The output of the expressions section changes even when running `context` multiple times after each other. The output variables in GDB are counted up when reexecuting the watched commands for example. The other sections don't change their output. This caused the history to be extended infinitely when using `ctxp` while having a `cwatch` command executed.
Special case the `expressions` context section in the history handling to avoid reevaluating the watched commands/expressions while browsing the history. This doesn't add the context output to the history when the expressions VALUES change somehow like it is done for the other sections, but since we cannot know if gdb counted up their output variable names from $1 to $2 or the value changed, this is a compromise.
* Fix#2549: block config.<name> assignments
This commit blocks `config.<param> = <value>` assignments as they should be done via `config.<param>.value = <value>`
instead.
* Update config.py
* add .value
* docs: Add missing testing toc entries and other minor formatting
* fix: Adjust nix dev shell packages to properly support jemalloc and qemu tests
* fix(jemalloc): Add more robust error handling to jemalloc commands and fix test
* fix: point JEMALLOC_PATH to correct jemalloc package path
* fix: Use correct aglib-compatible symbol resolution function
* fix: re-enable test_jemalloc_heap test and make slightly more forgiving
* Fix terminal width to 80 columns in tests
Set a `PWNDBG_IN_TEST`environment variable when running gdb in tests. Use the dimensions in `LINES` and `COLUMNS` when looking up the window size when that envvar is set.
This makes context output always be 80 columns wide which allows to compare to hardcoded output.
* Use `width` and `height` parameters for cmd window size
They are updated to the correct values when switching between tui and cli mode. So it's unnecessary to parse `info win`.
* Remove workarounds for terminal size in DEVELOPING
* Don't bother updating the env of the debuggee
* Add history of context output
Every context section is cached individually to allow to display older output again. You can scroll through the old context output using `contextprev` and `contextnext`.
This allows to "scroll" in TUI mode.
* Add button TUI window to control context history
* Simplify history status output generation
* Fix attachp tests when ptrace_scope is missing
Assume we can attach to any process when ptrace_scope is not available. This is the case in WSL2.
* Update tests/gdb-tests/tests/test_attachp.py
---------
Co-authored-by: Disconnect3d <dominik.b.czarnota@gmail.com>
* Make last-signal context section selectable
There exists a context section to display the reason for the last stop, but it wasn't registered anywhere.
Add it to the list of available sections to choose from.
* Fix default context sections "heap-tracker"
The section name is derived from the function name.
The function is called `context_heap_tracker` so `heap-tracker` wouldn't work:
```
pwndbg> set context-sections heap-tracker
Invalid section: heap-tracker, valid values: args, regs, disasm, stack, backtrace, code, expressions, ghidra, heap_tracker, threads
```
The validator doesn't appear to run for the default value :D
* Update help text to include all available context sections
* Replace bash with sleep in attachp commands to fix crash due to reading from terminal
* Update tests/gdb-tests/tests/test_attachp.py
* Update test_attachp.py
---------
Co-authored-by: Disconnect3d <dominik.b.czarnota@gmail.com>
* Fixed bug where plist()'s argument sentinel had incorrect type
* Fixed bug where if sentinel was different from zero, plist tried to dereference a null address
* Fixed bug where some error message in plist displayed 0x0x{address} due to incorrect formatting
* Added additional tests for plist's bugfixes
* Added types to hexdump()'s parameters
* hexdump(): upgraded format strings to f-strings, while fixing the offset when data is empty
* hexdump(): offset is now correctly computed and updated accross repeated calls to hexdump.
This fixes two bugs:
1. Offset was increased by the number of lines displayed by the hexdump() command,
which was incorrect, as some lines may be compressed;
2. When dumping a number of bytes that is not divisible by the number of bytes per line,
offset would be incorrectly updated.
* Linter
* hexdump: fixed type of retrieved config entry hexdump_group_use_big_endian
* Fixed test_hexdump_saved_address_and_offset test to account for random stack address
* Upgraded the linked-lists.c test program to 6 items in order to test the "deference-limit" setting
* Chain: changes to the deference-limit parameter were not reflected in the program due to an early cast to int()
* Linter
* Added support for `--offset` and `--count` for plist
* Added typing to commands.plist's prototype
* Updated documentation for plist tests
* Add informational message when plist --count is zero or under
Co-authored-by: Disconnect3d <dominik.b.czarnota@gmail.com>
* Handle case, in plist, where the default value for count is zero
---------
Co-authored-by: Disconnect3d <dominik.b.czarnota@gmail.com>
* Added types to hexdump()'s parameters
* hexdump(): upgraded format strings to f-strings, while fixing the offset when data is empty
* hexdump(): offset is now correctly computed and updated accross repeated calls to hexdump.
This fixes two bugs:
1. Offset was increased by the number of lines displayed by the hexdump() command,
which was incorrect, as some lines may be compressed;
2. When dumping a number of bytes that is not divisible by the number of bytes per line,
offset would be incorrectly updated.
* Linter
* hexdump: fixed type of retrieved config entry hexdump_group_use_big_endian
* Fixed test_hexdump_saved_address_and_offset test to account for random stack address
* Upgraded the linked-lists.c test program to 6 items in order to test the "deference-limit" setting
* Chain: changes to the deference-limit parameter were not reflected in the program due to an early cast to int()
* Linter
* Port Pwndbg to LLDB
* Fix splitting mistakes
* I love typos
* We already check for Python 3.11 manually
* Update pwndbg/dbg/lldb/__init__.py
Co-authored-by: patryk4815 <bux.patryk@gmail.com>
* Fix use of wrong variable name that was preventing the vmmap from being used in LLDB memory reads
* Use cached vmmap for `read_memory` in LLDB
* Fix Pwndbg CLI jank
* Try using SBAddress to resolve objfile names in LLDB vmmap
* Fix missing changes in `pwndbg.gdblib`
* Update pwndbg/aglib/arch.py
* Update pwndbg/aglib/arch.py
* Remove outdated comment about `pwndbg.aglib.arch`
* Update pwndbg/dbg/lldb/repl/io.py
Co-authored-by: Disconnect3d <dominik.b.czarnota@gmail.com>
* Fix lints
* Fix new test so that it points to `pwndbg.aglib.disasm`
* Fix lints
---------
Co-authored-by: patryk4815 <bux.patryk@gmail.com>
Co-authored-by: Disconnect3d <dominik.b.czarnota@gmail.com>
* Merge Arm tests
* Integrate MIPS, AArch64, RISC-V tests, and the rest of the ARM edges cases
* add fsbase/gsbase annotation test
* add function call / fin test
* lint
* Fix all bugs found - MIPS JAL, delay slots splits, arm/aarch unconditional jumps, and forcing targets when we know they go to the next instruction in memory, and add a bunch of comments. Arm now detects instructions that write to PC as branches too.
* Fix Arm exclusive stores
* comment update
* Fix arm/aarch64 post-indexed stores - add test
* AArch64 post-index stores
* Fix arm shifts
* comment typo
* Initial version of qemu-user tests
* Refactor testing files to reduce file duplication, introduce qemu-user-tests
* lint and edit github actions workflow file. Move old qemu-user tests to seperate directory
* Add iproute2 so ss command is available
* test ubuntu 24
* funkiness with current working directory...
* Further remote old test_qemu.sh and integrate into a Pytest fixture
* lint
* Disable ASLR, add test for aarch64 jumps
* Use Popen.kill() function to make sure it closes.
Co-authored-by: Disconnect3d <dominik.b.czarnota@gmail.com>
* qemu.kill() on the other fixture as well
* comment
* comment
* lint
* system test path stuff
* remove old try-catch block
* revert
* revert path change
* Use os._exit to pass return code, and move qemu-user tests above system tests because they run significantly faster
* lint
* Flush stdout before os._exit
* Comment out flaky check for the address of main in old qemu tests
* rename qemu-user to cross-arch
* rename qemu-user to cross-arch and hotfix to not run pytest when
cross-arch is used
* remove todo comment
* another comment
* Test pwndbg.gdblib.symbol.address is not None and revert setarch -R
* Revert os.exit change
* Revert os.exit change
* Revert os.exit change
* readd os.exit in new exit places
* lint
* rebase
* delete file introduced in rebase
* break up tests into 3 files to invoke separately. Update GitHub workflow, remove code duplication in existing test
* code coverage
* fix code coverage
* lint
* test difference between Ubuntu 22 and 24 in Kernel tests
* lint
---------
Co-authored-by: Disconnect3d <dominik.b.czarnota@gmail.com>
* basic go value dumping
* better error handling and misc improvements
* minor documentation changes
* satisfy mypy
* struct parsing and bug fix
* satisfy mypy
* deal with evacuated buckets
* better error message for invalid expressions
* convert bytearray to bytes before repr
* support for recursive types and better type dumping
* better QOL for go-dump command
* formatting options and debug/pretty printing
* add go dumping unit tests
* deal with differences in old go version
* lint
* old go versions missing any alias
* proper name dumping for go versions prior to 1.17
* lint
* go is being weird on CI
* warn instead of erroring
* function and interface dumping
* Add event system to the Debugger-agnostic API
* Move uses of `pwndbg.gdblib.events` to the Debugger-agnostic API
* Fix rebase mistakes
* Update pwndbg/commands/context.py
Co-authored-by: Disconnect3d <dominik.b.czarnota@gmail.com>
* Update context.py
* Update __init__.py
* Fix nits
---------
Co-authored-by: Disconnect3d <dominik.b.czarnota@gmail.com>
* basic binary ninja integration
* start centralizing integrations into uniform interface
* finish most integrations
* make binja integration contain all ida features
* allow switching servers
* remove unused functions
* format
* fix unit tests
* satisfy mypy + warn instead of error on invalid provider
* fix decompilation not working in function prologue
* better config options and il config
* lint
* fix bn_sym test
* add decomp command
* add bn_eval gdb function
* format binja_script.py
* satisfy mypy again
* doc update
* move integrations to separate subdir + make test_loads more versatile
* format
* forgot to push a file
* format
* highlight next instruction instead of previous one in decomp
* fix disassembly il level
* prefer function tags over data tags
* add bn_var function
* add variable labeling
* more configurability, documentation, bug fixes
* code cleanup
* format
* remove code duplication between chain.py and memory.py
* add short-lived cache for some integration functions
* use every register available to help with frame inference
* lint
* don't waste time checking irrelevant registers
* move performance hit in cache only to except case
* fix issue with conflict resolution
* lint
* fix rebase mistake
* Added types to hexdump()'s parameters
* hexdump(): upgraded format strings to f-strings, while fixing the offset when data is empty
* hexdump(): offset is now correctly computed and updated accross repeated calls to hexdump.
This fixes two bugs:
1. Offset was increased by the number of lines displayed by the hexdump() command,
which was incorrect, as some lines may be compressed;
2. When dumping a number of bytes that is not divisible by the number of bytes per line,
offset would be incorrectly updated.
* Linter
* hexdump: fixed type of retrieved config entry hexdump_group_use_big_endian
* Fixed test_hexdump_saved_address_and_offset test to account for random stack address
* Improve bin corruption checks
* Update pwndbg/heap/ptmalloc.py
* factor out and clean up bin corruption check
* check chunks even if bin is longer than limit
* add empty bin check
* lint.sh, remove testing if
* dont modify chain in check, allow corruption=0, cleanup
* typing, more reliable empty bin check
* cast params to int, otherwise not detected properly
* add regression test for corruption check
* lint.sh
---------
Co-authored-by: Gulshan Singh <gsingh2011@gmail.com>
Co-authored-by: Disconnect3d <dominik.b.czarnota@gmail.com>
* Fix caching for pwndbg.gdblib.elf functions
* Add test for cache clear priority
* Fix formatting in files
* switch priority to an enum instead of boolean
* fix logging for start event
* decouple gdblib.events import from lib.cache
* Move syscall number evaluation into instruction.py. This allows us to determine and display future syscalls
* Move string manipulation to color.disasm.py
* lint
* fix padding
* Fix x86 syscall
* comment
* disable debug mode
* Fix a test - we now remember previous syscalls as well
* Move x86 specific syscall logic to the x86 subclass
* lint
* @override decorator added to methods
* comments
* lint
* add test with emulation disabled for syscall annotation
* Fix x86/x86_64 edge cases with syscall register reading, and add test for emulation off for syscalls
* Update an outdated comment
* Tests depend on width of context banner
---------
Co-authored-by: Disconnect3d <dominik.b.czarnota@gmail.com>
* Modify python test scripts to work from nix
* Update utils.py
* address review feedback
---------
Co-authored-by: Disconnect3d <dominik.b.czarnota@gmail.com>
CptGibbon
patryk4815
peace-maker
Disconnect3d
Aaron Adams
patryk4815
jetchirag
Matt.
OBarronCS
Charles Fol
Jason An
Costinteo
k4lizen
Gulshan Singh