Hilda is a debugger which combines both the power of LLDB and iPython for easier debugging.
The name originates from the TV show "Hilda", which is the best friend of Frida. Both Frida and Hilda are meant for pretty much the same purpose, except Hilda takes the more " debugger-y" approach (based on LLDB).
Currently, the project is intended for iOS/OSX debugging, but in the future we will possibly add support for the following platforms as well:
- Linux
- Android
Since LLDB allows abstraction for both platform and architecture, it should be possible to make the necessary changes without too many modifications.
Pull requests are more than welcome 😊.
If you need help or have an amazing idea you would like to suggest, feel free to start a discussion 💬.
Requirements for remote iOS device (not required for debugging a local OSX process):
- Jailbroken iOS device
debugserver
in device's PATH- You can use this tool in order to obtain the binary
- After re-signing with new entitlements, you can put the binary in the following path:
/usr/bin/debugserver
In order to install please run:
xcrun python3 -m pip install --user -U hilda
You can may start a Hilda interactive shell by invoking any of the subcommand:
hilda launch /path/to/executable
- Launch given executable on current host
hilda attach [-p pid] [-n process-name]
- Attach to an already running process on current host (specified by either
pid
orprocess-name
)
- Attach to an already running process on current host (specified by either
hilda remote HOSTNAME PORT
- Attach to an already running process on a target host (sepcified by
HOSTNAME PORT
)
- Attach to an already running process on a target host (sepcified by
hilda bare
-
Only start an LLDB shell and load Hilda as a plugin.
-
Please refer to the following help page if you require help on the command available to you within the lldb shell:
As a cheatsheet, connecting to a remote platform like so:
platform connect connect://ip:port
... and attaching to a local process:
process attach -n proccess_name process attach -p proccess_pid
When you are ready, just execute
hilda
to move to Hilda's iPython shell.
-
Upon starting Hilda shell, you are greeted with:
Hilda has been successfully loaded! 😎
Use the p global to access all features.
Have a nice flight ✈️! Starting an IPython shell...
Here is a gist of methods you can access from p
:
-
hd
- Print an hexdump of given buffer
-
lsof
- Get dictionary of all open FDs
-
bt
- Print an improved backtrace.
-
disable_jetsam_memory_checks
- Disable jetsam memory checks, prevent raising:
error: Execution was interrupted, reason: EXC_RESOURCE RESOURCE_TYPE_MEMORY (limit=15 MB, unused=0x0).
when evaluating expression.
- Disable jetsam memory checks, prevent raising:
-
symbol
- Get symbol object for a given address
-
objc_symbol
- Get objc symbol wrapper for given address
-
inject
- Inject a single library into currently running process
-
rebind_symbols
- Reparse all loaded images symbols
-
poke
- Write data at given address
-
peek
- Read data at given address
-
peek_str
- Peek a buffer till null termination
-
stop
- Stop process.
-
cont
- Continue process.
-
run_for
- Run the process for given interval.
-
detach
- Detach from process. Useful in order to exit gracefully so process doesn't get killed while you exit
-
disass
- Print disassembly from a given address
-
file_symbol
- Calculate symbol address without ASLR
-
get_register
- Get value for register by its name
-
set_register
- Set value for register by its name
-
objc_call
- Simulate a call to an objc selector
-
call
- Call function at given address with given parameters
-
monitor
-
Monitor every time a given address is called
The following options are available:
regs={reg1: format} will print register values Available formats: x: hex s: string cf: use CFCopyDescription() to get more informative description of the object po: use LLDB po command User defined function, will be called like `format_function(hilda_client, value)`. For example: regs={'x0': 'x'} -> x0 will be printed in HEX format expr={lldb_expression: format} lldb_expression can be for example '$x0' or '$arg1' format behaves just like 'regs' option retval=format Print function's return value. The format is the same as regs format. stop=True force a stop at every hit bt=True print backtrace cmd=[cmd1, cmd2] run several LLDB commands, one by another force_return=value force a return from function with the specified value name=some_value use `some_name` instead of the symbol name automatically extracted from the calling frame override=True override previous break point at same location
-
-
show_current_source
- print current source code if possible
-
finish
- Run current frame till its end.
-
step_into
- Step into current instruction.
-
step_over
- Step over current instruction.
-
remove_all_hilda_breakpoints
- Remove all breakpoints created by Hilda
-
remove_hilda_breakpoint
- Remove a single breakpoint placed by Hilda
-
force_return
- Prematurely return from a stack frame, short-circuiting exection of newer frames and optionally yielding a specified value.
-
proc_info
- Print information about currently running mapped process.
-
print_proc_entitlements
- Get the plist embedded inside the process' __LINKEDIT section.
-
bp
- Add a breakpoint
-
show_hilda_breakpoints
- Show existing breakpoints created by Hilda.
-
save
- Save loaded symbols map (for loading later using the load() command)
-
load
- Load an existing symbols map (previously saved by the save() command)
-
po
-
Print given object using LLDB's po command Can also run big chunks of native code:
po('NSMutableString *s = [NSMutableString string]; [s appendString:@"abc"]; [s description]')
-
-
globalize_symbols
- Make all symbols in python's global scope
-
jump
- jump to given symbol
-
lldb_handle_command
- Execute an LLDB command For example: lldb_handle_command('register read')
-
objc_get_class
- Get ObjC class object
-
CFSTR
- Create CFStringRef object from given string
-
ns
- Create NSObject from given data
-
from_ns
- Create python object from NS object.
-
evaluate_expression
-
Wrapper for LLDB's EvaluateExpression. Used for quick code snippets.
Feel free to use local variables inside the expression using format string. For example: currentDevice = objc_get_class('UIDevice').currentDevice evaluate_expression(f'[[{currentDevice} systemName] hasPrefix:@"2"]')
-
-
import_module
- Import & reload given python module (intended mainly for external snippets)
-
unwind
- Unwind the stack (useful when get_evaluation_unwind() == False)
-
set_selected_thread
- sets the currently selected thread, which is used in other parts of the program, such as displaying disassembly or checking registers. This ensures the application focuses on the specified thread for these operations.
-
wait_for_module
- Wait for a module to be loaded (
dlopen
) by checking if given expression is contained within its filename
- Wait for a module to be loaded (
All these methods are available from the global p
within the newly created IPython shell. In addition, you may invoke any of the exported APIs described in the Python API
Sometimes accessing the Python API can be tiring, so we added some magic functions to help you out!
%objc <className>
- Equivalent to:
className = p.objc_get_class(className)
- Equivalent to:
%fbp <filename> <addressInHex>
- Equivalent to:
p.file_symbol(addressInHex, filename).bp()
- Equivalent to:
- F7: Step Into
- F8: Step Over
- F9: Continue
- F10: Stop
The global cfg
used to configure various settings for evaluation and monitoring.
These settings include:
evaluation_unwind_on_error
: Whether to unwind on error during evaluation. (Default:False
)evaluation_ignore_breakpoints
: Whether to ignore breakpoints during evaluation. (Default:False
)nsobject_exclusion
: Whether to excludeNSObject
during evaluation, reducing IPython autocomplete results. ( Default:False
)objc_verbose_monitor
: When set toTrue
, usingmonitor()
will automatically print Objective-C method arguments. ( Default:False
)
Hilda contains a minimal UI for examining the target state. The UI is divided into views:
- Registers
- Disassembly
- Stack
- Backtrace
This UI can be displayed at any time be executing:
ui.show()
By default step_into
and step_over
will show this UI automatically.
You may disable this behaviour by executing:
ui.active = False
Attentively, if you want to display UI after hitting a breakpoint, you can register ui.show
as callback:
p.symbol(0x7ff7b97c21b0).bp(ui.show)
Try playing with the UI settings by yourself:
# Disable stack view
ui.views.stack.active = False
# View words from the stack
ui.views.stack.depth = 10
# View last 10 frames
ui.views.backtrace.depth = 10
# Disassemble 5 instructions
ui.views.disassembly.instruction_count = 5
# Change disassembly syntax to AT&T
ui.views.disassembly.flavor = 'att'
# View floating point registers
ui.views.registers.rtype = 'float'
# Change addresses print color
ui.colors.address = 'red'
# Change titles color
ui.color.title = 'green'
Hilda provides a comprehensive API wrappers to access LLDB capabilities. This API may be used to access process memory, trigger functions, place breakpoints and much more!
Also, in addition to access this API using the Hilda shell, you may also use pure-python script using any of the create_hilda_client_using_*
APIs.
Consider the following snippet as an example of such usage:
from hilda.launch_lldb import create_hilda_client_using_attach_by_name
# attach to `sysmond`
p = create_hilda_client_using_attach_by_name('sysmond')
# allocate 10 bytes and print their address
print(p.symbols.malloc(10))
# detach
p.detach()
Please note this script must be executed using xcrun python3
in order for it to be able to access LLDB API.
In Hilda, almost everything is wrapped using the Symbol
Object. Symbol is just a nicer way for referring to addresses
encapsulated with an object allowing to deref the memory inside, or use these addresses as functions.
In order to create a symbol from a given address, please use:
s = p.symbol(0x12345678)
# the Symbol object extends `int`
True == isinstance(s, int)
# print the un-shifted file address
# (calculating the ASLR shift for you, so you can just view it in IDA)
print(s.file_address)
# or.. if you know the file address, but don't wanna mess
# with ASLR calculations
s = p.file_symbol(0x12345678)
# peek(/read) 20 bytes of memory
print(s.peek(20))
# write into this memory
s.poke('abc')
# let LLDB print-object (it should guess the type automatically
# based on its memory layout)
print(s.po())
# or you can help LLDB with telling it its type manually
print(s.po('char *'))
# jump to `s` as a function, passing (1, "string") as its args
s(1, "string")
# change the size of each item_size inside `s` for derefs
s.item_size = 1
# *(char *)s = 1
s[0] = 1
# *(((char *)s)+1) = 1
s[1] = 1
# symbol inherits from int, so all int operations apply
s += 4
# change s item size back to 8 to store pointers
s.item_size = 8
# *(intptr_t *)s = 1
s[0] = 1
# storing the return value of the function executed at `0x11223344`
# into `*s`
s[0] = p.symbol(0x11223344)() # calling symbols also returns symbols
# attempt to resolve symbol's name
print(p.symbol(0x11223344).lldb_symbol)
# monitor each time a symbol is called into console and print its backtrace (`bt` option)
# this will create a scripted breakpoint which prints your desired data and continue
s.monitor(bt=True)
# you can also:
# bt -> view the backtrace
# regs -> view registers upon each call in your desired format
# retval -> view the return value upon each call in your desired format
# cmd -> execute a list of LLDB commands on each hit
s.monitor(regs={'x0': 'x'}, # print `x0` in HEX form
retval='po', # use LLDB's `po` for printing the returned value
bt=True, # view backtrace (will also resolve ASLR addresses for you)
cmd=['thread list'], # show thread list
)
# we can also just `force_return` with a hard-coded value to practically disable
# a specific functionality
s.monitor(force_return=0) # cause the function to always return `0`
# as for everything, if you need help understanding each such feature,
# simply execute the following to view its help (many such features even contain examples)
s.monitor?
# create a scripted_breakpoint manually
def scripted_breakpoint(hilda, *args):
# like everything in hilda, registers are also
# just simple `Symbol` objects, so feel free to
# use them to your heart's content :)
if hilda.registers.x0.peek(4) == b'\x11\x22\x33\x44':
hilda.registers.x0 = hilda.symbols.malloc(200)
hilda.registers.x0.poke(b'\x22' * 200)
# just continue the process
hilda.cont()
s.bp(scripted_breakpoint)
# Place a breakpoint at a symbol not yet loaded by it's name
p.bp('symbol_name')
# In case you need to specify a specific library it's loaded from
p.bp('symbol_name', module_name='ModuleName')
Usually you would want/need to use the symbols already mapped into the currently running process. To do so, you can
access them using symbols.<symbol-name>
. The symbols
global object is of type SymbolsJar
, which is a wrapper
to dict
for accessing all exported symbols. For example, the following will generate a call to the exported
malloc
function with 20
as its only argument:
x = p.symbols.malloc(20)
You can also just write their name as if they already were in the global scope. Hilda will check if no name collision exists, and if so, will perform the following lazily for you:
x = malloc(20)
# is equivalent to:
malloc = p.symbols.malloc
x = malloc(20)
Sometimes you don't really know where to start your research. All you have is just theories of how your desired exported symbol should be called (if any).
For that reason alone, we have the rebind_symbols()
command - to help you find the symbol you are looking for.
p.rebind_symbols() # this might take some time
# find all symbols prefixed as `mem*` AND don't have `cpy`
# in their name
jar = p.symbols.startswith('mem') - p.symbols.find('cpy')
# filter only symbols of type "code" (removing data global for example)
jar = jar.code()
# monitor every time each one is called, print its `x0` in HEX
# form and show the backtrace
jar.monitor(regs={'x0': 'x'}, bt=True)
The same as symbols applies to Objective-C classes name resolution. You can either:
d = NSDictionary.new() # call its `new` selector
# which is equivalent to:
NSDictionary = p.objc_get_class('NSDictionary')
d = NSDictionary.new()
# Or you can use the IPython magic function
%objc
NSDictionary
This is possible only since NSDictionary
is exported. In case it is not, you must call objc_get_class()
explicitly.
As you can see, you can directly access all the class' methods.
Please look what more stuff you can do as shown below:
# show the class' ivars
print(NSDictionary.ivars)
# show the class' methods
print(NSDictionary.methods)
# show the class' proprties
print(NSDictionary.properties)
# view class' selectors which are prefixed with 'init'
print(NSDictionary.symbols_jar.startswith('-[NSDictionary init'))
# you can of course use any of `SymbolsJar` over them, for example:
# this will `po` (print object) all those selectors returned value
NSDictionary.symbols_jar.startswith('-[NSDictionary init').monitior(retval='po')
# monitor each time any selector in NSDictionary is called
NSDictionary.monitor()
# `force_return` for some specific selector with a hard-coded value (4)
NSDictionary.get_method('valueForKey:').address.monitor(force_return=4)
# capture the `self` object at the first hit of any selector
# `True` for busy-wait for object to be captured
dictionary = NSDictionary.capture_self(True)
# print a colored and formatted version for class layout
dictionary.show()
In order to work with ObjC objects, each symbol contains a property called
objc_symbol
. After calling, you can work better with each object:
dict = NSDictionary.new().objc_symbol
dict.show() # print object layout
# just like class, you can access its ivars, method, etc...
print(dict.ivars)
# except now they have values you can view
print(dict._ivarName)
# or edit
dict._ivarName = value
# and of course you can call the object's methods
# hilda will checks if the method returned an ObjC object:
# - if so, call `objc_symbol` upon it for you
# - otherwise, leave it as a simple `Symbol` object
arr = dict.objectForKey_('keyContainingNSArray')
# you can also call class-methods
# hilda will call it using either the instance object,
# or the class object respectively of the use
newDict = dict.dictionary()
# print the retrieved object
print(arr.po())
Also, working with Objective-C objects like this can be somewhat exhausting, so we created the ns
and from_ns
commands so you are able to use complicated types when parsing values and passing as arguments:
import datetime
# using the `ns` command we can just pass a python-native dictionary
function_requiring_a_specfic_dictionary(ns({
'key1': 'string', # will convert to NSString
'key2': True, # will convert to NSNumber
'key3': b'1234', # will convert to NSData
'key4': datetime.datetime(2021, 1, 1) # will convert to NSDate
}))
# and also parse one
normal_python_dict = p.cf({
'key1': 'string', # will convert to NSString
'key2': True, # will convert to NSNumber
'key3': b'1234', # will convert to NSData
'key4': datetime.datetime(2021, 1, 1) # will convert to NSDate
}).py()
On last resort, if the object is not serializable for this to work, you can just run pure Objective-C code:
# let LLDB compile and execute the expression
abc_string = p.evaluate_expression('[NSString stringWithFormat:@"abc"]')
# will print "abc"
print(abc_string.po())
Snippets are extensions for normal functionality used as quick cookbooks for day-to-day tasks of a debugger.
They all use the following concept to use:
from hilda.snippets import snippet_name
snippet_name.do_domething()
For example, XPC sniffing can be done using:
from hilda.snippets import xpc
xpc.sniff_all()
This will monitor all XPC related traffic in the given process.
Please run the tests as follows before submitting a PR:
xcrun python3 -m pytest