09 Oct 2016, by mrexodia
This is the seventh of (hopefully) many weekly digests. Basically it will highlight the things that happened to x64dbg and related projects during the week before.
Plugin page
There is now a wiki page available dedicated to x64dbg plugins. It contains various templates and also a list of plugins. If you wrote a plugin yourself, feel free to add it to the list!
Variable list will now be shown in the reference view
The command varlist will now show the available variables in the reference view instead of in the log.
Fixed a crash in the pluginload command
Previously the pluginload commands would not check the number of arguments and it would read in bad memory. See issue #1141 for more details.
Added undo in registers view
Atvaark added an ‘Undo’ option to revert register changes in pull request #1142.
Think there are too many entries in the disassembly context menu? You can now move menu entries you don’t use to the ‘More commands’ section to make your life less complicated. This also works for some other menus but some more work is required to make it possible everywhere.
Better character recognition in the info box
The info box will now recognize escaped characters in addition to printable characters.
Pull request #1145 added character recognition requested in issue #1128.
Goto origin in memory map
Atvaark has added the Goto -> Origin option in the memory map in pull request #1146. This will show you the memory page that EIP/RIP is currently in.
The branch lines in the sidebar are now highlighted when selecting the branch destination. This is in addition to the xref feature that was implemented some time ago. If you want xref analysis use the command analx, analyze a single function with the ‘A’ button or use the analr command. For more analysis commands, see the analysis section of the documentation.
Various updates to the mnemonic database
If you are looking for a quick description of every instruction you can use the ‘Show mnemonic brief’ (Ctrl+Alt+F1) option to get a brief description of every opcode. The mnemonic database used for this has been slightly updated and should give better results in some cases.
Open file/directory options for the source view
You can now open the file/directory of the source file you are currently debugging in to view the file in your favorite editor.
Next/Previous/Hide tab
The third and fourth(!!!) pull request by Atvaark this week (#1152 and #1153) added more flexibility with tabs. You can now easily hide tabs and switch between them.
Import/export database
It is now possible to use the dbload and dbsave commands to import/export databases to an arbitrary location. Once you have an exported the database you can import it in IDA with the x64dbgida plugin. This also works the other way around!
Better IsJumpGoingToExecute
The function that analyzes the flags to see if a jump is going to execute has been re-implemented and should now be faster. In addition to that the loop instruction is correctly analyzed now.
Usual stuff
That has been about it for this week again. If you have any questions, contact us on Telegram, Gitter or IRC. If you want to see the changes in more detail, check the commit log.
You can always get the latest release of x64dbg here. If you are interested in contributing, check out this page.
Finally, if someone is interested in hiring me to work on x64dbg more, please contact me!
Comments
04 Oct 2016, by torusrxxx
x64dbg has a complex architecture. There are three basic parts, namely DBG, BRIDGE and GUI, but in fact there is a fourth part, EXE. This is the main executable, it compiles into x64dbg.exe.
Bootstrapping
When the user starts x64dbg, it will follow this initialization path to get x64dbg running:
- WinMain
- BridgeInit
- LoadLibrary(“x64gui.dll”)
- Initialize global variables in the GUI
- LoadLibrary(“x64dbg.dll”)
- Initialize global variables in the DBG
- Load function pointers
- BridgeStart
Debugging
To start debugging, the GUI sends an init command to the DBG. Then the following things start:
Message passing from GUI to DBG
There are four methods to call DBG from GUI. They are commands, directly exported functions, bridge exported functions (messages) and DbgFunctions(). Currently the directly exported functions are frozen and no new ones should be added. The message flows for each way will be described below.
Commands dispatch
DbgCmdExec is relayed by the bridge to the DBG and eventually received by the cmdloop running in the command thread. This is done asynchronously (meaning DbgCmdExec will not wait until the command is completed).
DbgCmdExecDirect is relayed by the bridge to DBG and then directly in cmddirectexec. This will only return after the command is completed.
In both cases the command is parsed and dispatched to various registered command callbacks. A command callback is similar to main() functions. It can receive number of arguments (plus one), and pointer to every argument string.
Commands are registered in the registercommands function. If you want to get a total list of supported commands, or add your own, just go to that file. Make sure to put your command in the correct category and also make sure to add it to the documentation.
Directly exported functions
There are some legacy functions still unconverted to another method, these can be found in exports.h.
Export functions dispatch
Many Dbg*** functions are exported by the bridge. It then calls _dbg_sendmessage exported by DBG to pass information. Some Dbg*** functions have exports directly in DBG.
DbgFunctions
_dbgfunctions.cpp has a function table that is accessible by anyone. The GUI can call functions in DBG through this table directly.
Message flow from DBG to GUI
There are various Gui*** functions exported by the bridge. The control flow is described below:
- Gui*** export
- Bridge calls _gui_sendmessage
- Bridge calls Bridge::processMessage
- A long list of switch statements in
processMessage, basically to emit the corresponding signal. If you want to receive a system event, connect to one of the signals in Bridge::getBridge()
Important subsystems in GUI
Tables in GUI
There is three-level class architecture to support various tables. The first-level class is AbstractTableView, which only includes some basic functions. The second-level classes are Disassembly, HexDump and StdTable. They all inherit from AbstractTableView. Many basic and common functions are defined here, such as table painting, selection, content presentation and column reordering. The third-level classes inherit from the second-level classes. There are many third-level classes. The most common parent for these tables is StdTable.
There are two styles of context menu management. The traditional one builds actions in setupContextMenu and adds them into a menu object in contextMenuEvent. CPUStack uses this style currently. A newer way to manage context menu is to use MenuBuilder. You can see CPUDisassembly or this blog post for more details. It is the preferred way to manage context menu in newer tables, but it does not support non-table widgets out of the box. We want to convert traditional context menu systems into MenuBuilder to speed up development.
Configuration management
Configurations are stored in the Config() object which uses Utf8Ini in the bridge as its backend. When you want to add a new configuration, you have to modify the following files: Configuration.cpp and the SettingsDialog. If you are adding a color then you have to modify the AppearanceDialog as well. Config() can emit settings change signals.
Important subsystems in DBG
There are many subsystems in DBG. The following subsystems are important if you want to contribute:
threading.h
It includes various locks to prevent race condition. Without it, x64dbg will crash much more often. Don’t forget to acquire the lock when you are accessing a subsystem.
x64dbg.cpp
It registers all commands in x64dbg. The details of command processing is described above.
memory.h , module.h and thread.h, label.h and breakpoint.h, etc
They manages corresponding information of the debuggee.
scriptapi
It is intended to be used by plugins. It provides easy scripting experience for developers. x64dbg does not call any of these functions.
Comments
02 Oct 2016, by mrexodia
This is the sixth of (hopefully) many weekly digests. Basically it will highlight the things that happened to x64dbg and related projects during the week before.
Remove all breakpoints before detaching
When detaching x64dbg all breakpoints will be removed to prevent the debuggee from crashing when reaching breakpointed code.
Warnings when trying to set CIP to a non-executable page
When using the ‘Set New Origin Here’ option in the disassembler it will prompt you with a warning if the code there is not executable.
Fixed event filter plugin callbacks with Qt5
The PLUG_CB_WINEVENT callback is now working as intended again. This allows plugins to intercept and handle native window events that are usually handled by Qt. Multiline Ultimate Assembler uses this to handle hotkeys registered by the plugin. You can also use it to intercept mouse clicks and paint events for example.
All commands have been cleanly organized in separate source files (matching the categories in the documentation). This should help new contributers to find the code they are looking for more easily.
Import multiple patches
You can now import multiple patch files from the patch manager. Just select multiple files in the browse dialog and enjoy patching!
Adjust width of status label for translations
The debug status label will now scale to the biggest translation width.
Active view API
Issue #917 has been partially addresses with this. The following code allows you to query the active view. The titles are subject to change, but the class names shouldn’t change.
ACTIVEVIEW activeView;
GuiGetActiveView(&activeView);
printf("activeTitle: %s, activeClass: %s\n", activeView.title, activeView.className);
Highlight ud2 and ud2b as unusual instructions
The ud2 in addition to various other unusual instructions are marked in red to draw attention when encountered.
In his blog post, kao mentioned that the context menu in the register view was bloated. This has now been addresses and menu options are roughly ordered to how often they are used in practice. If you see more issues like this, please let us know here.
Before:
After:
Lots of code improvements
Various static analysis runs with Coverity, ReSharper and Visual Studio provided lots of small bugs or anomalies and these have been fixed.
Allow debugging of AnyCPU .NET files
Pull request #1124 addressed a bug where .NET files with AnyCPU would not load in x64dbg because their headers had the wrong PE architecture.
Clarified SetMemoryBPX command
The SetMemoryBPX command has been slightly changed. There was confusion in issue #1123 what read, write and access mean exactly. This has been clarified in the documentation and the correct type of memory breakpoint is now set for ‘Access’ when using the GUI.
Improved follow in memory map
When using the follow in memory map option it will now scroll to the entry requested, it will also show the memory map if you weren’t looking at it already.
Highlight active view in CPU
The view that is currently active will be highlighted with a thin black border.
Print symbolic name on expression command
When typing an unknown command in the command bar your text will be evaluated as an expression and the result will be printed. When the expression resolves to a symbolic address it will now also display the symbol name.
The disasm command is now more responsive.
Corrected width of the Hex short dump
The default width of the ‘hex short’ dump view didn’t show all values, this has now been corrected.
Fixed bug with endianness in the float register editor
Issue #1127 has been fixed. When showing FPU registers as big endian the editor would interpret the values incorrectly.
The plugin callback system didn’t have a separation on type, which meant that performance-critical locations that used plugin callbacks would pay for thing like menu callbacks. Every callback now has a separate list which solves the problem.
Type system
Issue #1108 that requests displaying types has been partially addressed. The basics of a type system have been implemented and you can now create and view types.
There is an example script available and the documentation is here. There will be a blog post later explaining the type system in more detail.
Fail assembling short jumps that don’t fit in 2 bytes
When assembling short jumps with keystone you could assemble short jumps that are not actually show. RaMMicHaeL has addressed this in pull request #1134.
Added plugin callback to filter symbols
Various symbols have no real meaning and these are filtered from view. Plugins can now register custom filter callbacks. This has been implemented by shamanas in pull request #1135.
When listing bookmarks your comments/labels will now be shown in the list. This was implemented by Atvaark in pull request #1136.
Use reference view for varlist
The varlist command will now show variables in the reference view instead of the console.
Allow allocation at a specified address
The alloc command now has an optional second parameter that allows you to specify an address to allocate memory at (similar to VirtualAlloc).
Use CIP per default in imageinfo
The imageinfo command now uses CIP when no address is specified.
Final words
That has been about it for this week again. If you have any questions, contact us on Telegram, Gitter or IRC. If you want to see the changes in more detail, check the commit log.
You can always get the latest release of x64dbg here. If you are interested in contributing, check out this page.
Finally, if someone is interested in hiring me to work on x64dbg more, please contact me!
Comments