- Rename SetThreadDescription to SetWin10ThreadDescription, to clarify that this function isn't actually useful to anyone. (ha ha, OK... but seriously, the same name is also used by the Windows SDK which apparently takes precedence and gets added as a static import, making it impossible to start the debugger on OSes other than Windows 10)
- Thread names are a good idea and they even kind of work on older Windows versions with NtQueryInformationThread(ThreadQuerySetWin32StartAddress), which is what e.g. Process Explorer and Process Hacker use. What *doesn't* work so well is lambdas. Added static functions SymbolsThread() and SourceLinesThread() to replace these. (before: x64dbg.dll!<lambda_fc00d3fb731b14a9b4857ac068d657c4>::<lambda_invoker_cdecl>. after: x64dbg.dll!SymbolSourceDIA::SymbolsThread). These should probably be file statics instead of class members, but they need access to private class functions
- GetModuleHandleA -> GetModuleHandleW. The former just calls the latter but with an extra string allocation and pointless unicode conversion
- Fix pedantic Clang warnings about member initialization order in ctor
- Qualify type name in call to virtual function in destructor, as this will be statically resolved and won't call any potential future implementations in derived classes (this can be further 'fixed' by making either the function or the class final so you'll get a compile time error if you try to do this later)
- Obtain the directory directly using RtlImageDirectoryEntryToData and ditch TitanEngine conversion helpers
- Use OFTs instead of FTs if possible, with FTs only as fallback
- Answer the pop quiz questions in comments re: ntdll loader behaviour and handle these cases appropriately
- Use THUNK_VAL() to obtain OFT/FT values independent of process and file bitness
- Always use ULONG64 for AddressOfData to be able to test for IMAGE_ORDINAL_FLAG64. Also return ULONG64 from RvaToVa(), and rva2offset too as a result of this. This makes these functions compatible with both 32 and 64 bit files regardless of process bitness. There shouldn't be any functional changes due to this, otherwise will revert/fix
- Require an import by name to have a non-null name in addition to not having the ordinal flag set. Otherwise treat it as an import by ordinal
- The ordinal value of an import by ordinal is obtained by (val & 0xffff), not (val &= ~ordinalFlag). The ordinal flag is now always removed to ensure the RVA is valid
- Give imports by ordinal a 'name' the same way dbghelp does, e.g. Ordinal57. Previously imports by ordinal were not being shown in the Symbols tab due to having no name. TODO: if we have the PDB for the file being imported from, we can overwrite or append the real function name later using the importee's export directory
- RvaToVa(): assert that RVA 0 always returns VA 0, because if this isn't the case something is seriously messed up
- symbolsourcedia.h: Add _global.h #include to prevent various macros like WINVER and _WIN32_WINNT from being redefined because Windows.h was indirectly included first
- Add ImageNtHeaders() (clone of RtlImageNtHeaderEx which doesn't exist on XP) to obtain PE headers given a VA
- Add HEADER_FIELD() and THUNK_VAL() macros to module.h to allow accessing header fields independent of process and file bitness
- Add IMAGE_NT_HEADERS pointer to MODINFO, since anything related to parsing PEs needs this struct
- Read PE headers in GetModuleInfo(). Currently the headers are being parsed every time a TitanEngine helper function is called, the goal is to reduce this to once per module load
- GetModuleInfo(): eliminate all TitanEngine calls now that we have the headers
- Add RvaToVa() for SEC_COMMIT mappings. This can simultaneously serve as replacement for rva2offset helpers (pass base = 0). Preferably SEC_IMAGE should be used though as that way neither of these would be needed
- ReadExportDirectory(): use RtlImageDirectoryEntryToData() to obtain a PIMAGE_EXPORT_DIRECTORY and its size in one go to eliminate TitanEngine helper calls and RVA to offset conversions
- Answer burning questions re: Windows loader behaviour when parsing exports in comments
- (Minor) fix '>= 0' comparison against unsigned as this will always evaluate to true
- Add comment re: PDB search path order since it's wrong atm but I'm too scared of breaking something if I change this code myself
debugLoopFunction: set fdProcessInfo->hProcess and fdProcessInfo->hThread to NULL as these shouldn't be used after this point. The actual CloseHandle calls on these two handles are done by kernel32!ContinueDebugEvent immediately after cbExitProcess
- CloseHandle() the fdProcessInfo->hProcess and fdProcessInfo->hThread handles and set them to NULL if CreateProcess was called (i.e. we are not attaching) just before entering the debug loop
- cbCreateProcess(): set fdProcessInfo->hProcess, fdProcessInfo->hThread and varset("$hp") to the correct handles prior to doing anything else
* Browse dialog and goto dialog supports auto-complete
* don't use unicode string size
* Auto complete only when expression is valid symbol name
* use dbgfunctions for better flexibility and performance
* buffer last auto complete
* disable auto completion
- Add LASTSTATUS struct
- Add LastStatus members to REGDUMP and THREADALLINFO
- Add ThreadGetLastStatus()/ThreadGetLastStatusTEB()
- Make "laststatus" a supported pseudo-register in isregister() / getregister() / setregister() similar to "lasterror"
- _dbg_getregdump(): copy the full name of the last NTSTATUS value
- ThreadGetList(): add the last status value to the thread list for each thread
- TraceRecordManager: account for the size change of REGDUMP to keep REGDUMPWORD the same size
Public SDK changes:
- Add LASTSTATUS struct
- Add REGDUMP_V2 struct with LASTSTATUS member
- Add DbgGetRegDumpEx(REGDUMP* regdump, size_t size);
Internal changes:
- Change typedef of _dbg_getregdump to add a size parameter
- Make DbgGetRegDump() pass sizeof(REGDUMP) to _dbg_getregdump to preserve existing behaviour. DbgGetRegDumpEx() forwards the size that was passed to it
* run trace file format
* record opcode
* Successfully recorded sample run trace
* fixed order of thread id and opcode
* use capstone in run trace
* Revised format;Stop tracing when stop debug;Changed ext name
* trace browser(draft)
* Lower bound
* Lower bound
* implemented more funcitons in trace file reader
* Initial trace browser
* trace browser works for single-page traces
* fixed page fault
* Multi-selection, fixed page faults
* copy disassembly
* resize columns
* address label;follow in disassembly
* highlight
* history,comment,breakpoint in trace view
* stricter validation to prevent buffer overflow
* MAX_MEMORY_OPERANDS=32
* fixing bug in memory access count
* Temporary info menu to view registers & memory
* assumed to fix thread id bug
* live update trace view
* Fixed a bug with registers recording (similar to thread id bug)
* Search for constant in trace file
* Fixed bugs with memory operands recording
* File header for trace file; Auto update trace viewer
* fix x64dbg_translations.pro
* Default filename for trace; Start trace from trace view
* Switch to Qt JSON
* Copy selection, file offset and RVA; recent files
* Properly implement MRU menu
* shortcut for tracing
* Fix file names with comma
* added interaction with tab closing
* change default directory for trace file
* fix minor issue
* DBG: added range checks to operand access
- previously, some instructions could trigger the `DebugBreak` path in `Zydis::operator[]`
* GUI: removed redundant semicolon
- While at it, added branch info logic to disassembler class
- Thus reduce direct checks by mnemonic in GUI and analysis code
- Replaced direct disassembler struct access with disassembler class calls where trivially possible
- Removed workarounds for empty segment registers
- Temp. disabled `cbInstrCapstone` command
- Temp. disabled flag stuff in `QBeaEngine`
Duncan Ogilvie
torusrxxx
Mattiwatti
ZehMatt
Mack Stump
Joel Höner
roL
Rajarshi Vaidya aka gmastergreatee