DBGPlugInLinux.cpp@ 62837

Last change on this file since 62837 was 62480, checked in by vboxsync, 8 years ago
(C) 2016
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 91.6 KB

Line
1	/* $Id: DBGPlugInLinux.cpp 62480 2016-07-22 18:29:41Z vboxsync $ */
2	/** @file
3	* DBGPlugInLinux - Debugger and Guest OS Digger Plugin For Linux.
4	*/
5
6	/*
7	* Copyright (C) 2008-2016 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/file.h>
28	#include <iprt/string.h>
29	#include <iprt/mem.h>
30	#include <iprt/stream.h>
31	#include <iprt/ctype.h>
32	#include <iprt/vfs.h>
33	#include <iprt/zip.h>
34
35
36	/*********************************************************************************************************************************
37	* Structures and Typedefs *
38	*********************************************************************************************************************************/
39
40	/** @name InternalLinux structures
41	* @{ */
42
43
44	/** @} */
45
46
47	/**
48	* Config item type.
49	*/
50	typedef enum DBGDIGGERLINUXCFGITEMTYPE
51	{
52	/** Invalid type. */
53	DBGDIGGERLINUXCFGITEMTYPE_INVALID = 0,
54	/** String. */
55	DBGDIGGERLINUXCFGITEMTYPE_STRING,
56	/** Number. */
57	DBGDIGGERLINUXCFGITEMTYPE_NUMBER,
58	/** Flag whether this feature is included in the
59	* kernel or as a module. */
60	DBGDIGGERLINUXCFGITEMTYPE_FLAG
61	} DBGDIGGERLINUXCFGITEMTYPE;
62
63	/**
64	* Item in the config database.
65	*/
66	typedef struct DBGDIGGERLINUXCFGITEM
67	{
68	/** String space core. */
69	RTSTRSPACECORE Core;
70	/** Config item type. */
71	DBGDIGGERLINUXCFGITEMTYPE enmType;
72	/** Data based on the type. */
73	union
74	{
75	/** Number. */
76	int64_t i64Num;
77	/** Flag. */
78	bool fModule;
79	/** String - variable in size. */
80	char aszString[1];
81	} u;
82	} DBGDIGGERLINUXCFGITEM;
83	/** Pointer to a config database item. */
84	typedef DBGDIGGERLINUXCFGITEM *PDBGDIGGERLINUXCFGITEM;
85	/** Pointer to a const config database item. */
86	typedef const DBGDIGGERLINUXCFGITEM *PCDBGDIGGERLINUXCFGITEM;
87
88	/**
89	* Linux guest OS digger instance data.
90	*/
91	typedef struct DBGDIGGERLINUX
92	{
93	/** Whether the information is valid or not.
94	* (For fending off illegal interface method calls.) */
95	bool fValid;
96	/** Set if 64-bit, clear if 32-bit. */
97	bool f64Bit;
98
99	/** The address of the linux banner.
100	* This is set during probing. */
101	DBGFADDRESS AddrLinuxBanner;
102	/** Kernel base address.
103	* This is set during probing, refined during kallsyms parsing. */
104	DBGFADDRESS AddrKernelBase;
105	/** The kernel size. */
106	uint32_t cbKernel;
107
108	/** The number of kernel symbols (kallsyms_num_syms).
109	* This is set during init. */
110	uint32_t cKernelSymbols;
111	/** The size of the kernel name table (sizeof(kallsyms_names)). */
112	uint32_t cbKernelNames;
113	/** Number of entries in the kernel_markers table. */
114	uint32_t cKernelNameMarkers;
115	/** The size of the kernel symbol token table. */
116	uint32_t cbKernelTokenTable;
117	/** The address of the encoded kernel symbol names (kallsyms_names). */
118	DBGFADDRESS AddrKernelNames;
119	/** The address of the kernel symbol addresses (kallsyms_addresses). */
120	DBGFADDRESS AddrKernelAddresses;
121	/** The address of the kernel symbol name markers (kallsyms_markers). */
122	DBGFADDRESS AddrKernelNameMarkers;
123	/** The address of the kernel symbol token table (kallsyms_token_table). */
124	DBGFADDRESS AddrKernelTokenTable;
125	/** The address of the kernel symbol token index table (kallsyms_token_index). */
126	DBGFADDRESS AddrKernelTokenIndex;
127
128	/** The kernel message log interface. */
129	DBGFOSIDMESG IDmesg;
130
131	/** The config database root. */
132	RTSTRSPACE hCfgDb;
133	} DBGDIGGERLINUX;
134	/** Pointer to the linux guest OS digger instance data. */
135	typedef DBGDIGGERLINUX *PDBGDIGGERLINUX;
136
137
138	/**
139	* The current printk_log structure.
140	*/
141	typedef struct LNXPRINTKHDR
142	{
143	/** Monotonic timestamp. */
144	uint64_t nsTimestamp;
145	/** The total size of this message record. */
146	uint16_t cbTotal;
147	/** The size of the text part (immediately follows the header). */
148	uint16_t cbText;
149	/** The size of the optional dictionary part (follows the text). */
150	uint16_t cbDict;
151	/** The syslog facility number. */
152	uint8_t bFacility;
153	/** First 5 bits are internal flags, next 3 bits are log level. */
154	uint8_t fFlagsAndLevel;
155	} LNXPRINTKHDR;
156	AssertCompileSize(LNXPRINTKHDR, 2*sizeof(uint64_t));
157	/** Pointer to linux printk_log header. */
158	typedef LNXPRINTKHDR *PLNXPRINTKHDR;
159	/** Pointer to linux const printk_log header. */
160	typedef LNXPRINTKHDR const *PCLNXPRINTKHDR;
161
162
163	/*********************************************************************************************************************************
164	* Defined Constants And Macros *
165	*********************************************************************************************************************************/
166	/** Validates a 32-bit linux kernel address */
167	#define LNX32_VALID_ADDRESS(Addr) ((Addr) > UINT32_C(0x80000000) && (Addr) < UINT32_C(0xfffff000))
168	/** Validates a 64-bit linux kernel address */
169	#define LNX64_VALID_ADDRESS(Addr) ((Addr) > UINT64_C(0xffff800000000000) && (Addr) < UINT64_C(0xfffffffffffff000))
170
171	/** The max kernel size. */
172	#define LNX_MAX_KERNEL_SIZE UINT32_C(0x0f000000)
173
174	/** The maximum size we expect for kallsyms_names. */
175	#define LNX_MAX_KALLSYMS_NAMES_SIZE UINT32_C(0x200000)
176	/** The maximum size we expect for kallsyms_token_table. */
177	#define LNX_MAX_KALLSYMS_TOKEN_TABLE_SIZE UINT32_C(0x10000)
178	/** The minimum number of symbols we expect in kallsyms_num_syms. */
179	#define LNX_MIN_KALLSYMS_SYMBOLS UINT32_C(2048)
180	/** The maximum number of symbols we expect in kallsyms_num_syms. */
181	#define LNX_MAX_KALLSYMS_SYMBOLS UINT32_C(1048576)
182	/** The min length an encoded symbol in kallsyms_names is expected to have. */
183	#define LNX_MIN_KALLSYMS_ENC_LENGTH UINT8_C(1)
184	/** The max length an encoded symbol in kallsyms_names is expected to have.
185	* @todo check real life here. */
186	#define LNX_MAX_KALLSYMS_ENC_LENGTH UINT8_C(28)
187	/** The approximate maximum length of a string token. */
188	#define LNX_MAX_KALLSYMS_TOKEN_LEN UINT16_C(32)
189	/** Maximum compressed config size expected. */
190	#define LNX_MAX_COMPRESSED_CFG_SIZE _1M
191
192	/** Module tag for linux ('linuxmod' on little endian ASCII systems). */
193	#define DIG_LNX_MOD_TAG UINT64_C(0x545f5d78758e898c)
194
195
196	/*********************************************************************************************************************************
197	* Internal Functions *
198	*********************************************************************************************************************************/
199	static DECLCALLBACK(int) dbgDiggerLinuxInit(PUVM pUVM, void *pvData);
200
201
202	/*********************************************************************************************************************************
203	* Global Variables *
204	*********************************************************************************************************************************/
205	/** Table of common linux kernel addresses. */
206	static uint64_t g_au64LnxKernelAddresses[] =
207	{
208	UINT64_C(0xc0100000),
209	UINT64_C(0x90100000),
210	UINT64_C(0xffffffff80200000)
211	};
212
213	static const uint8_t g_abLinuxVersion[] = "Linux version ";
214
215	/**
216	* Disassembles a simple getter returning the value for it.
217	*
218	* @returns VBox status code.
219	* @param pThis The Linux digger data.
220	* @param pUVM The VM handle.
221	* @param hMod The module to use.
222	* @param pszSymbol The symbol of the getter.
223	* @param pvVal Where to store the value on success.
224	* @param cbVal Size of the value in bytes.
225	*/
226	static int dbgDiggerLinuxDisassembleSimpleGetter(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
227	const char pszSymbol, void pvVal, uint32_t cbVal)
228	{
229	int rc = VINF_SUCCESS;
230
231	RTDBGSYMBOL SymInfo;
232	rc = RTDbgModSymbolByName(hMod, pszSymbol, &SymInfo);
233	if (RT_SUCCESS(rc))
234	{
235	/*
236	* Do the diassembling. Disassemble until a ret instruction is encountered
237	* or a limit is reached (don't want to disassemble for too long as the getter
238	* should be short).
239	* push and pop instructions are skipped as well as any mov instructions not
240	* touching the rax or eax register (depending on the size of the value).
241	*/
242	unsigned cInstrDisassembled = 0;
243	uint32_t offInstr = 0;
244	bool fRet = false;
245	DISSTATE DisState;
246	RT_ZERO(DisState);
247
248	do
249	{
250	DBGFADDRESS Addr;
251	RTGCPTR GCPtrCur = (RTGCPTR)SymInfo.Value + pThis->AddrKernelBase.FlatPtr + offInstr;
252	DBGFR3AddrFromFlat(pUVM, &Addr, GCPtrCur);
253
254	/* Prefetch the instruction. */
255	uint8_t abInstr[32];
256	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &Addr, &abInstr[0], sizeof(abInstr));
257	if (RT_SUCCESS(rc))
258	{
259	uint32_t cbInstr = 0;
260
261	rc = DISInstr(&abInstr[0], pThis->f64Bit ? DISCPUMODE_64BIT : DISCPUMODE_32BIT, &DisState, &cbInstr);
262	if (RT_SUCCESS(rc))
263	{
264	switch (DisState.pCurInstr->uOpcode)
265	{
266	case OP_PUSH:
267	case OP_POP:
268	case OP_NOP:
269	case OP_LEA:
270	break;
271	case OP_RETN:
272	/* Getter returned, abort disassembling. */
273	fRet = true;
274	break;
275	case OP_MOV:
276	/*
277	* Check that the destination is either rax or eax depending on the
278	* value size.
279	*
280	* Param1 is the destination and Param2 the source.
281	*/
282	if ( ( ( (DisState.Param1.fUse & (DISUSE_BASE \| DISUSE_REG_GEN32))
283	&& cbVal == sizeof(uint32_t))
284	\|\| ( (DisState.Param1.fUse & (DISUSE_BASE \| DISUSE_REG_GEN64))
285	&& cbVal == sizeof(uint64_t)))
286	&& DisState.Param1.Base.idxGenReg == DISGREG_RAX)
287	{
288	/* Parse the source. */
289	if (DisState.Param2.fUse & (DISUSE_IMMEDIATE32 \| DISUSE_IMMEDIATE64))
290	memcpy(pvVal, &DisState.Param2.uValue, cbVal);
291	else if (DisState.Param2.fUse & (DISUSE_RIPDISPLACEMENT32\|DISUSE_DISPLACEMENT32\|DISUSE_DISPLACEMENT64))
292	{
293	RTGCPTR GCPtrVal = 0;
294
295	if (DisState.Param2.fUse & DISUSE_RIPDISPLACEMENT32)
296	GCPtrVal = GCPtrCur + DisState.Param2.uDisp.i32 + cbInstr;
297	else if (DisState.Param2.fUse & DISUSE_DISPLACEMENT32)
298	GCPtrVal = (RTGCPTR)DisState.Param2.uDisp.u32;
299	else if (DisState.Param2.fUse & DISUSE_DISPLACEMENT64)
300	GCPtrVal = (RTGCPTR)DisState.Param2.uDisp.u64;
301	else
302	AssertMsgFailedBreakStmt(("Invalid displacement\n"), rc = VERR_INVALID_STATE);
303
304	DBGFADDRESS AddrVal;
305	rc = DBGFR3MemRead(pUVM, 0 /idCpu/,
306	DBGFR3AddrFromFlat(pUVM, &AddrVal, GCPtrVal),
307	pvVal, cbVal);
308	}
309	}
310	break;
311	default:
312	/* All other instructions will cause an error for now (playing safe here). */
313	rc = VERR_INVALID_PARAMETER;
314	break;
315	}
316	cInstrDisassembled++;
317	offInstr += cbInstr;
318	}
319	}
320	} while ( RT_SUCCESS(rc)
321	&& cInstrDisassembled < 20
322	&& !fRet);
323	}
324
325	return rc;
326	}
327
328	/**
329	* Try to get at the log buffer starting address and size by disassembling emit_log_char.
330	*
331	* @returns VBox status code.
332	* @param pThis The Linux digger data.
333	* @param pUVM The VM handle.
334	* @param hMod The module to use.
335	* @param pGCPtrLogBuf Where to store the log buffer pointer on success.
336	* @param pcbLogBuf Where to store the size of the log buffer on success.
337	*/
338	static int dbgDiggerLinuxQueryAsciiLogBufferPtrs(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
339	RTGCPTR pGCPtrLogBuf, uint32_t pcbLogBuf)
340	{
341	int rc = VINF_SUCCESS;
342
343	/**
344	* We disassemble emit_log_char to get at the log buffer address and size.
345	* This is used in case the symbols are not exported in kallsyms.
346	*
347	* This is what it typically looks like:
348	* vmlinux!emit_log_char:
349	* %00000000c01204a1 56 push esi
350	* %00000000c01204a2 8b 35 d0 1c 34 c0 mov esi, dword [0c0341cd0h]
351	* %00000000c01204a8 53 push ebx
352	* %00000000c01204a9 8b 1d 74 3b 3e c0 mov ebx, dword [0c03e3b74h]
353	* %00000000c01204af 8b 0d d8 1c 34 c0 mov ecx, dword [0c0341cd8h]
354	* %00000000c01204b5 8d 56 ff lea edx, [esi-001h]
355	* %00000000c01204b8 21 da and edx, ebx
356	* %00000000c01204ba 88 04 11 mov byte [ecx+edx], al
357	* %00000000c01204bd 8d 53 01 lea edx, [ebx+001h]
358	* %00000000c01204c0 89 d0 mov eax, edx
359	* [...]
360	*/
361	RTDBGSYMBOL SymInfo;
362	rc = RTDbgModSymbolByName(hMod, "emit_log_char", &SymInfo);
363	if (RT_SUCCESS(rc))
364	{
365	/*
366	* Do the diassembling. Disassemble until a ret instruction is encountered
367	* or a limit is reached (don't want to disassemble for too long as the getter
368	* should be short). Certain instructions found are ignored (push, nop, etc.).
369	*/
370	unsigned cInstrDisassembled = 0;
371	uint32_t offInstr = 0;
372	bool fRet = false;
373	DISSTATE DisState;
374	unsigned idxAddressesUsed = 0;
375	struct { size_t cb; RTGCPTR GCPtrOrigSrc; } aAddresses[5];
376	RT_ZERO(DisState);
377	RT_ZERO(aAddresses);
378
379	do
380	{
381	DBGFADDRESS Addr;
382	RTGCPTR GCPtrCur = (RTGCPTR)SymInfo.Value + pThis->AddrKernelBase.FlatPtr + offInstr;
383	DBGFR3AddrFromFlat(pUVM, &Addr, GCPtrCur);
384
385	/* Prefetch the instruction. */
386	uint8_t abInstr[32];
387	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &Addr, &abInstr[0], sizeof(abInstr));
388	if (RT_SUCCESS(rc))
389	{
390	uint32_t cbInstr = 0;
391
392	rc = DISInstr(&abInstr[0], pThis->f64Bit ? DISCPUMODE_64BIT : DISCPUMODE_32BIT, &DisState, &cbInstr);
393	if (RT_SUCCESS(rc))
394	{
395	switch (DisState.pCurInstr->uOpcode)
396	{
397	case OP_PUSH:
398	case OP_POP:
399	case OP_NOP:
400	case OP_LEA:
401	case OP_AND:
402	case OP_CBW:
403	break;
404	case OP_RETN:
405	/* emit_log_char returned, abort disassembling. */
406	rc = VERR_NOT_FOUND;
407	fRet = true;
408	break;
409	case OP_MOV:
410	case OP_MOVSXD:
411	/*
412	* If a mov is encountered writing to memory with al (or dil for amd64) being the source the
413	* character is stored and we can infer the base address and size of the log buffer from
414	* the source addresses.
415	*/
416	if ( (DisState.Param2.fUse & DISUSE_REG_GEN8)
417	&& ( (DisState.Param2.Base.idxGenReg == DISGREG_AL && !pThis->f64Bit)
418	\|\| (DisState.Param2.Base.idxGenReg == DISGREG_DIL && pThis->f64Bit))
419	&& DISUSE_IS_EFFECTIVE_ADDR(DisState.Param1.fUse))
420	{
421	RTGCPTR GCPtrLogBuf = 0;
422	size_t cbLogBuf = 0;
423
424	/*
425	* We can stop disassembling now and inspect all registers, look for a valid kernel address first.
426	* Only one of the accessed registers should hold a valid kernel address.
427	* For the log size look for the biggest non kernel address.
428	*/
429	for (unsigned i = 0; i < idxAddressesUsed; i++)
430	{
431	DBGFADDRESS AddrVal;
432	union { uint8_t abVal[8]; uint32_t u32Val; uint64_t u64Val; } Val;
433
434	rc = DBGFR3MemRead(pUVM, 0 /idCpu/,
435	DBGFR3AddrFromFlat(pUVM, &AddrVal, aAddresses[i].GCPtrOrigSrc),
436	&Val.abVal[0], aAddresses[i].cb);
437	if (RT_SUCCESS(rc))
438	{
439	if (pThis->f64Bit && aAddresses[i].cb == sizeof(uint64_t))
440	{
441	if (LNX64_VALID_ADDRESS(Val.u64Val))
442	{
443	if (GCPtrLogBuf == 0)
444	GCPtrLogBuf = Val.u64Val;
445	else
446	{
447	rc = VERR_NOT_FOUND;
448	break;
449	}
450	}
451	}
452	else
453	{
454	AssertMsgBreakStmt(aAddresses[i].cb == sizeof(uint32_t),
455	("Invalid value size\n"), rc = VERR_INVALID_STATE);
456
457	/* Might be a kernel address or a size indicator. */
458	if (!pThis->f64Bit && LNX32_VALID_ADDRESS(Val.u32Val))
459	{
460	if (GCPtrLogBuf == 0)
461	GCPtrLogBuf = Val.u32Val;
462	else
463	{
464	rc = VERR_NOT_FOUND;
465	break;
466	}
467	}
468	else
469	{
470	/*
471	* The highest value will be the log buffer because the other
472	* accessed variables are indexes into the buffer and hence
473	* always smaller than the size.
474	*/
475	if (cbLogBuf < Val.u32Val)
476	cbLogBuf = Val.u32Val;
477	}
478	}
479	}
480	}
481
482	if ( RT_SUCCESS(rc)
483	&& GCPtrLogBuf != 0
484	&& cbLogBuf != 0)
485	{
486	*pGCPtrLogBuf = GCPtrLogBuf;
487	*pcbLogBuf = cbLogBuf;
488	}
489	else if (RT_SUCCESS(rc))
490	rc = VERR_NOT_FOUND;
491
492	fRet = true;
493	break;
494	}
495	else
496	{
497	/*
498	* In case of a memory to register move store the destination register index and the
499	* source address in the relation table for later processing.
500	*/
501	if ( (DisState.Param1.fUse & (DISUSE_BASE \| DISUSE_REG_GEN32 \| DISUSE_REG_GEN64))
502	&& (DisState.Param2.cb == sizeof(uint32_t) \|\| DisState.Param2.cb == sizeof(uint64_t))
503	&& (DisState.Param2.fUse & (DISUSE_RIPDISPLACEMENT32\|DISUSE_DISPLACEMENT32\|DISUSE_DISPLACEMENT64)))
504	{
505	RTGCPTR GCPtrVal = 0;
506
507	if (DisState.Param2.fUse & DISUSE_RIPDISPLACEMENT32)
508	GCPtrVal = GCPtrCur + DisState.Param2.uDisp.i32 + cbInstr;
509	else if (DisState.Param2.fUse & DISUSE_DISPLACEMENT32)
510	GCPtrVal = (RTGCPTR)DisState.Param2.uDisp.u32;
511	else if (DisState.Param2.fUse & DISUSE_DISPLACEMENT64)
512	GCPtrVal = (RTGCPTR)DisState.Param2.uDisp.u64;
513	else
514	AssertMsgFailedBreakStmt(("Invalid displacement\n"), rc = VERR_INVALID_STATE);
515
516	if (idxAddressesUsed < RT_ELEMENTS(aAddresses))
517	{
518	/* movsxd reads always 32bits. */
519	if (DisState.pCurInstr->uOpcode == OP_MOVSXD)
520	aAddresses[idxAddressesUsed].cb = sizeof(uint32_t);
521	else
522	aAddresses[idxAddressesUsed].cb = DisState.Param2.cb;
523	aAddresses[idxAddressesUsed].GCPtrOrigSrc = GCPtrVal;
524	idxAddressesUsed++;
525	}
526	else
527	{
528	rc = VERR_INVALID_PARAMETER;
529	break;
530	}
531	}
532	}
533	break;
534	default:
535	/* All other instructions will cause an error for now (playing safe here). */
536	rc = VERR_INVALID_PARAMETER;
537	break;
538	}
539	cInstrDisassembled++;
540	offInstr += cbInstr;
541	}
542	}
543	} while ( RT_SUCCESS(rc)
544	&& cInstrDisassembled < 20
545	&& !fRet);
546	}
547
548	return rc;
549	}
550
551	/**
552	* Try to get at the log buffer starting address and size by disassembling some exposed helpers.
553	*
554	* @returns VBox status code.
555	* @param pThis The Linux digger data.
556	* @param pUVM The VM handle.
557	* @param hMod The module to use.
558	* @param pGCPtrLogBuf Where to store the log buffer pointer on success.
559	* @param pcbLogBuf Where to store the size of the log buffer on success.
560	*/
561	static int dbgDiggerLinuxQueryLogBufferPtrs(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
562	RTGCPTR pGCPtrLogBuf, uint32_t pcbLogBuf)
563	{
564	int rc = VINF_SUCCESS;
565
566	struct { void pvVar; size_t cbHost, cbGuest; const char pszSymbol; } aSymbols[] =
567	{
568	{ pGCPtrLogBuf, sizeof(RTGCPTR), pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t), "log_buf_addr_get" },
569	{ pcbLogBuf, sizeof(uint32_t), sizeof(uint32_t), "log_buf_len_get" }
570	};
571	for (uint32_t i = 0; i < RT_ELEMENTS(aSymbols) && RT_SUCCESS(rc); i++)
572	{
573	RT_BZERO(aSymbols[i].pvVar, aSymbols[i].cbHost);
574	Assert(aSymbols[i].cbHost >= aSymbols[i].cbGuest);
575	rc = dbgDiggerLinuxDisassembleSimpleGetter(pThis, pUVM, hMod, aSymbols[i].pszSymbol,
576	aSymbols[i].pvVar, aSymbols[i].cbGuest);
577	}
578
579	return rc;
580	}
581
582	/**
583	* Returns whether the log buffer is a simple ascii buffer or a record based implementation
584	* based on the kernel version found.
585	*
586	* @returns Flag whether the log buffer is the simple ascii buffer.
587	* @param pThis The Linux digger data.
588	* @param pUVM The user mode VM handle.
589	*/
590	static bool dbgDiggerLinuxLogBufferIsAsciiBuffer(PDBGDIGGERLINUX pThis, PUVM pUVM)
591	{
592	char szTmp[128];
593	char const *pszVer = &szTmp[sizeof(g_abLinuxVersion) - 1];
594
595	RT_ZERO(szTmp);
596	int rc = DBGFR3MemReadString(pUVM, 0, &pThis->AddrLinuxBanner, szTmp, sizeof(szTmp) - 1);
597	if ( RT_SUCCESS(rc)
598	&& RTStrVersionCompare(pszVer, "3.4") == -1)
599	return true;
600
601	return false;
602	}
603
604	/**
605	* Worker to get at the kernel log for pre 3.4 kernels where the log buffer was just a char buffer.
606	*
607	* @returns VBox status code.
608	* @param pThis The Linux digger data.
609	* @param pUVM The VM user mdoe handle.
610	* @param hMod The debug module handle.
611	* @param fFlags Flags reserved for future use, MBZ.
612	* @param cMessages The number of messages to retrieve, counting from the
613	* end of the log (i.e. like tail), use UINT32_MAX for all.
614	* @param pszBuf The output buffer.
615	* @param cbBuf The buffer size.
616	* @param pcbActual Where to store the number of bytes actually returned,
617	* including zero terminator. On VERR_BUFFER_OVERFLOW this
618	* holds the necessary buffer size. Optional.
619	*/
620	static int dbgDiggerLinuxLogBufferQueryAscii(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
621	uint32_t fFlags, uint32_t cMessages,
622	char pszBuf, size_t cbBuf, size_t pcbActual)
623	{
624	int rc = VINF_SUCCESS;
625	RTGCPTR GCPtrLogBuf;
626	uint32_t cbLogBuf;
627
628	struct { void pvVar; size_t cbHost, cbGuest; const char pszSymbol; } aSymbols[] =
629	{
630	{ &GCPtrLogBuf, sizeof(GCPtrLogBuf), pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t), "log_buf" },
631	{ &cbLogBuf, sizeof(cbLogBuf), sizeof(cbLogBuf), "log_buf_len" },
632	};
633	for (uint32_t i = 0; i < RT_ELEMENTS(aSymbols); i++)
634	{
635	RTDBGSYMBOL SymInfo;
636	rc = RTDbgModSymbolByName(hMod, aSymbols[i].pszSymbol, &SymInfo);
637	if (RT_SUCCESS(rc))
638	{
639	RT_BZERO(aSymbols[i].pvVar, aSymbols[i].cbHost);
640	Assert(aSymbols[i].cbHost >= aSymbols[i].cbGuest);
641	DBGFADDRESS Addr;
642	rc = DBGFR3MemRead(pUVM, 0 /idCpu/,
643	DBGFR3AddrFromFlat(pUVM, &Addr, (RTGCPTR)SymInfo.Value + pThis->AddrKernelBase.FlatPtr),
644	aSymbols[i].pvVar, aSymbols[i].cbGuest);
645	if (RT_SUCCESS(rc))
646	continue;
647	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: Reading '%s' at %RGv: %Rrc\n", aSymbols[i].pszSymbol, Addr.FlatPtr, rc));
648	}
649	else
650	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: Error looking up '%s': %Rrc\n", aSymbols[i].pszSymbol, rc));
651	rc = VERR_NOT_FOUND;
652	break;
653	}
654
655	/*
656	* Some kernels don't expose the variables in kallsyms so we have to try disassemble
657	* some public helpers to get at the addresses.
658	*
659	* @todo: Maybe cache those values so we don't have to do the heavy work every time?
660	*/
661	if (rc == VERR_NOT_FOUND)
662	{
663	rc = dbgDiggerLinuxQueryAsciiLogBufferPtrs(pThis, pUVM, hMod, &GCPtrLogBuf, &cbLogBuf);
664	if (RT_FAILURE(rc))
665	return rc;
666	}
667
668	/*
669	* Check if the values make sense.
670	*/
671	if (pThis->f64Bit ? !LNX64_VALID_ADDRESS(GCPtrLogBuf) : !LNX32_VALID_ADDRESS(GCPtrLogBuf))
672	{
673	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_buf' value %RGv is not valid.\n", GCPtrLogBuf));
674	return VERR_NOT_FOUND;
675	}
676	if ( cbLogBuf < 4096
677	\|\| !RT_IS_POWER_OF_TWO(cbLogBuf)
678	\|\| cbLogBuf > 16*_1M)
679	{
680	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_buf_len' value %#x is not valid.\n", cbLogBuf));
681	return VERR_NOT_FOUND;
682	}
683
684	/*
685	* Read the whole log buffer.
686	*/
687	uint8_t pbLogBuf = (uint8_t )RTMemAlloc(cbLogBuf);
688	if (!pbLogBuf)
689	{
690	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: Failed to allocate %#x bytes for log buffer\n", cbLogBuf));
691	return VERR_NO_MEMORY;
692	}
693	DBGFADDRESS Addr;
694	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, DBGFR3AddrFromFlat(pUVM, &Addr, GCPtrLogBuf), pbLogBuf, cbLogBuf);
695	if (RT_FAILURE(rc))
696	{
697	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: Error reading %#x bytes of log buffer at %RGv: %Rrc\n",
698	cbLogBuf, Addr.FlatPtr, rc));
699	RTMemFree(pbLogBuf);
700	return VERR_NOT_FOUND;
701	}
702
703	/** @todo: Try to parse where the single messages start to make use of cMessages. */
704	size_t cchLength = RTStrNLen((const char *)pbLogBuf, cbLogBuf);
705	memcpy(&pszBuf[0], pbLogBuf, RT_MIN(cbBuf, cchLength));
706
707	/* Done with the buffer. */
708	RTMemFree(pbLogBuf);
709
710	/* Set return size value. */
711	if (pcbActual)
712	*pcbActual = RT_MIN(cbBuf, cchLength);
713
714	return cbBuf <= cchLength ? VERR_BUFFER_OVERFLOW : VINF_SUCCESS;
715	}
716
717	/**
718	* Worker to get at the kernel log for post 3.4 kernels where the log buffer contains records.
719	*
720	* @returns VBox status code.
721	* @param pThis The Linux digger data.
722	* @param pUVM The VM user mdoe handle.
723	* @param hMod The debug module handle.
724	* @param fFlags Flags reserved for future use, MBZ.
725	* @param cMessages The number of messages to retrieve, counting from the
726	* end of the log (i.e. like tail), use UINT32_MAX for all.
727	* @param pszBuf The output buffer.
728	* @param cbBuf The buffer size.
729	* @param pcbActual Where to store the number of bytes actually returned,
730	* including zero terminator. On VERR_BUFFER_OVERFLOW this
731	* holds the necessary buffer size. Optional.
732	*/
733	static int dbgDiggerLinuxLogBufferQueryRecords(PDBGDIGGERLINUX pThis, PUVM pUVM, RTDBGMOD hMod,
734	uint32_t fFlags, uint32_t cMessages,
735	char pszBuf, size_t cbBuf, size_t pcbActual)
736	{
737	int rc = VINF_SUCCESS;
738	RTGCPTR GCPtrLogBuf;
739	uint32_t cbLogBuf;
740	uint32_t idxFirst;
741	uint32_t idxNext;
742
743	struct { void pvVar; size_t cbHost, cbGuest; const char pszSymbol; } aSymbols[] =
744	{
745	{ &GCPtrLogBuf, sizeof(GCPtrLogBuf), pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t), "log_buf" },
746	{ &cbLogBuf, sizeof(cbLogBuf), sizeof(cbLogBuf), "log_buf_len" },
747	{ &idxFirst, sizeof(idxFirst), sizeof(idxFirst), "log_first_idx" },
748	{ &idxNext, sizeof(idxNext), sizeof(idxNext), "log_next_idx" },
749	};
750	for (uint32_t i = 0; i < RT_ELEMENTS(aSymbols); i++)
751	{
752	RTDBGSYMBOL SymInfo;
753	rc = RTDbgModSymbolByName(hMod, aSymbols[i].pszSymbol, &SymInfo);
754	if (RT_SUCCESS(rc))
755	{
756	RT_BZERO(aSymbols[i].pvVar, aSymbols[i].cbHost);
757	Assert(aSymbols[i].cbHost >= aSymbols[i].cbGuest);
758	DBGFADDRESS Addr;
759	rc = DBGFR3MemRead(pUVM, 0 /idCpu/,
760	DBGFR3AddrFromFlat(pUVM, &Addr, (RTGCPTR)SymInfo.Value + pThis->AddrKernelBase.FlatPtr),
761	aSymbols[i].pvVar, aSymbols[i].cbGuest);
762	if (RT_SUCCESS(rc))
763	continue;
764	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: Reading '%s' at %RGv: %Rrc\n", aSymbols[i].pszSymbol, Addr.FlatPtr, rc));
765	}
766	else
767	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: Error looking up '%s': %Rrc\n", aSymbols[i].pszSymbol, rc));
768	rc = VERR_NOT_FOUND;
769	break;
770	}
771
772	/*
773	* Some kernels don't expose the variables in kallsyms so we have to try disassemble
774	* some public helpers to get at the addresses.
775	*
776	* @todo: Maybe cache those values so we don't have to do the heavy work every time?
777	*/
778	if (rc == VERR_NOT_FOUND)
779	{
780	idxFirst = 0;
781	idxNext = 0;
782	rc = dbgDiggerLinuxQueryLogBufferPtrs(pThis, pUVM, hMod, &GCPtrLogBuf, &cbLogBuf);
783	if (RT_FAILURE(rc))
784	return rc;
785	}
786
787	/*
788	* Check if the values make sense.
789	*/
790	if (pThis->f64Bit ? !LNX64_VALID_ADDRESS(GCPtrLogBuf) : !LNX32_VALID_ADDRESS(GCPtrLogBuf))
791	{
792	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_buf' value %RGv is not valid.\n", GCPtrLogBuf));
793	return VERR_NOT_FOUND;
794	}
795	if ( cbLogBuf < 4096
796	\|\| !RT_IS_POWER_OF_TWO(cbLogBuf)
797	\|\| cbLogBuf > 16*_1M)
798	{
799	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_buf_len' value %#x is not valid.\n", cbLogBuf));
800	return VERR_NOT_FOUND;
801	}
802	uint32_t const cbLogAlign = 4;
803	if ( idxFirst > cbLogBuf - sizeof(LNXPRINTKHDR)
804	\|\| (idxFirst & (cbLogAlign - 1)) != 0)
805	{
806	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_first_idx' value %#x is not valid.\n", idxFirst));
807	return VERR_NOT_FOUND;
808	}
809	if ( idxNext > cbLogBuf - sizeof(LNXPRINTKHDR)
810	\|\| (idxNext & (cbLogAlign - 1)) != 0)
811	{
812	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: 'log_next_idx' value %#x is not valid.\n", idxNext));
813	return VERR_NOT_FOUND;
814	}
815
816	/*
817	* Read the whole log buffer.
818	*/
819	uint8_t pbLogBuf = (uint8_t )RTMemAlloc(cbLogBuf);
820	if (!pbLogBuf)
821	{
822	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: Failed to allocate %#x bytes for log buffer\n", cbLogBuf));
823	return VERR_NO_MEMORY;
824	}
825	DBGFADDRESS Addr;
826	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, DBGFR3AddrFromFlat(pUVM, &Addr, GCPtrLogBuf), pbLogBuf, cbLogBuf);
827	if (RT_FAILURE(rc))
828	{
829	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: Error reading %#x bytes of log buffer at %RGv: %Rrc\n",
830	cbLogBuf, Addr.FlatPtr, rc));
831	RTMemFree(pbLogBuf);
832	return VERR_NOT_FOUND;
833	}
834
835	/*
836	* Count the messages in the buffer while doing some basic validation.
837	*/
838	uint32_t const cbUsed = idxFirst == idxNext ? cbLogBuf /* could be empty... */
839	: idxFirst < idxNext ? idxNext - idxFirst : cbLogBuf - idxFirst + idxNext;
840	uint32_t cbLeft = cbUsed;
841	uint32_t offCur = idxFirst;
842	uint32_t cLogMsgs = 0;
843
844	while (cbLeft > 0)
845	{
846	PCLNXPRINTKHDR pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
847	if (!pHdr->cbTotal)
848	{
849	/* Wrap around packet, most likely... */
850	if (cbLogBuf - offCur >= cbLeft)
851	break;
852	offCur = 0;
853	pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
854	}
855	if (RT_UNLIKELY( pHdr->cbTotal > cbLogBuf - sizeof(*pHdr) - offCur
856	\|\| pHdr->cbTotal > cbLeft
857	\|\| (pHdr->cbTotal & (cbLogAlign - 1)) != 0
858	\|\| pHdr->cbTotal < (uint32_t)pHdr->cbText + (uint32_t)pHdr->cbDict + sizeof(*pHdr) ))
859	{
860	Log(("dbgDiggerLinuxIDmsg_QueryKernelLog: Invalid printk_log record at %#x: cbTotal=%#x cbText=%#x cbDict=%#x cbLogBuf=%#x cbLeft=%#x\n",
861	offCur, pHdr->cbTotal, pHdr->cbText, pHdr->cbDict, cbLogBuf, cbLeft));
862	rc = VERR_INVALID_STATE;
863	break;
864	}
865
866	if (pHdr->cbText > 0)
867	cLogMsgs++;
868
869	/* next */
870	offCur += pHdr->cbTotal;
871	cbLeft -= pHdr->cbTotal;
872	}
873	if (RT_FAILURE(rc))
874	{
875	RTMemFree(pbLogBuf);
876	return rc;
877	}
878
879	/*
880	* Copy the messages into the output buffer.
881	*/
882	offCur = idxFirst;
883	cbLeft = cbUsed;
884
885	/* Skip messages that the caller doesn't want. */
886	if (cMessages < cLogMsgs)
887	{
888	uint32_t cToSkip = cLogMsgs - cMessages;
889	while (cToSkip > 0)
890	{
891	PCLNXPRINTKHDR pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
892	if (!pHdr->cbTotal)
893	{
894	offCur = 0;
895	pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
896	}
897	if (pHdr->cbText > 0)
898	cToSkip--;
899
900	/* next */
901	offCur += pHdr->cbTotal;
902	cbLeft -= pHdr->cbTotal;
903	}
904	}
905
906	/* Now copy the messages. */
907	size_t offDst = 0;
908	while (cbLeft > 0)
909	{
910	PCLNXPRINTKHDR pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
911	if (!pHdr->cbTotal)
912	{
913	if (cbLogBuf - offCur >= cbLeft)
914	break;
915	offCur = 0;
916	pHdr = (PCLNXPRINTKHDR)&pbLogBuf[offCur];
917	}
918
919	if (pHdr->cbText > 0)
920	{
921	char pchText = (char )(pHdr + 1);
922	size_t cchText = RTStrNLen(pchText, pHdr->cbText);
923	if (offDst + cchText < cbBuf)
924	{
925	memcpy(&pszBuf[offDst], pHdr + 1, cchText);
926	pszBuf[offDst + cchText] = '\n';
927	}
928	else if (offDst < cbBuf)
929	memcpy(&pszBuf[offDst], pHdr + 1, cbBuf - offDst);
930	offDst += cchText + 1;
931	}
932
933	/* next */
934	offCur += pHdr->cbTotal;
935	cbLeft -= pHdr->cbTotal;
936	}
937
938	/* Done with the buffer. */
939	RTMemFree(pbLogBuf);
940
941	/* Make sure we've reserved a char for the terminator. */
942	if (!offDst)
943	offDst = 1;
944
945	/* Set return size value. */
946	if (pcbActual)
947	*pcbActual = offDst;
948
949	if (offDst <= cbBuf)
950	return VINF_SUCCESS;
951	else
952	return VERR_BUFFER_OVERFLOW;
953	}
954
955	/**
956	* @interface_method_impl{DBGFOSIDMESG,pfnQueryKernelLog}
957	*/
958	static DECLCALLBACK(int) dbgDiggerLinuxIDmsg_QueryKernelLog(PDBGFOSIDMESG pThis, PUVM pUVM, uint32_t fFlags, uint32_t cMessages,
959	char pszBuf, size_t cbBuf, size_t pcbActual)
960	{
961	PDBGDIGGERLINUX pData = RT_FROM_MEMBER(pThis, DBGDIGGERLINUX, IDmesg);
962
963	if (cMessages < 1)
964	return VERR_INVALID_PARAMETER;
965
966	/*
967	* Resolve the symbols we need and read their values.
968	*/
969	RTDBGAS hAs = DBGFR3AsResolveAndRetain(pUVM, DBGF_AS_KERNEL);
970	RTDBGMOD hMod;
971	int rc = RTDbgAsModuleByName(hAs, "vmlinux", 0, &hMod);
972	if (RT_FAILURE(rc))
973	return VERR_NOT_FOUND;
974	RTDbgAsRelease(hAs);
975
976	size_t cbActual;
977	/*
978	* Check whether the kernel log buffer is a simple char buffer or the newer
979	* record based implementation.
980	* The record based implementation was presumably introduced with kernel 3.4,
981	* see: http://thread.gmane.org/gmane.linux.kernel/1284184
982	*/
983	if (dbgDiggerLinuxLogBufferIsAsciiBuffer(pData, pUVM))
984	rc = dbgDiggerLinuxLogBufferQueryAscii(pData, pUVM, hMod, fFlags, cMessages, pszBuf, cbBuf, &cbActual);
985	else
986	rc = dbgDiggerLinuxLogBufferQueryRecords(pData, pUVM, hMod, fFlags, cMessages, pszBuf, cbBuf, &cbActual);
987
988	/* Release the module in any case. */
989	RTDbgModRelease(hMod);
990
991	if (RT_FAILURE(rc) && rc != VERR_BUFFER_OVERFLOW)
992	return rc;
993
994	if (pcbActual)
995	*pcbActual = cbActual;
996
997	/*
998	* All VBox strings are UTF-8 and bad things may in theory happen if we
999	* pass bad UTF-8 to code which assumes it's all valid. So, we enforce
1000	* UTF-8 upon the guest kernel messages here even if they (probably) have
1001	* no defined code set in reality.
1002	*/
1003	if ( RT_SUCCESS(rc)
1004	&& cbActual <= cbBuf)
1005	{
1006	pszBuf[cbActual - 1] = '\0';
1007	RTStrPurgeEncoding(pszBuf);
1008	return VINF_SUCCESS;
1009	}
1010
1011	if (cbBuf)
1012	{
1013	pszBuf[cbBuf - 1] = '\0';
1014	RTStrPurgeEncoding(pszBuf);
1015	}
1016	return VERR_BUFFER_OVERFLOW;
1017	}
1018
1019
1020	/**
1021	* Worker destroying the config database.
1022	*/
1023	static DECLCALLBACK(int) dbgDiggerLinuxCfgDbDestroyWorker(PRTSTRSPACECORE pStr, void *pvUser)
1024	{
1025	PDBGDIGGERLINUXCFGITEM pCfgItem = (PDBGDIGGERLINUXCFGITEM)pStr;
1026	RTStrFree((char *)pCfgItem->Core.pszString);
1027	RTMemFree(pCfgItem);
1028	NOREF(pvUser);
1029	return 0;
1030	}
1031
1032
1033	/**
1034	* Destroy the config database.
1035	*
1036	* @returns nothing.
1037	* @param pThis The Linux digger data.
1038	*/
1039	static void dbgDiggerLinuxCfgDbDestroy(PDBGDIGGERLINUX pThis)
1040	{
1041	RTStrSpaceDestroy(&pThis->hCfgDb, dbgDiggerLinuxCfgDbDestroyWorker, NULL);
1042	}
1043
1044
1045	/**
1046	* @copydoc DBGFOSREG::pfnQueryInterface
1047	*/
1048	static DECLCALLBACK(void ) dbgDiggerLinuxQueryInterface(PUVM pUVM, void pvData, DBGFOSINTERFACE enmIf)
1049	{
1050	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
1051	switch (enmIf)
1052	{
1053	case DBGFOSINTERFACE_DMESG:
1054	return &pThis->IDmesg;
1055
1056	default:
1057	return NULL;
1058	}
1059	}
1060
1061
1062	/**
1063	* @copydoc DBGFOSREG::pfnQueryVersion
1064	*/
1065	static DECLCALLBACK(int) dbgDiggerLinuxQueryVersion(PUVM pUVM, void pvData, char pszVersion, size_t cchVersion)
1066	{
1067	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
1068	Assert(pThis->fValid);
1069
1070	/*
1071	* It's all in the linux banner.
1072	*/
1073	int rc = DBGFR3MemReadString(pUVM, 0, &pThis->AddrLinuxBanner, pszVersion, cchVersion);
1074	if (RT_SUCCESS(rc))
1075	{
1076	char *pszEnd = RTStrEnd(pszVersion, cchVersion);
1077	AssertReturn(pszEnd, VERR_BUFFER_OVERFLOW);
1078	while ( pszEnd > pszVersion
1079	&& RT_C_IS_SPACE(pszEnd[-1]))
1080	pszEnd--;
1081	*pszEnd = '\0';
1082	}
1083	else
1084	RTStrPrintf(pszVersion, cchVersion, "DBGFR3MemRead -> %Rrc", rc);
1085
1086	return rc;
1087	}
1088
1089
1090	/**
1091	* @copydoc DBGFOSREG::pfnTerm
1092	*/
1093	static DECLCALLBACK(void) dbgDiggerLinuxTerm(PUVM pUVM, void *pvData)
1094	{
1095	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
1096	Assert(pThis->fValid);
1097
1098	dbgDiggerLinuxCfgDbDestroy(pThis);
1099	pThis->fValid = false;
1100	}
1101
1102
1103	/**
1104	* @copydoc DBGFOSREG::pfnRefresh
1105	*/
1106	static DECLCALLBACK(int) dbgDiggerLinuxRefresh(PUVM pUVM, void *pvData)
1107	{
1108	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
1109	NOREF(pThis);
1110	Assert(pThis->fValid);
1111
1112	/*
1113	* For now we'll flush and reload everything.
1114	*/
1115	dbgDiggerLinuxTerm(pUVM, pvData);
1116	return dbgDiggerLinuxInit(pUVM, pvData);
1117	}
1118
1119
1120	/**
1121	* Worker for dbgDiggerLinuxFindStartOfNamesAndSymbolCount that update the
1122	* digger data.
1123	*
1124	* @returns VINF_SUCCESS.
1125	* @param pThis The Linux digger data to update.
1126	* @param pAddrKernelNames The kallsyms_names address.
1127	* @param cKernelSymbols The number of kernel symbol.
1128	* @param cbAddress The guest address size.
1129	*/
1130	static int dbgDiggerLinuxFoundStartOfNames(PDBGDIGGERLINUX pThis, PCDBGFADDRESS pAddrKernelNames,
1131	uint32_t cKernelSymbols, uint32_t cbAddress)
1132	{
1133	pThis->cKernelSymbols = cKernelSymbols;
1134	pThis->AddrKernelNames = *pAddrKernelNames;
1135	pThis->AddrKernelAddresses = *pAddrKernelNames;
1136	DBGFR3AddrSub(&pThis->AddrKernelAddresses, (cKernelSymbols + 1) * cbAddress);
1137
1138	Log(("dbgDiggerLinuxFoundStartOfNames: AddrKernelAddresses=%RGv\n"
1139	"dbgDiggerLinuxFoundStartOfNames: cKernelSymbols=%#x (at %RGv)\n"
1140	"dbgDiggerLinuxFoundStartOfNames: AddrKernelName=%RGv\n",
1141	pThis->AddrKernelAddresses.FlatPtr,
1142	pThis->cKernelSymbols, pThis->AddrKernelNames.FlatPtr - cbAddress,
1143	pThis->AddrKernelNames.FlatPtr));
1144	return VINF_SUCCESS;
1145	}
1146
1147
1148	/**
1149	* Tries to find the address of the kallsyms_names, kallsyms_num_syms and
1150	* kallsyms_addresses symbols.
1151	*
1152	* The kallsyms_num_syms is read and stored in pThis->cKernelSymbols, while the
1153	* addresses of the other two are stored as pThis->AddrKernelNames and
1154	* pThis->AddrKernelAddresses.
1155	*
1156	* @returns VBox status code, success indicating that all three variables have
1157	* been found and taken down.
1158	* @param pUVM The user mode VM handle.
1159	* @param pThis The Linux digger data.
1160	* @param pHitAddr An address we think is inside kallsyms_names.
1161	*/
1162	static int dbgDiggerLinuxFindStartOfNamesAndSymbolCount(PUVM pUVM, PDBGDIGGERLINUX pThis, PCDBGFADDRESS pHitAddr)
1163	{
1164	/*
1165	* Search backwards in chunks.
1166	*/
1167	union
1168	{
1169	uint8_t ab[0x1000];
1170	uint32_t au32[0x1000 / sizeof(uint32_t)];
1171	uint64_t au64[0x1000 / sizeof(uint64_t)];
1172	} uBuf;
1173	uint32_t cbLeft = LNX_MAX_KALLSYMS_NAMES_SIZE;
1174	uint32_t cbBuf = pHitAddr->FlatPtr & (sizeof(uBuf) - 1);
1175	DBGFADDRESS CurAddr = *pHitAddr;
1176	DBGFR3AddrSub(&CurAddr, cbBuf);
1177	cbBuf += sizeof(uint64_t) - 1; /* In case our kobj hit is in the first 4/8 bytes. */
1178	for (;;)
1179	{
1180	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &CurAddr, &uBuf, sizeof(uBuf));
1181	if (RT_FAILURE(rc))
1182	return rc;
1183
1184	/*
1185	* We assume that the three symbols are aligned on guest pointer boundrary.
1186	*
1187	* The boundrary between the two tables should be noticable as the number
1188	* is unlikely to be more than 16 millions, there will be at least one zero
1189	* byte where it is, 64-bit will have 5 zero bytes. Zero bytes aren't all
1190	* that common in the kallsyms_names table.
1191	*
1192	* Also the kallsyms_names table starts with a length byte, which means
1193	* we're likely to see a byte in the range 1..31.
1194	*
1195	* The kallsyms_addresses are mostly sorted (except for the start where the
1196	* absolute symbols are), so we'll spot a bunch of kernel addresses
1197	* immediately preceeding the kallsyms_num_syms field.
1198	*
1199	* Lazy bird: If kallsyms_num_syms is on a buffer boundrary, we skip
1200	* the check for kernel addresses preceeding it.
1201	*/
1202	if (pThis->f64Bit)
1203	{
1204	uint32_t i = cbBuf / sizeof(uint64_t);
1205	while (i-- > 0)
1206	if ( uBuf.au64[i] <= LNX_MAX_KALLSYMS_SYMBOLS
1207	&& uBuf.au64[i] >= LNX_MIN_KALLSYMS_SYMBOLS)
1208	{
1209	uint8_t pb = (uint8_t )&uBuf.au64[i + 1];
1210	if ( pb[0] <= LNX_MAX_KALLSYMS_ENC_LENGTH
1211	&& pb[0] >= LNX_MIN_KALLSYMS_ENC_LENGTH)
1212	{
1213	if ( (i <= 0 \|\| LNX64_VALID_ADDRESS(uBuf.au64[i - 1]))
1214	&& (i <= 1 \|\| LNX64_VALID_ADDRESS(uBuf.au64[i - 2]))
1215	&& (i <= 2 \|\| LNX64_VALID_ADDRESS(uBuf.au64[i - 3])))
1216	return dbgDiggerLinuxFoundStartOfNames(pThis,
1217	DBGFR3AddrAdd(&CurAddr, (i + 1) * sizeof(uint64_t)),
1218	(uint32_t)uBuf.au64[i], sizeof(uint64_t));
1219	}
1220	}
1221	}
1222	else
1223	{
1224	uint32_t i = cbBuf / sizeof(uint32_t);
1225	while (i-- > 0)
1226	if ( uBuf.au32[i] <= LNX_MAX_KALLSYMS_SYMBOLS
1227	&& uBuf.au32[i] >= LNX_MIN_KALLSYMS_SYMBOLS)
1228	{
1229	uint8_t pb = (uint8_t )&uBuf.au32[i + 1];
1230	if ( pb[0] <= LNX_MAX_KALLSYMS_ENC_LENGTH
1231	&& pb[0] >= LNX_MIN_KALLSYMS_ENC_LENGTH)
1232	{
1233	if ( (i <= 0 \|\| LNX32_VALID_ADDRESS(uBuf.au32[i - 1]))
1234	&& (i <= 1 \|\| LNX32_VALID_ADDRESS(uBuf.au32[i - 2]))
1235	&& (i <= 2 \|\| LNX32_VALID_ADDRESS(uBuf.au32[i - 3])))
1236	return dbgDiggerLinuxFoundStartOfNames(pThis,
1237	DBGFR3AddrAdd(&CurAddr, (i + 1) * sizeof(uint32_t)),
1238	uBuf.au32[i], sizeof(uint32_t));
1239	}
1240	}
1241	}
1242
1243	/*
1244	* Advance
1245	*/
1246	if (RT_UNLIKELY(cbLeft <= sizeof(uBuf)))
1247	{
1248	Log(("dbgDiggerLinuxFindStartOfNamesAndSymbolCount: failed (pHitAddr=%RGv)\n", pHitAddr->FlatPtr));
1249	return VERR_NOT_FOUND;
1250	}
1251	cbLeft -= sizeof(uBuf);
1252	DBGFR3AddrSub(&CurAddr, sizeof(uBuf));
1253	cbBuf = sizeof(uBuf);
1254	}
1255	}
1256
1257
1258	/**
1259	* Worker for dbgDiggerLinuxFindEndNames that records the findings.
1260	*
1261	* @returns VINF_SUCCESS
1262	* @param pThis The linux digger data to update.
1263	* @param pAddrMarkers The address of the marker (kallsyms_markers).
1264	* @param cbMarkerEntry The size of a marker entry (32-bit or 64-bit).
1265	*/
1266	static int dbgDiggerLinuxFoundMarkers(PDBGDIGGERLINUX pThis, PCDBGFADDRESS pAddrMarkers, uint32_t cbMarkerEntry)
1267	{
1268	pThis->cbKernelNames = pAddrMarkers->FlatPtr - pThis->AddrKernelNames.FlatPtr;
1269	pThis->AddrKernelNameMarkers = *pAddrMarkers;
1270	pThis->cKernelNameMarkers = RT_ALIGN_32(pThis->cKernelSymbols, 256) / 256;
1271	pThis->AddrKernelTokenTable = *pAddrMarkers;
1272	DBGFR3AddrAdd(&pThis->AddrKernelTokenTable, pThis->cKernelNameMarkers * cbMarkerEntry);
1273
1274	Log(("dbgDiggerLinuxFoundMarkers: AddrKernelNames=%RGv cbKernelNames=%#x\n"
1275	"dbgDiggerLinuxFoundMarkers: AddrKernelNameMarkers=%RGv cKernelNameMarkers=%#x\n"
1276	"dbgDiggerLinuxFoundMarkers: AddrKernelTokenTable=%RGv\n",
1277	pThis->AddrKernelNames.FlatPtr, pThis->cbKernelNames,
1278	pThis->AddrKernelNameMarkers.FlatPtr, pThis->cKernelNameMarkers,
1279	pThis->AddrKernelTokenTable.FlatPtr));
1280	return VINF_SUCCESS;
1281	}
1282
1283
1284	/**
1285	* Tries to find the end of kallsyms_names and thereby the start of
1286	* kallsyms_markers and kallsyms_token_table.
1287	*
1288	* The kallsyms_names size is stored in pThis->cbKernelNames, the addresses of
1289	* the two other symbols in pThis->AddrKernelNameMarkers and
1290	* pThis->AddrKernelTokenTable. The number of marker entries is stored in
1291	* pThis->cKernelNameMarkers.
1292	*
1293	* @returns VBox status code, success indicating that all three variables have
1294	* been found and taken down.
1295	* @param pUVM The user mode VM handle.
1296	* @param pThis The Linux digger data.
1297	* @param pHitAddr An address we think is inside kallsyms_names.
1298	*/
1299	static int dbgDiggerLinuxFindEndOfNamesAndMore(PUVM pUVM, PDBGDIGGERLINUX pThis, PCDBGFADDRESS pHitAddr)
1300	{
1301	/*
1302	* Search forward in chunks.
1303	*/
1304	union
1305	{
1306	uint8_t ab[0x1000];
1307	uint32_t au32[0x1000 / sizeof(uint32_t)];
1308	uint64_t au64[0x1000 / sizeof(uint64_t)];
1309	} uBuf;
1310	bool fPendingZeroHit = false;
1311	uint32_t cbLeft = LNX_MAX_KALLSYMS_NAMES_SIZE + sizeof(uBuf);
1312	uint32_t offBuf = pHitAddr->FlatPtr & (sizeof(uBuf) - 1);
1313	DBGFADDRESS CurAddr = *pHitAddr;
1314	DBGFR3AddrSub(&CurAddr, offBuf);
1315	for (;;)
1316	{
1317	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &CurAddr, &uBuf, sizeof(uBuf));
1318	if (RT_FAILURE(rc))
1319	return rc;
1320
1321	/*
1322	* The kallsyms_names table is followed by kallsyms_markers we assume,
1323	* using sizeof(unsigned long) alignment like the preceeding symbols.
1324	*
1325	* The kallsyms_markers table has entried sizeof(unsigned long) and
1326	* contains offsets into kallsyms_names. The kallsyms_markers used to
1327	* index kallsyms_names and reduce seek time when looking up the name
1328	* of an address/symbol. Each entry in kallsyms_markers covers 256
1329	* symbol names.
1330	*
1331	* Because of this, the first entry is always zero and all the entries
1332	* are ascending. It also follows that the size of the table can be
1333	* calculated from kallsyms_num_syms.
1334	*
1335	* Note! We could also have walked kallsyms_names by skipping
1336	* kallsyms_num_syms names, but this is faster and we will
1337	* validate the encoded names later.
1338	*/
1339	if (pThis->f64Bit)
1340	{
1341	if ( RT_UNLIKELY(fPendingZeroHit)
1342	&& uBuf.au64[0] >= (LNX_MIN_KALLSYMS_ENC_LENGTH + 1) * 256
1343	&& uBuf.au64[0] <= (LNX_MAX_KALLSYMS_ENC_LENGTH + 1) * 256)
1344	return dbgDiggerLinuxFoundMarkers(pThis, DBGFR3AddrSub(&CurAddr, sizeof(uint64_t)), sizeof(uint64_t));
1345
1346	uint32_t const cEntries = sizeof(uBuf) / sizeof(uint64_t);
1347	for (uint32_t i = offBuf / sizeof(uint64_t); i < cEntries; i++)
1348	if (uBuf.au64[i] == 0)
1349	{
1350	if (RT_UNLIKELY(i + 1 >= cEntries))
1351	{
1352	fPendingZeroHit = true;
1353	break;
1354	}
1355	if ( uBuf.au64[i + 1] >= (LNX_MIN_KALLSYMS_ENC_LENGTH + 1) * 256
1356	&& uBuf.au64[i + 1] <= (LNX_MAX_KALLSYMS_ENC_LENGTH + 1) * 256)
1357	return dbgDiggerLinuxFoundMarkers(pThis, DBGFR3AddrAdd(&CurAddr, i * sizeof(uint64_t)), sizeof(uint64_t));
1358	}
1359	}
1360	else
1361	{
1362	if ( RT_UNLIKELY(fPendingZeroHit)
1363	&& uBuf.au32[0] >= (LNX_MIN_KALLSYMS_ENC_LENGTH + 1) * 256
1364	&& uBuf.au32[0] <= (LNX_MAX_KALLSYMS_ENC_LENGTH + 1) * 256)
1365	return dbgDiggerLinuxFoundMarkers(pThis, DBGFR3AddrSub(&CurAddr, sizeof(uint32_t)), sizeof(uint32_t));
1366
1367	uint32_t const cEntries = sizeof(uBuf) / sizeof(uint32_t);
1368	for (uint32_t i = offBuf / sizeof(uint32_t); i < cEntries; i++)
1369	if (uBuf.au32[i] == 0)
1370	{
1371	if (RT_UNLIKELY(i + 1 >= cEntries))
1372	{
1373	fPendingZeroHit = true;
1374	break;
1375	}
1376	if ( uBuf.au32[i + 1] >= (LNX_MIN_KALLSYMS_ENC_LENGTH + 1) * 256
1377	&& uBuf.au32[i + 1] <= (LNX_MAX_KALLSYMS_ENC_LENGTH + 1) * 256)
1378	return dbgDiggerLinuxFoundMarkers(pThis, DBGFR3AddrAdd(&CurAddr, i * sizeof(uint32_t)), sizeof(uint32_t));
1379	}
1380	}
1381
1382	/*
1383	* Advance
1384	*/
1385	if (RT_UNLIKELY(cbLeft <= sizeof(uBuf)))
1386	{
1387	Log(("dbgDiggerLinuxFindEndOfNamesAndMore: failed (pHitAddr=%RGv)\n", pHitAddr->FlatPtr));
1388	return VERR_NOT_FOUND;
1389	}
1390	cbLeft -= sizeof(uBuf);
1391	DBGFR3AddrAdd(&CurAddr, sizeof(uBuf));
1392	offBuf = 0;
1393	}
1394	}
1395
1396
1397	/**
1398	* Locates the kallsyms_token_index table.
1399	*
1400	* Storing the address in pThis->AddrKernelTokenIndex and the size of the token
1401	* table in pThis->cbKernelTokenTable.
1402	*
1403	* @returns VBox status code.
1404	* @param pUVM The user mode VM handle.
1405	* @param pThis The Linux digger data.
1406	*/
1407	static int dbgDiggerLinuxFindTokenIndex(PUVM pUVM, PDBGDIGGERLINUX pThis)
1408	{
1409	/*
1410	* The kallsyms_token_table is very much like a string table. Due to the
1411	* nature of the compression algorithm it is reasonably short (one example
1412	* here is 853 bytes), so we'll not be reading it in chunks but in full.
1413	* To be on the safe side, we read 8KB, ASSUMING we won't run into unmapped
1414	* memory or any other nasty stuff...
1415	*/
1416	union
1417	{
1418	uint8_t ab[0x2000];
1419	uint16_t au16[0x2000 / sizeof(uint16_t)];
1420	} uBuf;
1421	DBGFADDRESS CurAddr = pThis->AddrKernelTokenTable;
1422	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &CurAddr, &uBuf, sizeof(uBuf));
1423	if (RT_FAILURE(rc))
1424	return rc;
1425
1426	/*
1427	* We've got two choices here, either walk the string table or look for
1428	* the next structure, kallsyms_token_index.
1429	*
1430	* The token index is a table of 256 uint16_t entries (index by bytes
1431	* from kallsyms_names) that gives offsets in kallsyms_token_table. It
1432	* starts with a zero entry and the following entries are sorted in
1433	* ascending order. The range of the entries are reasonably small since
1434	* kallsyms_token_table is small.
1435	*
1436	* The alignment seems to be sizeof(unsigned long), just like
1437	* kallsyms_token_table.
1438	*
1439	* So, we start by looking for a zero 16-bit entry.
1440	*/
1441	uint32_t cIncr = (pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t)) / sizeof(uint16_t);
1442
1443	for (uint32_t i = 0; i < sizeof(uBuf) / sizeof(uint16_t) - 16; i += cIncr)
1444	if ( uBuf.au16[i] == 0
1445	&& uBuf.au16[i + 1] > 0
1446	&& uBuf.au16[i + 1] <= LNX_MAX_KALLSYMS_TOKEN_LEN
1447	&& (uint16_t)(uBuf.au16[i + 2] - uBuf.au16[i + 1] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1448	&& (uint16_t)(uBuf.au16[i + 3] - uBuf.au16[i + 2] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1449	&& (uint16_t)(uBuf.au16[i + 4] - uBuf.au16[i + 3] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1450	&& (uint16_t)(uBuf.au16[i + 5] - uBuf.au16[i + 4] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1451	&& (uint16_t)(uBuf.au16[i + 6] - uBuf.au16[i + 5] - 1U) <= (uint16_t)LNX_MAX_KALLSYMS_TOKEN_LEN
1452	)
1453	{
1454	pThis->AddrKernelTokenIndex = CurAddr;
1455	DBGFR3AddrAdd(&pThis->AddrKernelTokenIndex, i * sizeof(uint16_t));
1456	pThis->cbKernelTokenTable = i * sizeof(uint16_t);
1457	return VINF_SUCCESS;
1458	}
1459
1460	Log(("dbgDiggerLinuxFindTokenIndex: Failed (%RGv..%RGv)\n", CurAddr.FlatPtr, CurAddr.FlatPtr + (RTGCUINTPTR)sizeof(uBuf)));
1461	return VERR_NOT_FOUND;
1462	}
1463
1464
1465	/**
1466	* Loads the kernel symbols from the kallsyms tables.
1467	*
1468	* @returns VBox status code.
1469	* @param pUVM The user mode VM handle.
1470	* @param pThis The Linux digger data.
1471	*/
1472	static int dbgDiggerLinuxLoadKernelSymbols(PUVM pUVM, PDBGDIGGERLINUX pThis)
1473	{
1474	/*
1475	* Allocate memory for temporary table copies, reading the tables as we go.
1476	*/
1477	uint32_t const cbGuestAddr = pThis->f64Bit ? sizeof(uint64_t) : sizeof(uint32_t);
1478	void pvAddresses = RTMemAllocZ(pThis->cKernelSymbols cbGuestAddr);
1479	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &pThis->AddrKernelAddresses, pvAddresses, pThis->cKernelSymbols * cbGuestAddr);
1480	if (RT_SUCCESS(rc))
1481	{
1482	uint8_t pbNames = (uint8_t )RTMemAllocZ(pThis->cbKernelNames);
1483	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &pThis->AddrKernelNames, pbNames, pThis->cbKernelNames);
1484	if (RT_SUCCESS(rc))
1485	{
1486	char pszzTokens = (char )RTMemAllocZ(pThis->cbKernelTokenTable);
1487	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &pThis->AddrKernelTokenTable, pszzTokens, pThis->cbKernelTokenTable);
1488	if (RT_SUCCESS(rc))
1489	{
1490	uint16_t paoffTokens = (uint16_t )RTMemAllocZ(256 * sizeof(uint16_t));
1491	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &pThis->AddrKernelTokenIndex, paoffTokens, 256 * sizeof(uint16_t));
1492	if (RT_SUCCESS(rc))
1493	{
1494	/*
1495	* Figure out the kernel start and end.
1496	*/
1497	RTGCUINTPTR uKernelStart = pThis->AddrKernelAddresses.FlatPtr;
1498	RTGCUINTPTR uKernelEnd = pThis->AddrKernelTokenIndex.FlatPtr + 256 * sizeof(uint16_t);
1499	uint32_t i;
1500	if (cbGuestAddr == sizeof(uint64_t))
1501	{
1502	uint64_t pauAddrs = (uint64_t )pvAddresses;
1503	for (i = 0; i < pThis->cKernelSymbols; i++)
1504	if ( pauAddrs[i] < uKernelStart
1505	&& LNX64_VALID_ADDRESS(pauAddrs[i])
1506	&& uKernelStart - pauAddrs[i] < LNX_MAX_KERNEL_SIZE)
1507	uKernelStart = pauAddrs[i];
1508
1509	for (i = pThis->cKernelSymbols - 1; i > 0; i--)
1510	if ( pauAddrs[i] > uKernelEnd
1511	&& LNX64_VALID_ADDRESS(pauAddrs[i])
1512	&& pauAddrs[i] - uKernelEnd < LNX_MAX_KERNEL_SIZE)
1513	uKernelEnd = pauAddrs[i];
1514	}
1515	else
1516	{
1517	uint32_t pauAddrs = (uint32_t )pvAddresses;
1518	for (i = 0; i < pThis->cKernelSymbols; i++)
1519	if ( pauAddrs[i] < uKernelStart
1520	&& LNX32_VALID_ADDRESS(pauAddrs[i])
1521	&& uKernelStart - pauAddrs[i] < LNX_MAX_KERNEL_SIZE)
1522	uKernelStart = pauAddrs[i];
1523
1524	for (i = pThis->cKernelSymbols - 1; i > 0; i--)
1525	if ( pauAddrs[i] > uKernelEnd
1526	&& LNX32_VALID_ADDRESS(pauAddrs[i])
1527	&& pauAddrs[i] - uKernelEnd < LNX_MAX_KERNEL_SIZE)
1528	uKernelEnd = pauAddrs[i];
1529	}
1530
1531	RTGCUINTPTR cbKernel = uKernelEnd - uKernelStart;
1532	pThis->cbKernel = (uint32_t)cbKernel;
1533	DBGFR3AddrFromFlat(pUVM, &pThis->AddrKernelBase, uKernelStart);
1534	Log(("dbgDiggerLinuxLoadKernelSymbols: uKernelStart=%RGv cbKernel=%#x\n", uKernelStart, cbKernel));
1535
1536	/*
1537	* Create a module for the kernel.
1538	*/
1539	RTDBGMOD hMod;
1540	rc = RTDbgModCreate(&hMod, "vmlinux", cbKernel, 0 /fFlags/);
1541	if (RT_SUCCESS(rc))
1542	{
1543	rc = RTDbgModSetTag(hMod, DIG_LNX_MOD_TAG); AssertRC(rc);
1544	rc = VINF_SUCCESS;
1545
1546	/*
1547	* Enumerate the symbols.
1548	*/
1549	uint8_t const pbCurAddr = (uint8_t const )pvAddresses;
1550	uint32_t offName = 0;
1551	uint32_t cLeft = pThis->cKernelSymbols;
1552	while (cLeft-- > 0 && RT_SUCCESS(rc))
1553	{
1554	/* Decode the symbol name first. */
1555	if (RT_LIKELY(offName < pThis->cbKernelNames))
1556	{
1557	uint8_t cbName = pbNames[offName++];
1558	if (RT_LIKELY(offName + cbName <= pThis->cbKernelNames))
1559	{
1560	char szSymbol[4096];
1561	uint32_t offSymbol = 0;
1562	while (cbName-- > 0)
1563	{
1564	uint8_t bEnc = pbNames[offName++];
1565	uint16_t offToken = paoffTokens[bEnc];
1566	if (RT_LIKELY(offToken < pThis->cbKernelTokenTable))
1567	{
1568	const char *pszToken = &pszzTokens[offToken];
1569	char ch;
1570	while ((ch = *pszToken++) != '\0')
1571	if (offSymbol < sizeof(szSymbol) - 1)
1572	szSymbol[offSymbol++] = ch;
1573	}
1574	else
1575	{
1576	rc = VERR_INVALID_UTF8_ENCODING;
1577	break;
1578	}
1579	}
1580	szSymbol[offSymbol < sizeof(szSymbol) ? offSymbol : sizeof(szSymbol) - 1] = '\0';
1581
1582	/* The address. */
1583	RTGCUINTPTR uSymAddr = cbGuestAddr == sizeof(uint64_t)
1584	? (uint64_t )pbCurAddr : (uint32_t )pbCurAddr;
1585	pbCurAddr += cbGuestAddr;
1586
1587	/* Add it without the type char. */
1588	if (uSymAddr - uKernelStart <= cbKernel)
1589	{
1590	rc = RTDbgModSymbolAdd(hMod, &szSymbol[1], RTDBGSEGIDX_RVA, uSymAddr - uKernelStart,
1591	0 /cb/, 0 /fFlags/, NULL);
1592	if (RT_FAILURE(rc))
1593	{
1594	if ( rc == VERR_DBG_SYMBOL_NAME_OUT_OF_RANGE
1595	\|\| rc == VERR_DBG_INVALID_RVA
1596	\|\| rc == VERR_DBG_ADDRESS_CONFLICT
1597	\|\| rc == VERR_DBG_DUPLICATE_SYMBOL)
1598	{
1599	Log2(("dbgDiggerLinuxLoadKernelSymbols: RTDbgModSymbolAdd(,%s,) failed %Rrc (ignored)\n", szSymbol, rc));
1600	rc = VINF_SUCCESS;
1601	}
1602	else
1603	Log(("dbgDiggerLinuxLoadKernelSymbols: RTDbgModSymbolAdd(,%s,) failed %Rrc\n", szSymbol, rc));
1604	}
1605	}
1606	}
1607	else
1608	{
1609	rc = VERR_END_OF_STRING;
1610	Log(("dbgDiggerLinuxLoadKernelSymbols: offName=%#x cLeft=%#x cbName=%#x cbKernelNames=%#x\n",
1611	offName, cLeft, cbName, pThis->cbKernelNames));
1612	}
1613	}
1614	else
1615	{
1616	rc = VERR_END_OF_STRING;
1617	Log(("dbgDiggerLinuxLoadKernelSymbols: offName=%#x cLeft=%#x cbKernelNames=%#x\n",
1618	offName, cLeft, pThis->cbKernelNames));
1619	}
1620	}
1621
1622	/*
1623	* Link the module into the address space.
1624	*/
1625	if (RT_SUCCESS(rc))
1626	{
1627	RTDBGAS hAs = DBGFR3AsResolveAndRetain(pUVM, DBGF_AS_KERNEL);
1628	if (hAs != NIL_RTDBGAS)
1629	rc = RTDbgAsModuleLink(hAs, hMod, uKernelStart, RTDBGASLINK_FLAGS_REPLACE);
1630	else
1631	rc = VERR_INTERNAL_ERROR;
1632	RTDbgAsRelease(hAs);
1633	}
1634	else
1635	Log(("dbgDiggerLinuxLoadKernelSymbols: Failed: %Rrc\n", rc));
1636	RTDbgModRelease(hMod);
1637	}
1638	else
1639	Log(("dbgDiggerLinuxLoadKernelSymbols: RTDbgModCreate failed: %Rrc\n", rc));
1640	}
1641	else
1642	Log(("dbgDiggerLinuxLoadKernelSymbols: Reading token index at %RGv failed: %Rrc\n",
1643	pThis->AddrKernelTokenIndex.FlatPtr, rc));
1644	RTMemFree(paoffTokens);
1645	}
1646	else
1647	Log(("dbgDiggerLinuxLoadKernelSymbols: Reading token table at %RGv failed: %Rrc\n",
1648	pThis->AddrKernelTokenTable.FlatPtr, rc));
1649	RTMemFree(pszzTokens);
1650	}
1651	else
1652	Log(("dbgDiggerLinuxLoadKernelSymbols: Reading encoded names at %RGv failed: %Rrc\n",
1653	pThis->AddrKernelNames.FlatPtr, rc));
1654	RTMemFree(pbNames);
1655	}
1656	else
1657	Log(("dbgDiggerLinuxLoadKernelSymbols: Reading symbol addresses at %RGv failed: %Rrc\n",
1658	pThis->AddrKernelAddresses.FlatPtr, rc));
1659	RTMemFree(pvAddresses);
1660	return rc;
1661	}
1662
1663
1664	/**
1665	* Checks if there is a likely kallsyms_names fragment at pHitAddr.
1666	*
1667	* @returns true if it's a likely fragment, false if not.
1668	* @param pUVM The user mode VM handle.
1669	* @param pHitAddr The address where paNeedle was found.
1670	* @param pabNeedle The fragment we've been searching for.
1671	* @param cbNeedle The length of the fragment.
1672	*/
1673	static bool dbgDiggerLinuxIsLikelyNameFragment(PUVM pUVM, PCDBGFADDRESS pHitAddr, uint8_t const *pabNeedle, uint8_t cbNeedle)
1674	{
1675	/*
1676	* Examples of lead and tail bytes of our choosen needle in a randomly
1677	* picked kernel:
1678	* k o b j
1679	* 22 6b 6f 62 6a aa
1680	* fc 6b 6f 62 6a aa
1681	* 82 6b 6f 62 6a 5f - ascii trail byte (_).
1682	* ee 6b 6f 62 6a aa
1683	* fc 6b 6f 62 6a 5f - ascii trail byte (_).
1684	* 0a 74 6b 6f 62 6a 5f ea - ascii lead (t) and trail (_) bytes.
1685	* 0b 54 6b 6f 62 6a aa - ascii lead byte (T).
1686	* ... omitting 29 samples similar to the last two ...
1687	* d8 6b 6f 62 6a aa
1688	* d8 6b 6f 62 6a aa
1689	* d8 6b 6f 62 6a aa
1690	* d8 6b 6f 62 6a aa
1691	* f9 5f 6b 6f 62 6a 5f 94 - ascii lead and trail bytes (_)
1692	* f9 5f 6b 6f 62 6a 0c - ascii lead byte (_).
1693	* fd 6b 6f 62 6a 0f
1694	* ... enough.
1695	*/
1696	uint8_t abBuf[32];
1697	DBGFADDRESS ReadAddr = *pHitAddr;
1698	DBGFR3AddrSub(&ReadAddr, 2);
1699	int rc = DBGFR3MemRead(pUVM, 0 /idCpu/, &ReadAddr, abBuf, 2 + cbNeedle + 2);
1700	if (RT_SUCCESS(rc))
1701	{
1702	if (memcmp(&abBuf[2], pabNeedle, cbNeedle) == 0) /* paranoia */
1703	{
1704	uint8_t const bLead = abBuf[1] == '_' \|\| abBuf[1] == 'T' \|\| abBuf[1] == 't' ? abBuf[0] : abBuf[1];
1705	uint8_t const offTail = 2 + cbNeedle;
1706	uint8_t const bTail = abBuf[offTail] == '_' ? abBuf[offTail] : abBuf[offTail + 1];
1707	if ( bLead >= 1 && (bLead < 0x20 \|\| bLead >= 0x80)
1708	&& bTail >= 1 && (bTail < 0x20 \|\| bTail >= 0x80))
1709	return true;
1710	Log(("dbgDiggerLinuxIsLikelyNameFragment: failed at %RGv: bLead=%#x bTail=%#x (offTail=%#x)\n",
1711	pHitAddr->FlatPtr, bLead, bTail, offTail));
1712	}
1713	else
1714	Log(("dbgDiggerLinuxIsLikelyNameFragment: failed at %RGv: Needle changed!\n", pHitAddr->FlatPtr));
1715	}
1716	else
1717	Log(("dbgDiggerLinuxIsLikelyNameFragment: failed at %RGv: %Rrc\n", pHitAddr->FlatPtr, rc));
1718
1719	return false;
1720	}
1721
1722	/**
1723	* Tries to find and load the kernel symbol table with the given needle.
1724	*
1725	* @returns VBox status code.
1726	* @param pThis The Linux digger data.
1727	* @param pUVM The user mode VM handle.
1728	* @param pabNeedle The needle to use for searching.
1729	* @param cbNeedle Size of the needle in bytes.
1730	*/
1731	static int dbgDiggerLinuxFindSymbolTableFromNeedle(PDBGDIGGERLINUX pThis, PUVM pUVM, uint8_t const *pabNeedle, size_t cbNeedle)
1732	{
1733	int rc = VINF_SUCCESS;
1734
1735	/*
1736	* Go looking for the kallsyms table. If it's there, it will be somewhere
1737	* after the linux_banner symbol, so use it for starting the search.
1738	*/
1739	DBGFADDRESS CurAddr = pThis->AddrLinuxBanner;
1740	uint32_t cbLeft = LNX_MAX_KERNEL_SIZE;
1741	while (cbLeft > 4096)
1742	{
1743	DBGFADDRESS HitAddr;
1744	rc = DBGFR3MemScan(pUVM, 0 /idCpu/, &CurAddr, cbLeft, 1 /uAlign/,
1745	pabNeedle, cbNeedle, &HitAddr);
1746	if (RT_FAILURE(rc))
1747	break;
1748	if (dbgDiggerLinuxIsLikelyNameFragment(pUVM, &HitAddr, pabNeedle, cbNeedle))
1749	{
1750	/* There will be another hit near by. */
1751	DBGFR3AddrAdd(&HitAddr, 1);
1752	rc = DBGFR3MemScan(pUVM, 0 /idCpu/, &HitAddr, LNX_MAX_KALLSYMS_NAMES_SIZE, 1 /uAlign/,
1753	pabNeedle, cbNeedle, &HitAddr);
1754	if ( RT_SUCCESS(rc)
1755	&& dbgDiggerLinuxIsLikelyNameFragment(pUVM, &HitAddr, pabNeedle, cbNeedle))
1756	{
1757	/*
1758	* We've got a very likely candidate for a location inside kallsyms_names.
1759	* Try find the start of it, that is to say, try find kallsyms_num_syms.
1760	* kallsyms_num_syms is aligned on sizeof(unsigned long) boundrary
1761	*/
1762	rc = dbgDiggerLinuxFindStartOfNamesAndSymbolCount(pUVM, pThis, &HitAddr);
1763	if (RT_SUCCESS(rc))
1764	rc = dbgDiggerLinuxFindEndOfNamesAndMore(pUVM, pThis, &HitAddr);
1765	if (RT_SUCCESS(rc))
1766	rc = dbgDiggerLinuxFindTokenIndex(pUVM, pThis);
1767	if (RT_SUCCESS(rc))
1768	rc = dbgDiggerLinuxLoadKernelSymbols(pUVM, pThis);
1769	if (RT_SUCCESS(rc))
1770	break;
1771	}
1772	}
1773
1774	/*
1775	* Advance.
1776	*/
1777	RTGCUINTPTR cbDistance = HitAddr.FlatPtr - CurAddr.FlatPtr + cbNeedle;
1778	if (RT_UNLIKELY(cbDistance >= cbLeft))
1779	{
1780	Log(("dbgDiggerLinuxInit: Failed to find kallsyms\n"));
1781	break;
1782	}
1783	cbLeft -= cbDistance;
1784	DBGFR3AddrAdd(&CurAddr, cbDistance);
1785
1786	}
1787
1788	return rc;
1789	}
1790
1791	/**
1792	* Skips whitespace and comments in the given config returning the pointer
1793	* to the first non whitespace character.
1794	*
1795	* @returns Pointer to the first non whitespace character or NULL if the end
1796	* of the string was reached.
1797	* @param pszCfg The config string.
1798	*/
1799	static const char dbgDiggerLinuxCfgSkipWhitespace(const char pszCfg)
1800	{
1801	do
1802	{
1803	while ( *pszCfg != '\0'
1804	&& ( RT_C_IS_SPACE(*pszCfg)
1805	\|\| *pszCfg == '\n'))
1806	pszCfg++;
1807
1808	/* Do we have a comment? Skip it. */
1809	if (*pszCfg == '#')
1810	{
1811	while ( *pszCfg != '\n'
1812	&& *pszCfg != '\0')
1813	pszCfg++;
1814	}
1815	} while ( *pszCfg != '\0'
1816	&& ( RT_C_IS_SPACE(*pszCfg)
1817	\|\| *pszCfg == '\n'
1818	\|\| *pszCfg == '#'));
1819
1820	return pszCfg;
1821	}
1822
1823	/**
1824	* Parses an identifier at the given position.
1825	*
1826	* @returns VBox status code.
1827	* @param pszCfg The config data.
1828	* @param ppszCfgNext Where to store the pointer to the data following the identifier.
1829	* @param ppszIde Where to store the pointer to the identifier on success.
1830	* Free with RTStrFree().
1831	*/
1832	static int dbgDiggerLinuxCfgParseIde(const char pszCfg, const char ppszCfgNext, char *ppszIde)
1833	{
1834	int rc = VINF_SUCCESS;
1835	size_t cchIde = 0;
1836
1837	while ( *pszCfg != '\0'
1838	&& ( RT_C_IS_ALNUM(*pszCfg)
1839	\|\| *pszCfg == '_'))
1840	{
1841	cchIde++;
1842	pszCfg++;
1843	}
1844
1845	if (cchIde)
1846	{
1847	*ppszIde = RTStrDupN(pszCfg - cchIde, cchIde);
1848	if (!*ppszIde)
1849	rc = VERR_NO_STR_MEMORY;
1850	}
1851
1852	*ppszCfgNext = pszCfg;
1853	return rc;
1854	}
1855
1856	/**
1857	* Parses a value for a config item.
1858	*
1859	* @returns VBox status code.
1860	* @param pszCfg The config data.
1861	* @param ppszCfgNext Where to store the pointer to the data following the identifier.
1862	* @param pcbCfg Where the initial size of the string is stored.
1863	* Contains the remaining string length on return.
1864	* @param ppCfgItem Where to store the created config item on success.
1865	*/
1866	static int dbgDiggerLinuxCfgParseVal(const char pszCfg, const char *ppszCfgNext,
1867	PDBGDIGGERLINUXCFGITEM *ppCfgItem)
1868	{
1869	int rc = VINF_SUCCESS;
1870	PDBGDIGGERLINUXCFGITEM pCfgItem = NULL;
1871
1872	if (RT_C_IS_DIGIT(pszCfg) \|\| pszCfg == '-')
1873	{
1874	/* Parse the number. */
1875	int64_t i64Num;
1876	rc = RTStrToInt64Ex(pszCfg, (char **)ppszCfgNext, 0, &i64Num);
1877	if ( RT_SUCCESS(rc)
1878	\|\| rc == VWRN_TRAILING_CHARS
1879	\|\| rc == VWRN_TRAILING_SPACES)
1880	{
1881	pCfgItem = (PDBGDIGGERLINUXCFGITEM)RTMemAllocZ(sizeof(DBGDIGGERLINUXCFGITEM));
1882	if (pCfgItem)
1883	{
1884	pCfgItem->enmType = DBGDIGGERLINUXCFGITEMTYPE_NUMBER;
1885	pCfgItem->u.i64Num = i64Num;
1886	}
1887	else
1888	rc = VERR_NO_MEMORY;
1889	}
1890	}
1891	else if (*pszCfg == '\"')
1892	{
1893	/* Parse a string. */
1894	const char *pszCfgCur = pszCfg + 1;
1895	while ( *pszCfgCur != '\0'
1896	&& *pszCfgCur != '\"')
1897	pszCfgCur++;
1898
1899	if (*pszCfgCur == '\"')
1900	{
1901	pCfgItem = (PDBGDIGGERLINUXCFGITEM)RTMemAllocZ(RT_OFFSETOF(DBGDIGGERLINUXCFGITEM, u.aszString[pszCfgCur - pszCfg + 1]));
1902	if (pCfgItem)
1903	{
1904	pCfgItem->enmType = DBGDIGGERLINUXCFGITEMTYPE_STRING;
1905	RTStrCopyEx(&pCfgItem->u.aszString[0], pszCfgCur - pszCfg + 1, pszCfg, pszCfgCur - pszCfg);
1906	*ppszCfgNext = pszCfgCur + 1;
1907	}
1908	else
1909	rc = VERR_NO_MEMORY;
1910	}
1911	else
1912	rc = VERR_INVALID_STATE;
1913	}
1914	else if ( *pszCfg == 'y'
1915	\|\| *pszCfg == 'm')
1916	{
1917	/* Included or module. */
1918	pCfgItem = (PDBGDIGGERLINUXCFGITEM)RTMemAllocZ(sizeof(DBGDIGGERLINUXCFGITEM));
1919	if (pCfgItem)
1920	{
1921	pCfgItem->enmType = DBGDIGGERLINUXCFGITEMTYPE_FLAG;
1922	pCfgItem->u.fModule = *pszCfg == 'm';
1923	}
1924	else
1925	rc = VERR_NO_MEMORY;
1926	pszCfg++;
1927	*ppszCfgNext = pszCfg;
1928	}
1929	else
1930	rc = VERR_INVALID_STATE;
1931
1932	if (RT_SUCCESS(rc))
1933	*ppCfgItem = pCfgItem;
1934	else if (pCfgItem)
1935	RTMemFree(pCfgItem);
1936
1937	return rc;
1938	}
1939
1940	/**
1941	* Parses the given kernel config and creates the config database.
1942	*
1943	* @returns VBox status code
1944	* @param pThis The Linux digger data.
1945	* @param pbCfg The config string.
1946	*/
1947	static int dbgDiggerLinuxCfgParse(PDBGDIGGERLINUX pThis, const char *pszCfg)
1948	{
1949	int rc = VINF_SUCCESS;
1950
1951	/*
1952	* The config is a text file with the following elements:
1953	* # starts a comment which goes till the end of the line
1954	* <Ide>=<val> where <Ide> is an identifier consisting of
1955	* alphanumerical characters (including _)
1956	* <val> denotes the value for the identifier and can have the following
1957	* formats:
1958	* (-)[0-9]* for numbers
1959	* "..." for a string value
1960	* m when a feature is enabled as a module
1961	* y when a feature is enabled
1962	* Newlines are used as a separator between values and mark the end
1963	* of a comment
1964	*/
1965	const char *pszCfgCur = pszCfg;
1966	while ( RT_SUCCESS(rc)
1967	&& *pszCfgCur != '\0')
1968	{
1969	/* Start skipping the whitespace. */
1970	pszCfgCur = dbgDiggerLinuxCfgSkipWhitespace(pszCfgCur);
1971	if ( pszCfgCur
1972	&& *pszCfgCur != '\0')
1973	{
1974	char *pszIde = NULL;
1975	/* Must be an identifier, parse it. */
1976	rc = dbgDiggerLinuxCfgParseIde(pszCfgCur, &pszCfgCur, &pszIde);
1977	if (RT_SUCCESS(rc))
1978	{
1979	/*
1980	* Skip whitespace again (shouldn't be required because = follows immediately
1981	* in the observed configs).
1982	*/
1983	pszCfgCur = dbgDiggerLinuxCfgSkipWhitespace(pszCfgCur);
1984	if ( pszCfgCur
1985	&& *pszCfgCur == '=')
1986	{
1987	pszCfgCur++;
1988	pszCfgCur = dbgDiggerLinuxCfgSkipWhitespace(pszCfgCur);
1989	if ( pszCfgCur
1990	&& *pszCfgCur != '\0')
1991	{
1992	/* Get the value. */
1993	PDBGDIGGERLINUXCFGITEM pCfgItem = NULL;
1994	rc = dbgDiggerLinuxCfgParseVal(pszCfgCur, &pszCfgCur, &pCfgItem);
1995	if (RT_SUCCESS(rc))
1996	{
1997	pCfgItem->Core.pszString = pszIde;
1998	bool fRc = RTStrSpaceInsert(&pThis->hCfgDb, &pCfgItem->Core);
1999	if (!fRc)
2000	{
2001	RTStrFree(pszIde);
2002	RTMemFree(pCfgItem);
2003	rc = VERR_INVALID_STATE;
2004	}
2005	}
2006	}
2007	else
2008	rc = VERR_EOF;
2009	}
2010	else
2011	rc = VERR_INVALID_STATE;
2012	}
2013
2014	if (RT_FAILURE(rc))
2015	RTStrFree(pszIde);
2016	}
2017	else
2018	break; /* Reached the end of the config. */
2019	}
2020
2021	if (RT_FAILURE(rc))
2022	dbgDiggerLinuxCfgDbDestroy(pThis);
2023
2024	return rc;
2025	}
2026
2027	/**
2028	* Decompresses the given config and validates the UTF-8 encoding.
2029	*
2030	* @returns VBox status code.
2031	* @param pbCfgComp The compressed config.
2032	* @param cbCfgComp Size of the compressed config.
2033	* @param ppszCfg Where to store the pointer to the decompressed config
2034	* on success.
2035	*/
2036	static int dbgDiggerLinuxCfgDecompress(const uint8_t pbCfgComp, size_t cbCfgComp, char *ppszCfg)
2037	{
2038	int rc = VINF_SUCCESS;
2039	RTVFSIOSTREAM hVfsIos = NIL_RTVFSIOSTREAM;
2040
2041	rc = RTVfsIoStrmFromBuffer(RTFILE_O_READ, pbCfgComp, cbCfgComp, &hVfsIos);
2042	if (RT_SUCCESS(rc))
2043	{
2044	RTVFSIOSTREAM hVfsIosDecomp = NIL_RTVFSIOSTREAM;
2045	rc = RTZipGzipDecompressIoStream(hVfsIos, RTZIPGZIPDECOMP_F_ALLOW_ZLIB_HDR, &hVfsIosDecomp);
2046	if (RT_SUCCESS(rc))
2047	{
2048	char *pszCfg = NULL;
2049	size_t cchCfg = 0;
2050	size_t cbRead = 0;
2051
2052	do
2053	{
2054	uint8_t abBuf[_64K];
2055	rc = RTVfsIoStrmRead(hVfsIosDecomp, abBuf, sizeof(abBuf), true /fBlocking/, &cbRead);
2056	if (rc == VINF_EOF && cbRead == 0)
2057	rc = VINF_SUCCESS;
2058	if ( RT_SUCCESS(rc)
2059	&& cbRead > 0)
2060	{
2061	/* Append data. */
2062	char *pszCfgNew = pszCfg;
2063	rc = RTStrRealloc(&pszCfgNew, cchCfg + cbRead + 1);
2064	if (RT_SUCCESS(rc))
2065	{
2066	pszCfg = pszCfgNew;
2067	memcpy(pszCfg + cchCfg, &abBuf[0], cbRead);
2068	cchCfg += cbRead;
2069	pszCfg[cchCfg] = '\0'; /* Enforce string termination. */
2070	}
2071	}
2072	} while (RT_SUCCESS(rc) && cbRead > 0);
2073
2074	if (RT_SUCCESS(rc))
2075	*ppszCfg = pszCfg;
2076	else if (RT_FAILURE(rc) && pszCfg)
2077	RTStrFree(pszCfg);
2078
2079	RTVfsIoStrmRelease(hVfsIosDecomp);
2080	}
2081	RTVfsIoStrmRelease(hVfsIos);
2082	}
2083
2084	return rc;
2085	}
2086
2087	/**
2088	* Reads and decodes the compressed kernel config.
2089	*
2090	* @returns VBox status code.
2091	* @param pThis The Linux digger data.
2092	* @param pUVM The user mode VM handle.
2093	* @param pAddrStart The start address of the compressed config.
2094	* @param cbCfgComp The size of the compressed config.
2095	*/
2096	static int dbgDiggerLinuxCfgDecode(PDBGDIGGERLINUX pThis, PUVM pUVM,
2097	PCDBGFADDRESS pAddrStart, size_t cbCfgComp)
2098	{
2099	int rc = VINF_SUCCESS;
2100	uint8_t pbCfgComp = (uint8_t )RTMemTmpAlloc(cbCfgComp);
2101	if (!pbCfgComp)
2102	return VERR_NO_MEMORY;
2103
2104	rc = DBGFR3MemRead(pUVM, 0 /idCpu/, pAddrStart, pbCfgComp, cbCfgComp);
2105	if (RT_SUCCESS(rc))
2106	{
2107	char *pszCfg = NULL;
2108	rc = dbgDiggerLinuxCfgDecompress(pbCfgComp, cbCfgComp, &pszCfg);
2109	if (RT_SUCCESS(rc))
2110	{
2111	if (RTStrIsValidEncoding(pszCfg))
2112	rc = dbgDiggerLinuxCfgParse(pThis, pszCfg);
2113	else
2114	rc = VERR_INVALID_UTF8_ENCODING;
2115	RTStrFree(pszCfg);
2116	}
2117	}
2118
2119	RTMemFree(pbCfgComp);
2120	return rc;
2121	}
2122
2123	/**
2124	* Tries to find the compressed kernel config in the kernel address space
2125	* and sets up the config database.
2126	*
2127	* @returns VBox status code.
2128	* @param pThis The Linux digger data.
2129	* @param pUVM The user mode VM handle.
2130	*/
2131	static int dbgDiggerLinuxCfgFind(PDBGDIGGERLINUX pThis, PUVM pUVM)
2132	{
2133	int rc = VINF_SUCCESS;
2134
2135	/*
2136	* Go looking for the IKCFG_ST string which indicates the start
2137	* of the compressed config file.
2138	*/
2139	static const uint8_t s_abCfgNeedleStart[] = "IKCFG_ST";
2140	static const uint8_t s_abCfgNeedleEnd[] = "IKCFG_ED";
2141	DBGFADDRESS CurAddr = pThis->AddrLinuxBanner;
2142	uint32_t cbLeft = LNX_MAX_KERNEL_SIZE;
2143	while (cbLeft > 4096)
2144	{
2145	DBGFADDRESS HitAddrStart;
2146	rc = DBGFR3MemScan(pUVM, 0 /idCpu/, &CurAddr, cbLeft, 1 /uAlign/,
2147	s_abCfgNeedleStart, sizeof(s_abCfgNeedleStart) - 1, &HitAddrStart);
2148	if (RT_FAILURE(rc))
2149	break;
2150
2151	/* Check for the end marker which shouldn't be that far away. */
2152	DBGFR3AddrAdd(&HitAddrStart, sizeof(s_abCfgNeedleStart) - 1);
2153	DBGFADDRESS HitAddrEnd;
2154	rc = DBGFR3MemScan(pUVM, 0 /* idCpu */, &HitAddrStart, LNX_MAX_COMPRESSED_CFG_SIZE,
2155	1 /* uAlign */, s_abCfgNeedleEnd, sizeof(s_abCfgNeedleEnd) - 1, &HitAddrEnd);
2156	if (RT_SUCCESS(rc))
2157	{
2158	/* Allocate a buffer to hold the compressed data between the markers and fetch it. */
2159	RTGCUINTPTR cbCfg = HitAddrEnd.FlatPtr - HitAddrStart.FlatPtr;
2160	Assert(cbCfg == (size_t)cbCfg);
2161	rc = dbgDiggerLinuxCfgDecode(pThis, pUVM, &HitAddrStart, cbCfg);
2162	if (RT_SUCCESS(rc))
2163	break;
2164	}
2165
2166	/*
2167	* Advance.
2168	*/
2169	RTGCUINTPTR cbDistance = HitAddrStart.FlatPtr - CurAddr.FlatPtr + sizeof(s_abCfgNeedleStart) - 1;
2170	if (RT_UNLIKELY(cbDistance >= cbLeft))
2171	{
2172	LogFunc(("Failed to find compressed kernel config\n"));
2173	break;
2174	}
2175	cbLeft -= cbDistance;
2176	DBGFR3AddrAdd(&CurAddr, cbDistance);
2177
2178	}
2179
2180	return rc;
2181	}
2182
2183	/**
2184	* @copydoc DBGFOSREG::pfnInit
2185	*/
2186	static DECLCALLBACK(int) dbgDiggerLinuxInit(PUVM pUVM, void *pvData)
2187	{
2188	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
2189	Assert(!pThis->fValid);
2190
2191	/*
2192	* Assume 64-bit kernels all live way beyond 32-bit address space.
2193	*/
2194	pThis->f64Bit = pThis->AddrLinuxBanner.FlatPtr > UINT32_MAX;
2195
2196	pThis->hCfgDb = NULL;
2197
2198	/*
2199	* Try to find the compressed kernel config and parse it before we try
2200	* to get the symbol table, the config database is required to select
2201	* the method to use.
2202	*/
2203	int rc = dbgDiggerLinuxCfgFind(pThis, pUVM);
2204	if (RT_FAILURE(rc))
2205	LogFlowFunc(("Failed to find kernel config (%Rrc), no config database available\n", rc));
2206
2207	static const uint8_t s_abNeedle[] = "kobj";
2208	rc = dbgDiggerLinuxFindSymbolTableFromNeedle(pThis, pUVM, s_abNeedle, sizeof(s_abNeedle) - 1);
2209	if (RT_FAILURE(rc))
2210	{
2211	/* Try alternate needle (seen on older x86 Linux kernels). */
2212	static const uint8_t s_abNeedleAlt[] = "kobjec";
2213	rc = dbgDiggerLinuxFindSymbolTableFromNeedle(pThis, pUVM, s_abNeedleAlt, sizeof(s_abNeedleAlt) - 1);
2214	if (RT_FAILURE(rc))
2215	{
2216	static const uint8_t s_abNeedleOSuseX86[] = "nmi"; /* OpenSuSe 10.2 x86 */
2217	rc = dbgDiggerLinuxFindSymbolTableFromNeedle(pThis, pUVM, s_abNeedleOSuseX86, sizeof(s_abNeedleOSuseX86) - 1);
2218	}
2219	}
2220
2221	pThis->fValid = true;
2222	return VINF_SUCCESS;
2223	}
2224
2225
2226	/**
2227	* @copydoc DBGFOSREG::pfnProbe
2228	*/
2229	static DECLCALLBACK(bool) dbgDiggerLinuxProbe(PUVM pUVM, void *pvData)
2230	{
2231	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
2232
2233	/*
2234	* Look for "Linux version " at the start of the rodata segment.
2235	* Hope that this comes before any message buffer or other similar string.
2236	*/
2237	for (unsigned i = 0; i < RT_ELEMENTS(g_au64LnxKernelAddresses); i++)
2238	{
2239	DBGFADDRESS KernelAddr;
2240	DBGFR3AddrFromFlat(pUVM, &KernelAddr, g_au64LnxKernelAddresses[i]);
2241	DBGFADDRESS HitAddr;
2242	int rc = DBGFR3MemScan(pUVM, 0, &KernelAddr, LNX_MAX_KERNEL_SIZE, 1,
2243	g_abLinuxVersion, sizeof(g_abLinuxVersion) - 1, &HitAddr);
2244	if (RT_SUCCESS(rc))
2245	{
2246	char szTmp[128];
2247	char const *pszX = &szTmp[sizeof(g_abLinuxVersion) - 1];
2248	rc = DBGFR3MemReadString(pUVM, 0, &HitAddr, szTmp, sizeof(szTmp));
2249	if ( RT_SUCCESS(rc)
2250	&& ( ( pszX[0] == '2' /* 2.x.y with x in {0..6} */
2251	&& pszX[1] == '.'
2252	&& pszX[2] >= '0'
2253	&& pszX[2] <= '6')
2254	\|\| ( pszX[0] >= '3' /* 3.x, 4.x, ... 9.x */
2255	&& pszX[0] <= '9'
2256	&& pszX[1] == '.'
2257	&& pszX[2] >= '0'
2258	&& pszX[2] <= '9')
2259	)
2260	)
2261	{
2262	pThis->AddrKernelBase = KernelAddr;
2263	pThis->AddrLinuxBanner = HitAddr;
2264	return true;
2265	}
2266	}
2267	}
2268	return false;
2269	}
2270
2271
2272	/**
2273	* @copydoc DBGFOSREG::pfnDestruct
2274	*/
2275	static DECLCALLBACK(void) dbgDiggerLinuxDestruct(PUVM pUVM, void *pvData)
2276	{
2277
2278	}
2279
2280
2281	/**
2282	* @copydoc DBGFOSREG::pfnConstruct
2283	*/
2284	static DECLCALLBACK(int) dbgDiggerLinuxConstruct(PUVM pUVM, void *pvData)
2285	{
2286	PDBGDIGGERLINUX pThis = (PDBGDIGGERLINUX)pvData;
2287	pThis->IDmesg.u32Magic = DBGFOSIDMESG_MAGIC;
2288	pThis->IDmesg.pfnQueryKernelLog = dbgDiggerLinuxIDmsg_QueryKernelLog;
2289	pThis->IDmesg.u32EndMagic = DBGFOSIDMESG_MAGIC;
2290
2291	return VINF_SUCCESS;
2292	}
2293
2294
2295	const DBGFOSREG g_DBGDiggerLinux =
2296	{
2297	/* .u32Magic = */ DBGFOSREG_MAGIC,
2298	/* .fFlags = */ 0,
2299	/* .cbData = */ sizeof(DBGDIGGERLINUX),
2300	/* .szName = */ "Linux",
2301	/* .pfnConstruct = */ dbgDiggerLinuxConstruct,
2302	/* .pfnDestruct = */ dbgDiggerLinuxDestruct,
2303	/* .pfnProbe = */ dbgDiggerLinuxProbe,
2304	/* .pfnInit = */ dbgDiggerLinuxInit,
2305	/* .pfnRefresh = */ dbgDiggerLinuxRefresh,
2306	/* .pfnTerm = */ dbgDiggerLinuxTerm,
2307	/* .pfnQueryVersion = */ dbgDiggerLinuxQueryVersion,
2308	/* .pfnQueryInterface = */ dbgDiggerLinuxQueryInterface,
2309	/* .u32EndMagic = */ DBGFOSREG_MAGIC
2310	};
2311

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

source: vbox/trunk/src/VBox/Debugger/DBGPlugInLinux.cpp@ 62837

Download in other formats: