LNX_PTR_T uint64_t
1949	#define LNX_64BIT 1
1950	#include "DBGPlugInLinuxModuleVerTmpl.cpp.h"
1951
1952	#define LNX_BIT_SUFFIX _x86
1953	#define LNX_PTR_T uint32_t
1954	#define LNX_64BIT 0
1955	#include "DBGPlugInLinuxModuleVerTmpl.cpp.h"
1956
1957	#undef LNX_TEMPLATE_HEADER
1958
1959	static const struct
1960	{
1961	uint32_t uVersion;
1962	bool f64Bit;
1963	uint64_t (*pfnProcessModule)(PDBGDIGGERLINUX pThis, PUVM pUVM, PDBGFADDRESS pAddrModule);
1964	} g_aModVersions[] =
1965	{
1966	#define LNX_TEMPLATE_HEADER "DBGPlugInLinuxModuleTableEntryTmpl.cpp.h"
1967
1968	#define LNX_BIT_SUFFIX _amd64
1969	#define LNX_64BIT 1
1970	#include "DBGPlugInLinuxModuleVerTmpl.cpp.h"
1971
1972	#define LNX_BIT_SUFFIX _x86
1973	#define LNX_64BIT 0
1974	#include "DBGPlugInLinuxModuleVerTmpl.cpp.h"
1975
1976	#undef LNX_TEMPLATE_HEADER
1977	};
1978
1979
1980	/**
1981	* Tries to find and process the module list.
1982	*
1983	* @returns VBox status code.
1984	* @param pThis The Linux digger data.
1985	* @param pUVM The user mode VM handle.
1986	*/
1987	static int dbgDiggerLinuxLoadModules(PDBGDIGGERLINUX pThis, PUVM pUVM)
1988	{
1989	/*
1990	* Locate the list head.
1991	*/
1992	RTDBGAS hAs = DBGFR3AsResolveAndRetain(pUVM, DBGF_AS_KERNEL);
1993	RTDBGSYMBOL SymInfo;
1994	int rc = RTDbgAsSymbolByName(hAs, "vmlinux!modules", &SymInfo, NULL);
1995	RTDbgAsRelease(hAs);
1996	if (RT_FAILURE(rc))
1997	return VERR_NOT_FOUND;
1998
1999	if (RT_FAILURE(rc))
2000	{
2001	LogRel(("dbgDiggerLinuxLoadModules: Failed to locate the module list (%Rrc).\n", rc));
2002	return VERR_NOT_FOUND;
2003	}
2004
2005	/*
2006	* Read the list anchor.
2007	*/
2008	union
2009	{
2010	uint32_t volatile u32Pair[2];
2011	uint64_t u64Pair[2];
2012	} uListAnchor;
2013	DBGFADDRESS Addr;
2014	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, DBGFR3AddrFromFlat(pUVM, &Addr, SymInfo.Value),
2015	&uListAnchor, pThis->f64Bit ? sizeof(uListAnchor.u64Pair) : sizeof(uListAnchor.u32Pair));
2016	if (RT_FAILURE(rc))
2017	{
2018	LogRel(("dbgDiggerLinuxLoadModules: Error reading list anchor at %RX64: %Rrc\n", SymInfo.Value, rc));
2019	return VERR_NOT_FOUND;
2020	}
2021	if (!pThis->f64Bit)
2022	{
2023	uListAnchor.u64Pair[1] = uListAnchor.u32Pair[1];
2024	ASMCompilerBarrier();
2025	uListAnchor.u64Pair[0] = uListAnchor.u32Pair[0];
2026	}
2027
2028	/*
2029	* Get a numerical version number.
2030	*/
2031	char szVersion[256] = "Linux version 4.19.0";
2032	bool fValid = pThis->fValid;
2033	pThis->fValid = true;
2034	dbgDiggerLinuxQueryVersion(pUVM, pThis, szVersion, sizeof(szVersion));
2035	pThis->fValid = fValid;
2036
2037	const char *pszVersion = szVersion;
2038	while (pszVersion && !RT_C_IS_DIGIT(pszVersion))
2039	pszVersion++;
2040
2041	size_t offVersion = 0;
2042	uint32_t uMajor = 0;
2043	while (pszVersion[offVersion] && RT_C_IS_DIGIT(pszVersion[offVersion]))
2044	uMajor = uMajor * 10 + pszVersion[offVersion++] - '0';
2045
2046	if (pszVersion[offVersion] == '.')
2047	offVersion++;
2048
2049	uint32_t uMinor = 0;
2050	while (pszVersion[offVersion] && RT_C_IS_DIGIT(pszVersion[offVersion]))
2051	uMinor = uMinor * 10 + pszVersion[offVersion++] - '0';
2052
2053	if (pszVersion[offVersion] == '.')
2054	offVersion++;
2055
2056	uint32_t uBuild = 0;
2057	while (pszVersion[offVersion] && RT_C_IS_DIGIT(pszVersion[offVersion]))
2058	uBuild = uBuild * 10 + pszVersion[offVersion++] - '0';
2059
2060	uint32_t const uGuestVer = LNX_MK_VER(uMajor, uMinor, uBuild);
2061	if (uGuestVer == 0)
2062	{
2063	LogRel(("dbgDiggerLinuxLoadModules: Failed to parse version string: %s\n", pszVersion));
2064	return VERR_NOT_FOUND;
2065	}
2066
2067	/*
2068	* Find the g_aModVersion entry that fits the best.
2069	* ASSUMES strict descending order by bitcount and version.
2070	*/
2071	Assert(g_aModVersions[0].f64Bit == true);
2072	unsigned i = 0;
2073	if (!pThis->f64Bit)
2074	while (i < RT_ELEMENTS(g_aModVersions) && g_aModVersions[i].f64Bit)
2075	i++;
2076	while ( i < RT_ELEMENTS(g_aModVersions)
2077	&& g_aModVersions[i].f64Bit == pThis->f64Bit
2078	&& uGuestVer < g_aModVersions[i].uVersion)
2079	i++;
2080	if (i >= RT_ELEMENTS(g_aModVersions))
2081	{
2082	LogRel(("dbgDiggerLinuxLoadModules: Failed to find anything matching version: %u.%u.%u (%s)\n",
2083	uMajor, uMinor, uBuild, pszVersion));
2084	return VERR_NOT_FOUND;
2085	}
2086
2087	/*
2088	* Walk the list.
2089	*/
2090	uint64_t uModAddr = uListAnchor.u64Pair[0];
2091	for (size_t iModule = 0; iModule < 4096 && uModAddr != SymInfo.Value && uModAddr != 0; iModule++)
2092	uModAddr = g_aModVersions[i].pfnProcessModule(pThis, pUVM, DBGFR3AddrFromFlat(pUVM, &Addr, uModAddr));
2093
2094	return VINF_SUCCESS;
2095	}
2096
2097
2098	/**
2099	* Checks if there is a likely kallsyms_names fragment at pHitAddr.
2100	*
2101	* @returns true if it's a likely fragment, false if not.
2102	* @param pUVM The user mode VM handle.
2103	* @param pHitAddr The address where paNeedle was found.
2104	* @param pabNeedle The fragment we've been searching for.
2105	* @param cbNeedle The length of the fragment.
2106	*/
2107	static bool dbgDiggerLinuxIsLikelyNameFragment(PUVM pUVM, PCDBGFADDRESS pHitAddr, uint8_t const *pabNeedle, uint8_t cbNeedle)
2108	{
2109	/*
2110	* Examples of lead and tail bytes of our choosen needle in a randomly
2111	* picked kernel:
2112	* k o b j
2113	* 22 6b 6f 62 6a aa
2114	* fc 6b 6f 62 6a aa
2115	* 82 6b 6f 62 6a 5f - ascii trail byte (_).
2116	* ee 6b 6f 62 6a aa
2117	* fc 6b 6f 62 6a 5f - ascii trail byte (_).
2118	* 0a 74 6b 6f 62 6a 5f ea - ascii lead (t) and trail (_) bytes.
2119	* 0b 54 6b 6f 62 6a aa - ascii lead byte (T).
2120	* ... omitting 29 samples similar to the last two ...
2121	* d8 6b 6f 62 6a aa
2122	* d8 6b 6f 62 6a aa
2123	* d8 6b 6f 62 6a aa
2124	* d8 6b 6f 62 6a aa
2125	* f9 5f 6b 6f 62 6a 5f 94 - ascii lead and trail bytes (_)
2126	* f9 5f 6b 6f 62 6a 0c - ascii lead byte (_).
2127	* fd 6b 6f 62 6a 0f
2128	* ... enough.
2129	*/
2130	uint8_t abBuf[32];
2131	DBGFADDRESS ReadAddr = *pHitAddr;
2132	DBGFR3AddrSub(&ReadAddr, 2);
2133	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &ReadAddr, abBuf, 2 + cbNeedle + 2);
2134	if (RT_SUCCESS(rc))
2135	{
2136	if (memcmp(&abBuf[2], pabNeedle, cbNeedle) == 0) /* paranoia */
2137	{
2138	uint8_t const bLead = abBuf[1] == '_' \|\| abBuf[1] == 'T' \|\| abBuf[1] == 't' ? abBuf[0] : abBuf[1];
2139	uint8_t const offTail = 2 + cbNeedle;
2140	uint8_t const bTail = abBuf[offTail] == '_' ? abBuf[offTail] : abBuf[offTail + 1];
2141	if ( bLead >= 1 && (bLead < 0x20 \|\| bLead >= 0x80)
2142	&& bTail >= 1 && (bTail < 0x20 \|\| bTail >= 0x80))
2143	return true;
2144	Log(("dbgDiggerLinuxIsLikelyNameFragment: failed at %RGv: bLead=%#x bTail=%#x (offTail=%#x)\n",
2145	pHitAddr->FlatPtr, bLead, bTail, offTail));
2146	}
2147	else
2148	Log(("dbgDiggerLinuxIsLikelyNameFragment: failed at %RGv: Needle changed!\n", pHitAddr->FlatPtr));
2149	}
2150	else
2151	Log(("dbgDiggerLinuxIsLikelyNameFragment: failed at %RGv: %Rrc\n", pHitAddr->FlatPtr, rc));
2152
2153	return false;
2154	}
2155
2156	/**
2157	* Tries to find and load the kernel symbol table with the given needle.
2158	*
2159	* @returns VBox status code.
2160	* @param pThis The Linux digger data.
2161	* @param pUVM The user mode VM handle.
2162	* @param pabNeedle The needle to use for searching.
2163	* @param cbNeedle Size of the needle in bytes.
2164	*/
2165	static int dbgDiggerLinuxFindSymbolTableFromNeedle(PDBGDIGGERLINUX pThis, PUVM pUVM, uint8_t const *pabNeedle, uint8_t cbNeedle)
2166	{
2167	int rc = VINF_SUCCESS;
2168
2169	/*
2170	* Go looking for the kallsyms table. If it's there, it will be somewhere
2171	* after the linux_banner symbol, so use it for starting the search.
2172	*/
2173	DBGFADDRESS CurAddr = pThis->AddrLinuxBanner;
2174	uint32_t cbLeft = LNX_MAX_KERNEL_SIZE;
2175	while (cbLeft > 4096)
2176	{
2177	DBGFADDRESS HitAddr;
2178	rc = DBGFR3MemScan(pUVM, 0 /idCpu/, &CurAddr, cbLeft, 1 /uAlign/,
2179	pabNeedle, cbNeedle, &HitAddr);
2180	if (RT_FAILURE(rc))
2181	break;
2182	if (dbgDiggerLinuxIsLikelyNameFragment(pUVM, &HitAddr, pabNeedle, cbNeedle))
2183	{
2184	/* There will be another hit near by. */
2185	DBGFR3AddrAdd(&HitAddr, 1);
2186	rc = DBGFR3MemScan(pUVM, 0 /idCpu/, &HitAddr, LNX_MAX_KALLSYMS_NAMES_SIZE, 1 /uAlign/,
2187	pabNeedle, cbNeedle, &HitAddr);
2188	if ( RT_SUCCESS(rc)
2189	&& dbgDiggerLinuxIsLikelyNameFragment(pUVM, &HitAddr, pabNeedle, cbNeedle))
2190	{
2191	/*
2192	* We've got a very likely candidate for a location inside kallsyms_names.
2193	* Try find the start of it, that is to say, try find kallsyms_num_syms.
2194	* kallsyms_num_syms is aligned on sizeof(unsigned long) boundrary
2195	*/
2196	rc = dbgDiggerLinuxFindStartOfNamesAndSymbolCount(pUVM, pThis, &HitAddr);
2197	if (RT_SUCCESS(rc))
2198	rc = dbgDiggerLinuxFindEndOfNamesAndMore(pUVM, pThis, &HitAddr);
2199	if (RT_SUCCESS(rc))
2200	rc = dbgDiggerLinuxFindTokenIndex(pUVM, pThis);
2201	if (RT_SUCCESS(rc))
2202	rc = dbgDiggerLinuxLoadKernelSymbols(pUVM, pThis);
2203	if (RT_SUCCESS(rc))
2204	{
2205	rc = dbgDiggerLinuxLoadModules(pThis, pUVM);
2206	break;
2207	}
2208	}
2209	}
2210
2211	/*
2212	* Advance.
2213	*/
2214	RTGCUINTPTR cbDistance = HitAddr.FlatPtr - CurAddr.FlatPtr + cbNeedle;
2215	if (RT_UNLIKELY(cbDistance >= cbLeft))
2216	{
2217	Log(("dbgDiggerLinuxInit: Failed to find kallsyms\n"));
2218	break;
2219	}
2220	cbLeft -= cbDistance;
2221	DBGFR3AddrAdd(&CurAddr, cbDistance);
2222
2223	}
2224
2225	return rc;
2226	}
2227
2228	/**
2229	* Skips whitespace and comments in the given config returning the pointer
2230	* to the first non whitespace character.
2231	*
2232	* @returns Pointer to the first non whitespace character or NULL if the end
2233	* of the string was reached.
2234	* @param pszCfg The config string.
2235	*/
2236	static const char dbgDiggerLinuxCfgSkipWhitespace(const char pszCfg)
2237	{
2238	do
2239	{
2240	while ( *pszCfg != '\0'
2241	&& ( RT_C_IS_SPACE(*pszCfg)
2242	\|\| *pszCfg == '\n'))
2243	pszCfg++;
2244
2245	/* Do we have a comment? Skip it. */
2246	if (*pszCfg == '#')
2247	{
2248	while ( *pszCfg != '\n'
2249	&& *pszCfg != '\0')
2250	pszCfg++;
2251	}
2252	} while ( *pszCfg != '\0'
2253	&& ( RT_C_IS_SPACE(*pszCfg)
2254	\|\| *pszCfg == '\n'
2255	\|\| *pszCfg == '#'));
2256
2257	return pszCfg;
2258	}
2259
2260	/**
2261	* Parses an identifier at the given position.
2262	*
2263	* @returns VBox status code.
2264	* @param pszCfg The config data.
2265	* @param ppszCfgNext Where to store the pointer to the data following the identifier.
2266	* @param ppszIde Where to store the pointer to the identifier on success.
2267	* Free with RTStrFree().
2268	*/
2269	static int dbgDiggerLinuxCfgParseIde(const char pszCfg, const char ppszCfgNext, char *ppszIde)
2270	{
2271	int rc = VINF_SUCCESS;
2272	size_t cchIde = 0;
2273
2274	while ( *pszCfg != '\0'
2275	&& ( RT_C_IS_ALNUM(*pszCfg)
2276	\|\| *pszCfg == '_'))
2277	{
2278	cchIde++;
2279	pszCfg++;
2280	}
2281
2282	if (cchIde)
2283	{
2284	*ppszIde = RTStrDupN(pszCfg - cchIde, cchIde);
2285	if (!*ppszIde)
2286	rc = VERR_NO_STR_MEMORY;
2287	}
2288
2289	*ppszCfgNext = pszCfg;
2290	return rc;
2291	}
2292
2293	/**
2294	* Parses a value for a config item.
2295	*
2296	* @returns VBox status code.
2297	* @param pszCfg The config data.
2298	* @param ppszCfgNext Where to store the pointer to the data following the identifier.
2299	* @param ppCfgItem Where to store the created config item on success.
2300	*/
2301	static int dbgDiggerLinuxCfgParseVal(const char pszCfg, const char *ppszCfgNext,
2302	PDBGDIGGERLINUXCFGITEM *ppCfgItem)
2303	{
2304	int rc = VINF_SUCCESS;
2305	PDBGDIGGERLINUXCFGITEM pCfgItem = NULL;
2306
2307	if (RT_C_IS_DIGIT(pszCfg) \|\| pszCfg == '-')
2308	{
2309	/* Parse the number. */
2310	int64_t i64Num;
2311	rc = RTStrToInt64Ex(pszCfg, (char **)ppszCfgNext, 0, &i64Num);
2312	if ( RT_SUCCESS(rc)
2313	\|\| rc == VWRN_TRAILING_CHARS
2314	\|\| rc == VWRN_TRAILING_SPACES)
2315	{
2316	pCfgItem = (PDBGDIGGERLINUXCFGITEM)RTMemAllocZ(sizeof(DBGDIGGERLINUXCFGITEM));
2317	if (pCfgItem)
2318	{
2319	pCfgItem->enmType = DBGDIGGERLINUXCFGITEMTYPE_NUMBER;
2320	pCfgItem->u.i64Num = i64Num;
2321	}
2322	else
2323	rc = VERR_NO_MEMORY;
2324	}
2325	}
2326	else if (*pszCfg == '\"')
2327	{
2328	/* Parse a string. */
2329	const char *pszCfgCur = pszCfg + 1;
2330	while ( *pszCfgCur != '\0'
2331	&& *pszCfgCur != '\"')
2332	pszCfgCur++;
2333
2334	if (*pszCfgCur == '\"')
2335	{
2336	pCfgItem = (PDBGDIGGERLINUXCFGITEM)RTMemAllocZ(RT_UOFFSETOF_DYN(DBGDIGGERLINUXCFGITEM,
2337	u.aszString[pszCfgCur - pszCfg + 1]));
2338	if (pCfgItem)
2339	{
2340	pCfgItem->enmType = DBGDIGGERLINUXCFGITEMTYPE_STRING;
2341	RTStrCopyEx(&pCfgItem->u.aszString[0], pszCfgCur - pszCfg + 1, pszCfg, pszCfgCur - pszCfg);
2342	*ppszCfgNext = pszCfgCur + 1;
2343	}
2344	else
2345	rc = VERR_NO_MEMORY;
2346	}
2347	else
2348	rc = VERR_INVALID_STATE;
2349	}
2350	else if ( *pszCfg == 'y'
2351	\|\| *pszCfg == 'm')
2352	{
2353	/* Included or module. */
2354	pCfgItem = (PDBGDIGGERLINUXCFGITEM)RTMemAllocZ(sizeof(DBGDIGGERLINUXCFGITEM));
2355	if (pCfgItem)
2356	{
2357	pCfgItem->enmType = DBGDIGGERLINUXCFGITEMTYPE_FLAG;
2358	pCfgItem->u.fModule = *pszCfg == 'm';
2359	}
2360	else
2361	rc = VERR_NO_MEMORY;
2362	pszCfg++;
2363	*ppszCfgNext = pszCfg;
2364	}
2365	else
2366	rc = VERR_INVALID_STATE;
2367
2368	if (RT_SUCCESS(rc))
2369	*ppCfgItem = pCfgItem;
2370	else if (pCfgItem)
2371	RTMemFree(pCfgItem);
2372
2373	return rc;
2374	}
2375
2376	/**
2377	* Parses the given kernel config and creates the config database.
2378	*
2379	* @returns VBox status code
2380	* @param pThis The Linux digger data.
2381	* @param pszCfg The config string.
2382	*/
2383	static int dbgDiggerLinuxCfgParse(PDBGDIGGERLINUX pThis, const char *pszCfg)
2384	{
2385	int rc = VINF_SUCCESS;
2386
2387	/*
2388	* The config is a text file with the following elements:
2389	* # starts a comment which goes till the end of the line
2390	* <Ide>=<val> where <Ide> is an identifier consisting of
2391	* alphanumerical characters (including _)
2392	* <val> denotes the value for the identifier and can have the following
2393	* formats:
2394	* (-)[0-9]* for numbers
2395	* "..." for a string value
2396	* m when a feature is enabled as a module
2397	* y when a feature is enabled
2398	* Newlines are used as a separator between values and mark the end
2399	* of a comment
2400	*/
2401	const char *pszCfgCur = pszCfg;
2402	while ( RT_SUCCESS(rc)
2403	&& *pszCfgCur != '\0')
2404	{
2405	/* Start skipping the whitespace. */
2406	pszCfgCur = dbgDiggerLinuxCfgSkipWhitespace(pszCfgCur);
2407	if ( pszCfgCur
2408	&& *pszCfgCur != '\0')
2409	{
2410	char *pszIde = NULL;
2411	/* Must be an identifier, parse it. */
2412	rc = dbgDiggerLinuxCfgParseIde(pszCfgCur, &pszCfgCur, &pszIde);
2413	if (RT_SUCCESS(rc))
2414	{
2415	/*
2416	* Skip whitespace again (shouldn't be required because = follows immediately
2417	* in the observed configs).
2418	*/
2419	pszCfgCur = dbgDiggerLinuxCfgSkipWhitespace(pszCfgCur);
2420	if ( pszCfgCur
2421	&& *pszCfgCur == '=')
2422	{
2423	pszCfgCur++;
2424	pszCfgCur = dbgDiggerLinuxCfgSkipWhitespace(pszCfgCur);
2425	if ( pszCfgCur
2426	&& *pszCfgCur != '\0')
2427	{
2428	/* Get the value. */
2429	PDBGDIGGERLINUXCFGITEM pCfgItem = NULL;
2430	rc = dbgDiggerLinuxCfgParseVal(pszCfgCur, &pszCfgCur, &pCfgItem);
2431	if (RT_SUCCESS(rc))
2432	{
2433	pCfgItem->Core.pszString = pszIde;
2434	bool fRc = RTStrSpaceInsert(&pThis->hCfgDb, &pCfgItem->Core);
2435	if (!fRc)
2436	{
2437	RTStrFree(pszIde);
2438	RTMemFree(pCfgItem);
2439	rc = VERR_INVALID_STATE;
2440	}
2441	}
2442	}
2443	else
2444	rc = VERR_EOF;
2445	}
2446	else
2447	rc = VERR_INVALID_STATE;
2448	}
2449
2450	if (RT_FAILURE(rc))
2451	RTStrFree(pszIde);
2452	}
2453	else
2454	break; /* Reached the end of the config. */
2455	}
2456
2457	if (RT_FAILURE(rc))
2458	dbgDiggerLinuxCfgDbDestroy(pThis);
2459
2460	return rc;
2461	}
2462
2463	/**
2464	* Decompresses the given config and validates the UTF-8 encoding.
2465	*
2466	* @returns VBox status code.
2467	* @param pbCfgComp The compressed config.
2468	* @param cbCfgComp Size of the compressed config.
2469	* @param ppszCfg Where to store the pointer to the decompressed config
2470	* on success.
2471	*/
2472	static int dbgDiggerLinuxCfgDecompress(const uint8_t pbCfgComp, size_t cbCfgComp, char *ppszCfg)
2473	{
2474	int rc = VINF_SUCCESS;
2475	RTVFSIOSTREAM hVfsIos = NIL_RTVFSIOSTREAM;
2476
2477	rc = RTVfsIoStrmFromBuffer(RTFILE_O_READ, pbCfgComp, cbCfgComp, &hVfsIos);
2478	if (RT_SUCCESS(rc))
2479	{
2480	RTVFSIOSTREAM hVfsIosDecomp = NIL_RTVFSIOSTREAM;
2481	rc = RTZipGzipDecompressIoStream(hVfsIos, RTZIPGZIPDECOMP_F_ALLOW_ZLIB_HDR, &hVfsIosDecomp);
2482	if (RT_SUCCESS(rc))
2483	{
2484	char *pszCfg = NULL;
2485	size_t cchCfg = 0;
2486	size_t cbRead = 0;
2487
2488	do
2489	{
2490	uint8_t abBuf[_64K];
2491	rc = RTVfsIoStrmRead(hVfsIosDecomp, abBuf, sizeof(abBuf), true /fBlocking/, &cbRead);
2492	if (rc == VINF_EOF && cbRead == 0)
2493	rc = VINF_SUCCESS;
2494	if ( RT_SUCCESS(rc)
2495	&& cbRead > 0)
2496	{
2497	/* Append data. */
2498	char *pszCfgNew = pszCfg;
2499	rc = RTStrRealloc(&pszCfgNew, cchCfg + cbRead + 1);
2500	if (RT_SUCCESS(rc))
2501	{
2502	pszCfg = pszCfgNew;
2503	memcpy(pszCfg + cchCfg, &abBuf[0], cbRead);
2504	cchCfg += cbRead;
2505	pszCfg[cchCfg] = '\0'; /* Enforce string termination. */
2506	}
2507	}
2508	} while (RT_SUCCESS(rc) && cbRead > 0);
2509
2510	if (RT_SUCCESS(rc))
2511	*ppszCfg = pszCfg;
2512	else if (RT_FAILURE(rc) && pszCfg)
2513	RTStrFree(pszCfg);
2514
2515	RTVfsIoStrmRelease(hVfsIosDecomp);
2516	}
2517	RTVfsIoStrmRelease(hVfsIos);
2518	}
2519
2520	return rc;
2521	}
2522
2523	/**
2524	* Reads and decodes the compressed kernel config.
2525	*
2526	* @returns VBox status code.
2527	* @param pThis The Linux digger data.
2528	* @param pUVM The user mode VM handle.
2529	* @param pAddrStart The start address of the compressed config.
2530	* @param cbCfgComp The size of the compressed config.
2531	*/
2532	static int dbgDiggerLinuxCfgDecode(PDBGDIGGERLINUX pThis, PUVM pUVM,
2533	PCDBGFADDRESS pAddrStart, size_t cbCfgComp)
2534	{
2535	int rc = VINF_SUCCESS;
2536	uint8_t pbCfgComp = (uint8_t )RTMemTmpAlloc(cbCfgComp);
2537	if (!pbCfgComp)
2538	return VERR_NO_MEMORY;
2539
2540	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, pAddrStart, pbCfgComp, cbCfgComp);
2541	if (RT_SUCCESS(rc))
2542	{
2543	char *pszCfg = NULL;
2544	rc = dbgDiggerLinuxCfgDecompress(pbCfgComp, cbCfgComp, &pszCfg);
2545	if (RT_SUCCESS(rc))
2546	{
2547	if (RTStrIsValidEncoding(pszCfg))
2548	rc = dbgDiggerLinuxCfgParse(pThis, pszCfg);
2549	else
2550	rc = VERR_INVALID_UTF8_ENCODING;
2551	RTStrFree(pszCfg);
2552	}
2553	}
2554
2555	RTMemFree(pbCfgComp);
2556	return rc;
2557	}
2558
2559	/**
2560	* Tries to find the compressed kernel config in the kernel address space
2561	* and sets up the config database.
2562	*
2563	* @returns VBox status code.
2564	* @param pThis The Linux digger data.
2565	* @param pUVM The user mode VM handle.
2566	*/
2567	static int dbgDiggerLinuxCfgFind(PDBGDIGGERLINUX pThis, PUVM pUVM)
2568	{
2569	int rc = VINF_SUCCESS;
2570
2571	/*
2572	* Go looking for the IKCFG_ST string which indicates the start
2573	* of the compressed config file.
2574	*/
2575	static const uint8_t s_abCfgNeedleStart[] = "IKCFG_ST";
2576	static const uint8_t s_abCfgNeedleEnd[] = "IKCFG_ED";
2577	DBGFADDRESS CurAddr = pThis->AddrLinuxBanner;
2578	uint32_t cbLeft = LNX_MAX_KERNEL_SIZE;
2579	while (cbLeft > 4096)
2580	{
2581	DBGFADDRESS HitAddrStart;
2582	rc = DBGFR3MemScan(pUVM, 0 /idCpu/, &CurAddr, cbLeft, 1 /uAlign/,
2583	s_abCfgNeedleStart, sizeof(s_abCfgNeedleStart) - 1, &HitAddrStart);
2584	if (RT_FAILURE(rc))
2585	break;
2586
2587	/* Check for the end marker which shouldn't be that far away. */
2588	DBGFR3AddrAdd(&HitAddrStart, sizeof(s_abCfgNeedleStart) - 1);
2589	DBGFADDRESS HitAddrEnd;
2590	rc = DBGFR3MemScan(pUVM, 0 /* idCpu */, &HitAddrStart, LNX_MAX_COMPRESSED_CFG_SIZE,
2591	1 /* uAlign */, s_abCfgNeedleEnd, sizeof(s_abCfgNeedleEnd) - 1, &HitAddrEnd);
2592	if (RT_SUCCESS(rc))
2593	{
2594	/* Allocate a buffer to hold the compressed data between the markers and fetch it. */
2595	RTGCUINTPTR cbCfg = HitAddrEnd.FlatPtr - HitAddrStart.FlatPtr;
2596	Assert(cbCfg == (size_t)cbCfg);
2597	rc = dbgDiggerLinuxCfgDecode(pThis, pUVM, &HitAddrStart, cbCfg);
2598	if (RT_SUCCESS(rc))
2599	break;
2600	}
2601
2602	/*
2603	* Advance.
2604	*/
2605	RTGCUINTPTR cbDistance = HitAddrStart.FlatPtr - CurAddr.FlatPtr + sizeof(s_abCfgNeedleStart) - 1;
2606	if (RT_UNLIKELY(cbDistance >= cbLeft))
2607	{
2608	LogFunc(("Failed to find compressed kernel config\n"));
2609	break;
2610	}
2611	cbLeft -= cbDistance;
2612	DBGFR3AddrAdd(&CurAddr, cbDistance);
2613
2614	}
2615
2616	return rc;
2617	}
2618
2619	/**
2620	* Probes for a Linux kernel starting at the given address.
2621	*
2622	* @returns Flag whether something which looks like a valid Linux kernel was found.
2623	* @param pThis The Linux digger data.
2624	* @param pUVM The user mode VM handle.
2625	* @param uAddrStart The address to start scanning at.
2626	* @param cbScan How much to scan.
2627	*/
2628	static bool dbgDiggerLinuxProbeWithAddr(PDBGDIGGERLINUX pThis, PUVM pUVM, RTGCUINTPTR uAddrStart, size_t cbScan)
2629	{
2630	/*
2631	* Look for "Linux version " at the start of the rodata segment.
2632	* Hope that this comes before any message buffer or other similar string.
2633	*/
2634	DBGFADDRESS KernelAddr;
2635	DBGFR3AddrFromFlat(pUVM, &KernelAddr, uAddrStart);
2636	DBGFADDRESS HitAddr;
2637	int rc = DBGFR3MemScan(pUVM, 0, &KernelAddr, cbScan, 1,
2638	g_abLinuxVersion, sizeof(g_abLinuxVersion) - 1, &HitAddr);
2639	if (RT_SUCCESS(rc))
2640	{
2641	char szTmp[128];
2642	char const *pszX = &szTmp[sizeof(g_abLinuxVersion) - 1];
2643	rc = DBGFR3MemReadString(pUVM, 0, &HitAddr, szTmp, sizeof(szTmp));
2644	if ( RT_SUCCESS(rc)
2645	&& ( ( pszX[0] == '2' /* 2.x.y with x in {0..6} */
2646	&& pszX[1] == '.'
2647	&& pszX[2] >= '0'
2648	&& pszX[2] <= '6')
2649	\|\| ( pszX[0] >= '3' /* 3.x, 4.x, ... 9.x */
2650	&& pszX[0] <= '9'
2651	&& pszX[1] == '.'
2652	&& pszX[2] >= '0'
2653	&& pszX[2] <= '9')
2654	)
2655	)
2656	{
2657	pThis->AddrKernelBase = KernelAddr;
2658	pThis->AddrLinuxBanner = HitAddr;
2659	return true;
2660	}
2661	}
2662
2663	return false;
2664	}
2665
2666	/**
2667	* Probes for a Linux kernel which has KASLR enabled.
2668	*
2669	* @returns Flag whether a possible candidate location was found.
2670	* @param pThis The Linux digger data.
2671	* @param pUVM The user mode VM handle.
2672	* @param uAddrKernelStart The first address the kernel is expected at.
2673	*/
2674	static bool dbgDiggerLinuxProbeKaslr(PDBGDIGGERLINUX pThis, PUVM pUVM, RTGCUINTPTR uAddrKernelStart)
2675	{
2676	/**
2677	* With KASLR the kernel is loaded at a different address at each boot making detection
2678	* more difficult for us.
2679	*
2680	* The randomization is done in arch/x86/boot/compressed/kaslr.c:choose_random_location() (as of Nov 2017).
2681	* At the end of the method a random offset is chosen using find_random_virt_addr() which is added to the
2682	* kernel map start in the caller (the start of the kernel depends on the bit size, see LNX32_KERNEL_ADDRESS_START
2683	* and LNX64_KERNEL_ADDRESS_START for 32bit and 64bit kernels respectively).
2684	* The lowest offset possible is LOAD_PHYSICAL_ADDR which is defined in arch/x86/include/asm/boot.h
2685	* using CONFIG_PHYSICAL_START aligned to CONFIG_PHYSICAL_ALIGN.
2686	* The default CONFIG_PHYSICAL_START and CONFIG_PHYSICAL_ALIGN are both 0x1000000 no matter whether a 32bit
2687	* or a 64bit kernel is used. So the lowest offset to the kernel start address is 0x1000000.
2688	* The find_random_virt_addr() the number of possible slots where the kernel can be placed based on the image size
2689	* is calculated using the following formula:
2690	* cSlots = ((KERNEL_IMAGE_SIZE - 0x1000000 (minimum) - image_size) / 0x1000000 (CONFIG_PHYSICAL_ALIGN)) + 1
2691	*
2692	* KERNEL_IMAGE_SIZE is 1GB for 64bit kernels and 512MB for 32bit kernels, so the maximum number of slots (resulting
2693	* in the largest possible offset) can be achieved when image_size (which contains the real size of the kernel image
2694	* which is unknown for us) goes to 0 and a 1GB KERNEL_IMAGE_SIZE is assumed. With that the biggest cSlots which can be
2695	* achieved is 64. The chosen random offset is taken from a random long integer using kaslr_get_random_long() modulo the
2696	* number of slots which selects a slot between 0 and 63. The final offset is calculated using:
2697	* offAddr = random_addr * 0x1000000 (CONFIG_PHYSICAL_ALIGN) + 0x1000000 (minimum)
2698	*
2699	* So the highest offset the kernel can start is 0x40000000 which is 1GB (plus the maximum kernel size we defined).
2700	*/
2701	if (dbgDiggerLinuxProbeWithAddr(pThis, pUVM, uAddrKernelStart, _1G + LNX_MAX_KERNEL_SIZE))
2702	return true;
2703
2704	return false;
2705	}
2706
2707	/**
2708	* @copydoc DBGFOSREG::pfnInit
2709	*/
2710	static DECLCALLBACK(int) dbgDiggerLinuxInit(PUVM pUVM, void *pvData)
2711	{
2712	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
2713	Assert(!pThis->fValid);
2714
2715	/*
2716	* Assume 64-bit kernels all live way beyond 32-bit address space.
2717	*/
2718	pThis->f64Bit = pThis->AddrLinuxBanner.FlatPtr > UINT32_MAX;
2719	pThis->fRelKrnlAddr = false;
2720
2721	pThis->hCfgDb = NULL;
2722
2723	/*
2724	* Try to find the compressed kernel config and parse it before we try
2725	* to get the symbol table, the config database is required to select
2726	* the method to use.
2727	*/
2728	int rc = dbgDiggerLinuxCfgFind(pThis, pUVM);
2729	if (RT_FAILURE(rc))
2730	LogFlowFunc(("Failed to find kernel config (%Rrc), no config database available\n", rc));
2731
2732	static const uint8_t s_abNeedle[] = "kobj";
2733	rc = dbgDiggerLinuxFindSymbolTableFromNeedle(pThis, pUVM, s_abNeedle, sizeof(s_abNeedle) - 1);
2734	if (RT_FAILURE(rc))
2735	{
2736	/* Try alternate needle (seen on older x86 Linux kernels). */
2737	static const uint8_t s_abNeedleAlt[] = "kobjec";
2738	rc = dbgDiggerLinuxFindSymbolTableFromNeedle(pThis, pUVM, s_abNeedleAlt, sizeof(s_abNeedleAlt) - 1);
2739	if (RT_FAILURE(rc))
2740	{
2741	static const uint8_t s_abNeedleOSuseX86[] = "nmi"; /* OpenSuSe 10.2 x86 */
2742	rc = dbgDiggerLinuxFindSymbolTableFromNeedle(pThis, pUVM, s_abNeedleOSuseX86, sizeof(s_abNeedleOSuseX86) - 1);
2743	}
2744	}
2745
2746	pThis->fValid = true;
2747	return VINF_SUCCESS;
2748	}
2749
2750
2751	/**
2752	* @copydoc DBGFOSREG::pfnProbe
2753	*/
2754	static DECLCALLBACK(bool) dbgDiggerLinuxProbe(PUVM pUVM, void *pvData)
2755	{
2756	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
2757
2758	for (unsigned i = 0; i < RT_ELEMENTS(g_au64LnxKernelAddresses); i++)
2759	{
2760	if (dbgDiggerLinuxProbeWithAddr(pThis, pUVM, g_au64LnxKernelAddresses[i], LNX_MAX_KERNEL_SIZE))
2761	return true;
2762	}
2763
2764	/* Maybe the kernel uses KASLR. */
2765	if (dbgDiggerLinuxProbeKaslr(pThis, pUVM, LNX32_KERNEL_ADDRESS_START))
2766	return true;
2767
2768	if (dbgDiggerLinuxProbeKaslr(pThis, pUVM, LNX64_KERNEL_ADDRESS_START))
2769	return true;
2770
2771	return false;
2772	}
2773
2774
2775	/**
2776	* @copydoc DBGFOSREG::pfnDestruct
2777	*/
2778	static DECLCALLBACK(void) dbgDiggerLinuxDestruct(PUVM pUVM, void *pvData)
2779	{
2780	RT_NOREF2(pUVM, pvData);
2781	}
2782
2783
2784	/**
2785	* @copydoc DBGFOSREG::pfnConstruct
2786	*/
2787	static DECLCALLBACK(int) dbgDiggerLinuxConstruct(PUVM pUVM, void *pvData)
2788	{
2789	RT_NOREF1(pUVM);
2790	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
2791	pThis->IDmesg.u32Magic = DBGFOSIDMESG_MAGIC;
2792	pThis->IDmesg.pfnQueryKernelLog = dbgDiggerLinuxIDmsg_QueryKernelLog;
2793	pThis->IDmesg.u32EndMagic = DBGFOSIDMESG_MAGIC;
2794
2795	return VINF_SUCCESS;
2796	}
2797
2798
2799	const DBGFOSREG g_DBGDiggerLinux =
2800	{
2801	/* .u32Magic = */ DBGFOSREG_MAGIC,
2802	/* .fFlags = */ 0,
2803	/* .cbData = */ sizeof(DBGDIGGERLINUX),
2804	/* .szName = */ "Linux",
2805	/* .pfnConstruct = */ dbgDiggerLinuxConstruct,
2806	/* .pfnDestruct = */ dbgDiggerLinuxDestruct,
2807	/* .pfnProbe = */ dbgDiggerLinuxProbe,
2808	/* .pfnInit = */ dbgDiggerLinuxInit,
2809	/* .pfnRefresh = */ dbgDiggerLinuxRefresh,
2810	/* .pfnTerm = */ dbgDiggerLinuxTerm,
2811	/* .pfnQueryVersion = */ dbgDiggerLinuxQueryVersion,
2812	/* .pfnQueryInterface = */ dbgDiggerLinuxQueryInterface,
2813	/* .pfnStackUnwindAssist = */ dbgDiggerLinuxStackUnwindAssist,
2814	/* .u32EndMagic = */ DBGFOSREG_MAGIC
2815	};
2816

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

source: vbox/trunk/src/VBox/Debugger/DBGPlugInLinux.cpp@ 77874

Download in other formats: