This is a fork of GEF that includes three major improvements:
- Adds heuristic commands for kernel debugging without requiring a symbolized
vmlinux(forqemu-system, supports Linux kernel 3.x-7.x). - Expands support to many architectures (for
qemu-user). - Provides heap dump commands for multiple memory allocators.
Numerous other commands have been added and enhanced. Enjoy!
- Recommended: Ubuntu 24.04-26.04, Debian stable, and Arch Linux.
- May work, but not recommended: Ubuntu 20.04-23.10.
The installer has been refactored. I recommend re-installing. (2026-10-17)
-
Run the following command:
# For various reasons, I recommend installing it as the root user wget -q https://raw.githubusercontent.com/bata24/gef/dev/install-uv.sh -O- | sudo sh # For non-root user wget -q https://raw.githubusercontent.com/bata24/gef/dev/install-uv.sh -O- | sh # Install GEF in a different directory wget -q https://raw.githubusercontent.com/bata24/gef/dev/install-uv.sh -O- | sudo env GEF_INSTALL_DIR="/path/to/gef" sh wget -q https://raw.githubusercontent.com/bata24/gef/dev/install-uv.sh -O- | env GEF_INSTALL_DIR="/path/to/gef" sh
- Notes
- By default,
gef.pyis installed to$HOME/.gef/gef.py. - The required Python packages are installed to
$HOME/.gef/venv-gef. - The installer selects CPython matching GDB's Python major and minor version.
- GEF queries the environment at startup; activation is not required.
- An existing environment with a different Python version must be recreated before installation.
- GEF's installation directory is registered in
$HOME/.gdbinit. - For more installation options and non-
rootuser's limitations, see docs/FAQ.md.
- By default,
- Notes
-
Or, quick trial (no installation):
wget https://raw.githubusercontent.com/bata24/gef/dev/gef.py echo "source $(pwd)/gef.py" >> ~/.gdbinit
- Notes
- Most features work fine even without external tools.
- Extra Python packages are optional; commands requiring them remain unavailable until installed.
- Notes
python3 "$HOME/.gef/gef.py" --upgrade- Note
- If you get errors after upgrading, it may be due to old config. Try renaming
$HOME/.gef.rc. - Restart GDB after upgrading. Old
gef.venv.confandskip-venv-checkfiles are no longer used. - If GDB's Python major or minor version changes, recreate
venv-gefwith a matching Python and reinstall its packages.
- If you get errors after upgrading, it may be due to old config. Try renaming
rm -rf "$HOME/.gef"
rm -f "$HOME/.gef.rc"
rm -rf /tmp/gef
sed -i -e '/from gef import/d' "$HOME/.gdbinit"If you used GEF_INSTALL_DIR, remove that directory instead of $HOME/.gef.
Please refer to install-uv.sh for installation requirements.
- Standard debugging
- Attaching to a running process
- Attaching to a process in an isolated namespace (e.g., attaching from outside a Docker container)
- Connecting to Gdbserver
- Connecting to the GDB stub of Qemu-system
- Connecting to the GDB stub of Qemu-user
- Connecting to the GDB stub of Intel Pin
- Connecting to the GDB stub of Intel SDE
- Connecting to the GDB stub of Qiling framework
- Connecting to the GDB stub of KGDB (requires GDB version 12 or later)
- Connecting to the GDB stub of VMWare
- Connecting to the GDB stub of Wine
- Debugging with Record and replay (
rr replay)
For a comprehensive list and additional details, see docs/SUPPORTED-MODE.md.
For the usage and options of each command, see docs/COMMANDS.md.
pagewalk, kvmmap
pagewalk: dumps page tables.- x64 (Supported: 4-Level/5-Level Paging)
- x86 (Supported: PAE/Non-PAE)
- ARM64 (Supported: Cortex-A only, EL0-EL3, Stage1-2)
- ARM v8.7 base. 32bit mode is NOT supported.

- Here is a sample of each level pagewalk from HITCON CTF 2018
super_hexagon. 
- Secure memory scanning is also supported, but you have to break in the secure world.

- Pseudo memory map without detailed flags and permissions can be output even in the normal world (when OP-TEE).

- ARM (Supported: Cortex-A only, LPAE/Non-LPAE, PL0/PL1/PL2)
- RISCV64 / RISCV32
- MIPS32 / MIPS64 (Supported: Linux only, 2-Level to 4-Level, 4KB/16KB/64KB page)
- 64-bit PTE on MIPS32 (XPA, etc.) is NOT supported.
- PPC32 / PPC64 (Supported: POWER Radix/Hash, G3/G4, e500, 44x, 4KB/64KB page)
- POWER Hash, e500 and 44x are Linux only. G3/G4 shows only the pages in the hash table.
- S390X (Supported: 2-Level to 5-Level, 4KB/1MB/2GB page)
- The kernel and user address spaces overlap, so both are shown.
- LOONGARCH64 (Supported: 2-Level to 5-Level, 4KB/16KB/64KB page, DMW)
- SPARC32 (Supported: SRMMU, 4KB/256KB/16MB page)
kvmmap: prints kernel memory map.
qreg, msr, cet, vbar, gdtinfo, idtinfo, read-system-register-for-qemu-arm, read-system-register-for-kgdb, switch-el
qreg: displays the register values from qemu-monitor (allows getting values like$cseven under qemu 2.x).msr: reads/writes MSR (Model Specific Registers) value by embedding/executing dynamic assembly.cet: displays Intel CET settings.vbar: displays ARM/ARM64 vector table.gdtinfo: pretty prints GDT entries. If userland, show sample entries.idtinfo: pretty prints IDT entries. If userland, show sample entries.read-system-register-for-qemu-arm: reads system register for oldqemu-system-arm.read-system-register-for-kgdb: reads system register for kgdb (x64/ARM64 only).switch-el: switches EL (Exception Level).- Supported on ARM64.
kbase, kversion, kcmdline, kcurrent, kchecksec, kmagic, kpercpu, ks-selftest
kbase: displays the kernel base address.kversion: displays the kernel version.kcmdline: displays the kernel cmdline used at boot time.kcurrent: displays current task address.kchecksec: checks kernel security.kmagic: displays useful addresses in the kernel.kpercpu: resolves the per-cpu variables and the per-cpu areas.ks-selftest: compares the results of kernel-address heuristic finders with kallsyms.
v2p, p2v, xp, page, slab-virtual, pageinfo, highmem-dump
v2p/p2v: displays the transformation between virtual addresses and physical addresses.xp: is a shortcut for physical memory dump.page: displays the transformation between astruct pageand its virtual/physical address.slab-virtual: displays the transformation between slab-meta and its slab-data/struct pageaddress (forCONFIG_SLAB_VIRTUAL=y).pageinfo: dumpsstruct page->{flags,page_type}.highmem-dump: dumpsHighMemmappings.
ksymaddr-remote, ksymaddr-remote-apply, vmlinux-to-elf-apply, ktypes, ktypes-load, kload, kmod-load
ksymaddr-remote: displays kallsyms information from scanning kernel memory.ksymaddr-remote-apply/vmlinux-to-elf-apply: applies kallsyms information obtained byksymaddr-remoteorvmlinux-to-elfto gdb.

- Once you get a symbolized pseudo ELF file, you can reuse and apply it automatically even after rebooting qemu-system.
vmlinux-to-elf-applyandksymaddr-remote-applyprovide almost the same functionality.vmlinux-to-elf-apply: Requires installation of external tools. Createvmlinuxwith symbols.ksymaddr-remote-apply: Requires no external tools. Create a blank ELF with embedded symbols only.
ktypes: displays kernel type information from scanning kernel memory.ktypes-load: loads kernel type information from scanning kernel memory.kload: loadsvmlinuxwithout a load address.- It is useful if you have a
vmlinuxwithdebuginfoat hand.
- It is useful if you have a
kmod-load: loads the kernel module without a load address.- It is useful if you have a kernel module with
debuginfoat hand.
- It is useful if you have a kernel module with
ktask, ksched, kcred, kmem-task, kbt
ktask: displays each task's address.
- It also displays the memory map of the userland process.

- It also displays the register values saved on the kstack of the userland process.

- It also displays the file descriptors of the userland process.

- It also displays the signal handlers of the userland process.

- It also displays the namespaces of the userland process.

- It also displays the seccomp-filter.

ksched: dumps the scheduler state of each task (cpu, state, class, priority, policy).kcred: dumps the credentials of each task.kmem-task: dumps the userland memory of the specified process through its own page table, likexortelescope.kbt: scans inactive tasks' kernel stacks from the saved stack pointer for text addresses.
kconfig, ksyscalls, ksysctl, kmod, kops, kcdev, kbdev, kfilesystems, kmount, kclock-source, kdmesg, kpipe, kpagecache, kbpf, ktimer, kpcidev, kipcs, kdevio, kdmabuf, kirq, knetdev, kworkqueue, kwalk, kpath, kvfs, ksock, kskb, klsm, knft, kio-uring, kuffd, krefs, kkeyring, kcfi-hash, kprocfs, kcgroup, kbinder, klocks
kconfig: dumps the kernel config if available.ksyscalls: displays the system call table.ksysctl: dumps the sysctl parameters.kmod: displays each module's address.kops: displays each operation's member.kcdev: displays information for each character device.kbdev: displays information for each block device.kfilesystems: dumps supported file systems.kmount: dumps the mount tree of each mount namespace.kclock-source: dumps the clocksource list.kdmesg: dumps the ring buffer of the dmesg area.kpipe: displays information for each pipe.kpagecache: dumps file contents from the page cache.kbpf: dumps the BPF information.ktimer: dumps the timer.kpcidev: dumps the PCI devices.kipcs: dumps IPCs information (System V semaphore, message queue and shared memory).kdevio: dumps I/O-port and I/O-memory information.kdmabuf: dumps DMA-BUF information.kirq: dumps irq information.knetdev: displays net devices.kworkqueue: dumps workqueue items and inspects embedded work_struct objects.kwalk: dumps the entries held by the well-known kernel data structures.kpath: reconstructs the pathname of a dentry, path, file or mount.kvfs: displays the VFS object graph of a file descriptor or VFS object.ksock: walks from a file descriptor (or a raw struct sock) through socket, sock and its skb queues.kskb: parses a single sk_buff and shows its buffer layout, refcount and fragment information.klsm: dumps the registered Linux Security Module hooks.knft: dumps the nftables (netfilter) object graph.kio-uring: displays the kernel-side io_uring object graph.kuffd: displays the kernel-side userfaultfd object graph (registered ranges and the threads waiting on faults).krefs: searches the kernel pointers that reference the specified address.kkeyring: displays the keyrings referenced by tasks, or inspects an arbitrary struct key.kcfi-hash: lists the functions that share a kCFI type hash with a call target.kprocfs: displays the proc_dir_entry tree. It helps to find the entries a rootkit hides from /proc.kcgroup: displays the cgroup hierarchies, the cgroups of each task and the settings of the main controllers.kbinder: displays the binder_proc, thread, node, ref and transaction of Android binder.klocks: lists task lock records and automatically resolves mutex/rwsem owners and waiters.
slub-dump, slab-dump, slob-dump, slub-tiny-dump, slab-contains, kmem-cache-alias, buddy-dump, buddy-contains, vmalloc-dump, kobj
slub-dump: dumps slub free-list.- Supported on x64/x86/ARM64/ARM +
SLUB+ no-symbol + kASLR, and RISC-V 64/32 +SLUB+ no-symbol. - Supported regardless of whether
CONFIG_SLAB_FREELIST_HARDENEDisyorn. - Supported regardless of whether
CONFIG_SLAB_VIRTUALisyorn(x64 only). - It supports dumping partial pages (
-v) and NUMA node pages (-vv). - Since
page_to_virtis difficult to implement, it will heuristically determine the virtual address from the free-list. 
- It supports
sheaf/barnmechanism for linux 6.18~. 
- Supported on x64/x86/ARM64/ARM +
slab-dump: dumps slab free-list.slob-dump: dumps slob free-list.slub-tiny-dump: dumps slub-tiny free-list.slab-contains: resolves the slab cache (kmem_cache) that a certain address (object) belongs to (forSLUB/SLUB-TINY/SLAB).kmem-cache-alias: dumpskmem_cachealias name.buddy-dump: dumps the zone of the page allocator (buddy allocator) free-list.buddy-contains: resolves which buddy allocator block contains it.vmalloc-dump: dumpsvmallocused-list and freed-list.kobj: identifies an arbitrary kernel address.
ktrace, thunk-tracer, usermodehelper-tracer, kmalloc-tracer, kmalloc-allocated-by, kuaf-watch, kpage-watch
ktrace: traces arguments and returns of the specified kernel functions.thunk-tracer: collects and displays the thunk function addresses that are called automatically (x64/x86 only).usermodehelper-tracer: collects and displays the information that is executed bycall_usermodehelper_setup.kmalloc-tracer: collects and displays information whenkmalloc/kfree.kmalloc-allocated-by: calls a predefined set of system calls and prints structures allocated bykmallocor freed bykfree.kuaf-watch: tracks the alloc/free/reuse lifecycle of a slab object (or an entire cache) for UAF analysis.kpage-watch: tracks the alloc/free/slab lifecycle of a physical page (PFN) for cross-cache analysis.
xsm, wsm, bsm, optee-{break-ta,smc-service-dump,ta-dump,shm-list,bget-dump}
xsm: dumps secure memory when gdb is in normal world.wsm: writes the value to secure memory when gdb is in normal world.bsm: sets the breakpoint to secure memory when gdb is in normal world.optee-break-ta: sets the breakpoint to the offset of OPTEE-Trusted-App when gdb is in normal world.optee-smc-service-dump: dumps OPTEE SMC services.optee-ta-dump: dumps the information of OPTEE-Trusted-Apps from the memory or specified host directory.optee-shm-list: shows the information of dynamic shared-memory buffers.optee-bget-dump: dumps bget allocator of OPTEE-Trusted-App.
smm-status, smm-dump
ksearch-code-ptr, kdiff, uefi-ovmf-info, qemu-device-info
ksearch-code-ptr: searches for the code pointer in kernel data area.kdiff: compares kernel information (function pointers, etc.) at two points in time.uefi-ovmf-info: dumps addresses of some important structures in each boot phase of UEFI when OVMF is used.qemu-device-info: dumps device information for qemu-escape.
heap {arena,arenas,bins,bins-simple,calc-protected-fd,chunk,chunks,dump-image,extract-heap-addr,find-fake-fast,parse,snapshot,snapshot-compare,tcache-index-helper,timeline,top,tracer,try-calloc,try-free,try-malloc,try-realloc,visual-heap}
- Glibc heap commands have been improved. Supports up to glibc 2.43.
- It changes the color and prints a symbol if it exists.
- They print bins information if the chunk is in free-list.
- Thread arena is supported for all
heapcommands.- Use
-aoption.
- Use
- It supports some new sub-commands.
heap arenasheap bins-simple: displays the heap like pwndbg style.heap parse: displays the heap in the style of Pwngdb/angelheap.heap calc-protected-fd: calculates a valid value as protected fd.heap dump-image: visualizes chunks on a heap as composition image.heap extract-heap-addr: analyzes tcache-protected-fd introduced from glibc-2.32.heap find-fake-fast: searches for a memory with a size-like value that can be linked to the fastbin free-list.heap snapshot: takes a snapshot of the heap.heap snapshot-compare: compares current heap with a previously saved heap-snapshot.heap tcache-index-helperheap timeline(x86/x64/ARM32/ARM64, glibc): records heap states athttp://0.0.0.0:61234/.heap topheap tracer: has been integrated fromheap-analysis-helperand rewritten.heap try-mallocheap try-freeheap try-reallocheap try-callocheap visual-heap: is a colorized heap viewer.
uclibc-ng-{heap-dump,visual-heap,top,arena,chunk,chunks,parse,bins-simple,dump-image,snapshot,snapshot-compare}
uclibc-ng-heap-dump: dumps uClibc-ng heap chunks.- Based on uClibc-ng malloc-standard, without an architecture restriction.
- Tested on x64, x86, ARM, ARM64, MIPS, PPC, RISC-V, Alpha, SPARC64 and OR1K.
- Without the
__malloc_statesymbol, architectures other than x64/x86 need--malloc_state.
- Use
-sfor builds with Safe-Linking (e.g. uClibc-ng 1.0.54 and 1.0.58).- The shift follows
PAGE_SHIFTof the build (e.g. 13 on Alpha, 16 on ARM64 with 64KB pages).
- The shift follows

- How to test (x64):
- Download and extract
x86-64--uclibc--stable-2025.08-1.tar.bz2from https://toolchains.bootlin.com/ - Add
/PATH/TO/x86_64-buildroot-linux-uclibc/binto$PATH, then build asx86_64-linux-gcc test.c. - Fix interpreter by
patchelf --set-interpreter /PATH/TO/x86_64-buildroot-linux-uclibc/sysroot/lib/ld64-uClibc.so.0 a.out.
- Download and extract
- Based on uClibc-ng malloc-standard, without an architecture restriction.
uclibc-ng-visual-heap: is a colorized heap viewer for uClibc-ng.uclibc-ng-top,uclibc-ng-arena,uclibc-ng-chunk,uclibc-ng-chunks,uclibc-ng-parse,uclibc-ng-bins-simple,uclibc-ng-dump-image,uclibc-ng-snapshot,uclibc-ng-snapshot-compare: are ports of the same-nameheapsub-commands.uclibc-ng-arenadisplaysmalloc_state, which also holds the parameters that glibc keeps inmp_.-smeans Safe-Linking decoding as inuclibc-ng-heap-dump, so--skip-sizeofuclibc-ng-bins-simpleand-pofuclibc-ng-dump-imagereplace the glibc-s.- Snapshots have the same format as
heap snapshot, soheap snapshot-comparecan also compare them.
partition-alloc-dump, v8, cage, v8-list-maps, v8-dump-space
partition-alloc-dump: dumps Partition-Alloc free-list for chromium.
- This command is reserved for the implementation of the latest version of Chromium.
- Currently tested: v160.x (x64) / 1742397 / 473f38565a3a1d796d7bf99ff39152800c81213f
- Supported on Linux x64, x86, ARM64 and ARM32.
- Prebuilt sources: official Chromium (Linux x64 / Android ARM32, ARM64) and Debian (all four architectures). Android heap parsing is unverified.
- It will try heuristic search if the binary has no symbol.
- How to test:
v8: displays v8 (Chromium andd8) tagged object.cage: displays v8 (Chromium andd8) ubercage.v8-list-maps: lists v8 (Chromium andd8) built-in maps.v8-dump-space: dumps v8 (Chromium andd8) heap objects in each space.
{tcmalloc,musl,go,tlsf,hoard,mimalloc,scalloc,snmalloc,ssmalloc,jemalloc,sfmalloc,rpmalloc,tbbmalloc,zig,mesh,streamflow,supermalloc,dlmalloc,scudo,mozjemalloc,bmalloc,libpas,zend,obmalloc,newlib}-heap-dump
tcmalloc-heap-dump: dumps TCMalloc (gperftools, formerly google-perftools, v0.1~) free-list.musl-heap-dump: dumps Musl-Libc v0.5.0~ heap chunks (mallocng of v1.2.1~, and the legacy malloc).
- How to test:
- Get and extract the upstream source from https://musl.libc.org/
- Build with
./configure && make install.- For the legacy malloc, use v1.2.0 or earlier, or build with
./configure --with-malloc=oldmalloc && make install.
- For the legacy malloc, use v1.2.0 or earlier, or build with
- Then build with
/usr/local/musl/bin/musl-gcc test.c.
go-heap-dump: dumps Go Language v1.7~ mheap.tlsf-heap-dump: dumps TLSF (Two-Level Segregated Fit) v2.2.0 (2006/06/30) ~ v2.4.6 (2009/09/10) free-list (x64 only).
- How to test (x64):
- Get and extract the upstream source from http://www.gii.upv.es/tlsf/
- Build with
cd TLSF-2.4.6/src && make && cd ../examples && makethen usetest1etc.
hoard-heap-dump: dumps Hoard v3.10 (2013/10/18) ~ v3.2 (2025/12/31) free-list (x64 only).
- How to test (x64):
- Get and extract the upstream source from https://github.com/emeryberger/Hoard
- Build with
cd Hoard/src && make. - Run your test program with
LD_PRELOAD=/PATH/TO/libhoard.so ./a.out.
mimalloc-heap-dump: dumps mimalloc free-list (x64 only).
- How to test (x64):
- Get and extract the upstream source from https://github.com/microsoft/mimalloc
- Build with
mkdir build && cd build && cmake -DCMAKE_BUILD_TYPE=Debug .. && make. - Run your test program with
LD_PRELOAD=/PATH/TO/libmimalloc.so ./a.out.
scalloc-heap-dump: dumps scalloc free-list (x64 only).
- How to test (x64):
- Get and extract the upstream source from https://github.com/cksystemsgroup/scalloc
- Fix the bug with
sed -i -e 's/\(strncat(.*\), 1);/\1, 2);/' src/log.h. - Build with
gyp --depth . scalloc.gyp && make. - Enable overcommit with
echo 1 > /proc/sys/vm/overcommit_memory. - Run your test program with
LD_PRELOAD=/PATH/TO/libscalloc.so ./a.out.
snmalloc-heap-dump: dumps snmalloc free-list.- Use
-rto display remote message links and the retained tail. 
- How to test (x64):
- Get and extract the upstream source from https://github.com/microsoft/snmalloc
- Build with
mkdir build && cd build && cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Debug && ninja. - Run your test program with
LD_PRELOAD=/PATH/TO/libsnmallocshim.so ./a.out.
- Use
ssmalloc-heap-dump: dumps ssmalloc free-list (x64 only).
- How to test (x64):
- Get and extract the upstream source from https://github.com/Naruil/SSMalloc
- Patch
new_delete.cppbecause the original source uses dynamic exception specifications, which are rejected by C++17 and later:sed -i -e 's/ throw *(std::bad_alloc)//g' -e 's/^void operator delete(void \*ptr)$/void operator delete(void *ptr) noexcept/' -e 's/^void operator delete\[\](void \*ptr)$/void operator delete[](void *ptr) noexcept/' new_delete.cpp
- Build with
make - Run your test program with
LD_PRELOAD=/PATH/TO/libssmalloc.so ./a.out.
jemalloc-heap-dump: dumps jemalloc v3.0~ arenas, slabs and tcache (x64 only).
- How to test (x64):
- Install
libjemalloc2(e.g.apt install libjemalloc2), or get and extract the upstream source from https://github.com/jemalloc/jemalloc/releases - Build with
./configure && make build_lib_shared. - Run your test program with
LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so.2 ./a.out(or/PATH/TO/lib/libjemalloc.so.2).
- Install
sfmalloc-heap-dump: dumps sfmalloc free-list (x64 only).
- How to test (x64):
- Get and extract the upstream source from https://github.com/jeffhammond/sfmalloc
- Build with
cd sfmalloc && make CXX='g++ -std=gnu++03' DEFS='-D_REENTRANT -include stddef.h'. - Run your test program with
LD_PRELOAD=/PATH/TO/libsfmalloc.so ./a.out.
rpmalloc-heap-dump: dumps rpmalloc free-list.
- How to test (x64):
- Get and extract the upstream source from https://github.com/mjansson/rpmalloc
- Build with
cd rpmalloc && git switch main && python3 configure.py && ninja. - Run your test program with
LD_PRELOAD=/PATH/TO/bin/linux/release/librpmalloc.so ./a.out.
tbbmalloc-heap-dump: dumps tbbmalloc free-list.
- How to test (x64):
- Get and extract the upstream source from https://github.com/uxlfoundation/oneTBB
- Build with
mkdir build && cd build && cmake .. && make. - Run your test program with
LD_PRELOAD=/PATH/TO/libtbbmalloc_proxy.so ./a.out.
zig-heap-dump: dumps Zig Language heap.mesh-heap-dump: dumps Mesh free-list (x64/ARM64 only).
- How to test (x64):
- Get and extract the upstream source from https://github.com/plasma-umass/Mesh
- Build with
make build BAZEL_CONFIG='--config=modern-amd64 --linkopt=-fuse-ld=bfd'.- The default gold linker is no longer provided by recent binutils, so the bfd linker is used.
- To include debug symbols:
make build BAZEL_CONFIG='--config=modern-amd64 --linkopt=-fuse-ld=bfd --copt=-g'.
- Run your test program with
LD_PRELOAD=/PATH/TO/bazel-bin/src/libmesh.so ./a.out.
streamflow-heap-dump: dumps StreamFlow free-list (x64/x86/PPC64 only).
- How to test (x64):
- Get and extract the upstream source from https://github.com/scotts/streamflow
- Create an empty
numa.hwithcd streamflow && touch numa.h, because only the NUMA build uses libnuma. - Build with
make libstreamflow.so CXX='g++ -std=gnu++98 -include stddef.h' GCC_OPT='-O3 -ggdb -fpermissive -I. -fno-builtin'.- The original source is rejected by recent compilers, so the old C++ standard and
-fpermissiveare used. - Without
-fno-builtin, GCC may turnmalloc()+memset()incalloc()into a recursivecalloc()call.
- The original source is rejected by recent compilers, so the old C++ standard and
- Run your test program with
LD_PRELOAD=/PATH/TO/libstreamflow.so ./a.out.- Common commands such as
/bin/lsdo not work. Whenrealloc()of StreamFlow shrinks an object, it copies the old size into the new object, so they abort withbuffer overflow detected(or corrupt the heap if_FORTIFY_SOURCEis disabled).
- Common commands such as
supermalloc-heap-dump: dumps SuperMalloc free-list (x64 only).
- How to test (x64):
- Get and extract the upstream source from https://github.com/kuszmaul/SuperMalloc
- Build with
cd SuperMalloc/release && make. - Run your test program with
LD_PRELOAD=/PATH/TO/release/lib/libsupermalloc.so ./a.out.
dlmalloc-heap-dump: dumps dlmalloc (Doug Lea's malloc) v2.8.x free-list.
- How to test (x64):
- Get the upstream source from https://gee.cs.oswego.edu/pub/misc/malloc.c
- Build with
gcc -O2 -g -fno-builtin -shared -fPIC malloc.c -o libdlmalloc.so.- Without
-fno-builtin(and_FORTIFY_SOURCE), GCC may turnmalloc()+memset()incalloc()into a recursivecalloc()call.
- Without
- Run your test program with
LD_PRELOAD=/PATH/TO/libdlmalloc.so ./a.out.
scudo-heap-dump: dumps Scudo (the standalone allocator of LLVM 11~) free-list.
- How to test (x64):
- Get and extract the upstream source from https://github.com/llvm/llvm-project/releases/download/llvmorg-21.1.8/compiler-rt-21.1.8.src.tar.xz
- Build with
cd compiler-rt-21.1.8.src/lib/scudo/standalone && clang++ -std=c++17 -O2 -fPIC -shared-ftls-model=initial-exec -nostdinc++ -nostdlib++ -fno-exceptions -fno-rtti -I . -I include *.cpp -o libscudo.so. - Or use
libclang_rt.scudo_standalone-x86_64.soof the distribution (e.g.libclang-rt-21-devof Ubuntu). - Run your test program with
LD_PRELOAD=/PATH/TO/libscudo.so ./a.out.
mozjemalloc-heap-dump: dumps mozjemalloc (the allocator of Firefox 58~) arenas, runs, large and huge allocations.
- How to test (x64):
- Get and extract the official JS shell, e.g. https://archive.mozilla.org/pub/firefox/releases/159.0/jsshell/jsshell-linux-x86_64.zip
- Save the following as
test.js. It allocates 100 buffers of 0x300 bytes and frees half of them withgc().a=[];for(i=0;i<100;i++)a.push(new ArrayBuffer(0x300));for(i=0;i<100;i+=2)a[i]=null;gc();print(1) - Run
LD_LIBRARY_PATH=. gdb --args ./js -f test.js,catch syscall writeand thenrun.
bmalloc-heap-dump: dumps the old WebKit bmalloc (before it was replaced by mimalloc in 2026/01) free lists and per-threadCacheobjects.
- How to test:
- bmalloc is part of WebKit, so check out the last commit that still has the old allocator and build it alone:
git clone --filter=blob:none --sparse https://github.com/WebKit/WebKit.git && cd WebKit,git checkout 0a23246677 && git sparse-checkout set Source/bmalloc && cd Source/bmalloc. - Compile its core as a freestanding library with libpas disabled (GCC needs
-O2, not-O0/-O1):for f in $(ls bmalloc/*.cpp | grep -vE 'TZone|mbmalloc|Zone'); do clang++ -std=c++20 -O2 -g -mcx16 -Ibmalloc -I. -DBENABLE_LIBPAS=0 -DBUSE_LIBPAS=0 -c $f -o $f.o; done. - Build the test program (
g_configis normally provided by WTF, so-DBMALLOC_STANDALONEdefines it here):clang++ -std=c++20 -O0 -g -mcx16 -Ibmalloc -I. -DBENABLE_LIBPAS=0 -DBUSE_LIBPAS=0 -DBMALLOC_STANDALONE /PATH/TO/tests/test.cpp bmalloc/*.o -lpthread -ldl -latomic. - Run
gdb ./a.out.
- bmalloc is part of WebKit, so check out the last commit that still has the old allocator and build it alone:
libpas-heap-dump: dumps libpas (the WebKit allocator that replaced the old bmalloc), the per-thread caches, the small size-class free objects and the large free ranges.
- How to test:
- libpas is part of WebKit; a release branch works too (
webkitglib/2.50..webkitglib/2.54, ormain):git clone --filter=blob:none --sparse https://github.com/WebKit/WebKit.git && cd WebKit,git sparse-checkout set Source/bmalloc. - Build libpas with the bmalloc glue as a static library (clang; GCC also works with
-latomic):cd Source/bmalloc, then for everylibpas/src/libpas/*.candbmalloc/*.cppnamed inCMakeLists.txtexceptZone.cpp(Darwin only):clang++ -std=c++20 -O2 -g -Ibmalloc -I. -Ilibpas/src/libpas -D_GNU_SOURCE -DPAS_BMALLOC=1 -DNDEBUG -DBUILDING_bmalloc -c <file>(compile the.cfiles the same way), andar rcs libbmalloc.a *.o. - Build the test program against it:
clang++ -O0 -g /PATH/TO/tests/test.cpp libbmalloc.a -lpthread -latomic. - Run
gdb ./a.out.
- libpas is part of WebKit; a release branch works too (
zend-heap-dump: dumps Zend MM (the PHP allocator of PHP 7.0~) free lists, chunks and huge blocks.
- How to test:
- Install
php-cli(e.g.apt install php-cli), or get and extract the upstream source from https://www.php.net/downloads and build with./configure --disable-all --enable-cli && make(the binary issapi/cli/php). - Save the following as
test.php. It allocates 100 strings of 0x100 bytes and frees half of them.<?php $a=[];for($i=0;$i<100;$i++)$a[]=str_repeat("A",0x100);for($i=0;$i<100;$i+=2)unset($a[$i]);echo 1; - Run
gdb --args php -n test.php,catch syscall writethenrun.
- Install
obmalloc-heap-dump: dumps obmalloc (pymalloc, the small object allocator of CPython) arenas, pools and free blocks.
- How to test:
- Install
python3(e.g.apt install python3), or get and extract the upstream source from https://www.python.org/downloads/source/ and build with./configure && make python. - Save the following as
test.py. It allocates 1000 bytes objects of 0x40 bytes and frees half of them.import os,signal;a=[b"A"*n for n in [0x40]*1000];del a[::2];os.kill(os.getpid(),signal.SIGTRAP) - Run
gdb --args python3 test.pythenrun.
- Install
newlib-heap-dump: dumps the free chunks of newlib (the C library for bare-metal and embedded systems), for both the default malloc (based on dlmalloc 2.6.4) and the nano malloc.
- How to test:
- Get and extract the Arm GNU Toolchain for
arm-none-eabifrom https://developer.arm.com/downloads/-/arm-gnu-toolchain-downloads. - Save the following 2 lines as
test.c. It allocates 100 chunks of 0x40 bytes and frees half of them.#include <stdlib.h>int main(){void*a[100];for(int i=0;i<100;i++)a[i]=malloc(64);for(int i=1;i<100;i+=2)free(a[i]);__builtin_trap();} - Build with
arm-none-eabi-gcc --specs=rdimon.specs -g test.c(add--specs=nano.specsfor the nano malloc). - Run
qemu-arm -g 1234 ./a.outandgdb-multiarch ./a.out -ex 'target remote :1234'thenc.
- Get and extract the Arm GNU Toolchain for
For the supported architectures, see docs/QEMU-USER-SUPPORTED-ARCH.md.
si, ni, c
si/ni: are wrappers for nativesi/niif OpenRISC 1000 or CRIS.- On OpenRISC 1000 architecture, branch operations don't work well, so GEF uses breakpoints to simulate.
- On CRIS architecture,
stepi/nexticommands don't work well, so GEF uses breakpoints to simulate. - If you want to use native
si/ni, use the full formstepi/nexti.
c: is the wrapper for nativecif gdb is connected toqemu-userorIntel Pin.- When connecting to gdb stub of
qemu-userorIntel Pin, gdb does not trapSIGINTduringcontinue. - If you want to trap, you need to issue
SIGTRAPon theqemu-userorpinside, but switching screens is annoying. - This command realizes a pseudo
SIGTRAPtrap by trappingSIGINTon the Python side and throwingSIGTRAPback toqemu-userorIntel Pin. - It works only for local
qemu-userorIntel Pin. - If you want to use native
c, use the full formcontinue.
- When connecting to gdb stub of
context, registers, telescope
context(improved)- It supports automatic display of system call arguments when calling a system call.
- It supports new commands:
context on,context off. - It supports automatic display of address and value when accessing memory.
- It supports smart symbol printing for C++ function.
registers(improved)telescope(improved)- It displays ordinal numbers as well as offsets.
- It displays if there are canaries and return addresses in the target area.
- It supports blacklist address features (to avoid dying when touching the address mapped to the serial device).
- It also shows the symbol if available.
- It supports some new options:
--is-addr,--is-not-addr,--perm,--is-zero,--is-not-zero,--mask-hits,--no-mask-hits,--tag,--frame-split,--uniq,--interval,--depth,--phys,--list-head,--slab-contains,--slab-contains-unaligned.
break-rva, command-break, main-break, load-break, regdump-break, multi-break, break-if-taken, break-if-not-taken
break-rva: sets a breakpoint at relative offset from codebase.command-break: sets a breakpoint which executes user defined command if hit.main-break: sets a breakpoint atmainwith or without symbols, then continue.- This is useful when you just want to run to
mainusingqemu-userorpin, or debugging no-symbol ELF.
- This is useful when you just want to run to
load-break: breaks if something is loaded.regdump-break: sets a breakpoint which dumps specified registers if hit.multi-break: sets multiple breakpoints easily.break-if-taken/break-if-not-taken: sets a breakpoint which breaks if branch is taken (or not taken).
up, down, time, multi-line, stepi-for-kgdb
exec-until, call-trace, xuntil, strings-continue, xskip
exec-until: executes until specified operation.call-trace: traces call, ret, and syscall instructions.xuntil: executes until specified address.- It is slightly easier to use than the original until command.
strings-continue: single-steps and prints ASCII strings referenced by general-purpose registers.xskip: skips instructions easily.
ii, dasm, asm-list
ii: is a shortcut forx/50i $pcwith opcode bytes.dasm: disassembles the code by capstone.asm-list: lists instructions. (x64/x86 only)
- This command uses x86data.js from https://github.com/asmjit/asmdb
extra, comment, smart-cpp-function-name
extra: manages user specified command to execute when each step.comment: manages user specified temporary comment.smart-cpp-function-name: togglescontext.smart_cpp_function_namesetting.
mprotect, hijack-fd, call-syscall, mmap, munmap, kill-threads, syscall-search, xtap
mprotect(improved)- Rewritten to use
call-syscall.
- Rewritten to use
hijack-fd(improved)call-syscall: calls system call with specified values.mmap: allocates a new memory bycall-syscall.munmap: unmaps a memory bycall-syscall.kill-threads: kills specific or all threads (forpthread).syscall-search: searches for system call by regex.xtap: taps read/write syscalls on specific file descriptors and hexdumps the transferred data.
unicorn-emulate, unicorn-emulate-script, angr, future-calls
unicorn-emulate(improved)unicorn-emulate-script: generates and runs a standalone script that emulates the binary by Unicorn-Engine.angr: finds simple constraints byangr.future-calls: displays future function calls from the current function.
ropper, format-string-helper, add-symbol-temporary, seccomp, onegadget, rp, follow
ropper(improved)- It does not reset autocomplete settings after calling imported
ropper.
- It does not reset autocomplete settings after calling imported
format-string-helper(improved)- It supports more
printf-like functions.
- It supports more
add-symbol-temporary: adds symbol information from command-line.seccomp: invokesceccomporseccomp-tools.onegadget: invokesone_gadget.rp: invokesrp++with commonly used options.follow: changesfollow-fork-modesetting.
proc-info, elf-info, fds, proc-dump
proc-info(improved)elf-info(improved)fds: shows opened file descriptors.proc-dump: dumps each file under/proc/PID/.
tls, fsbase, gsbase, libc, ld, heapbase, codebase
vmmap, xinfo
vmmap(improved)- It displays the memory map information even when connecting to gdb stub of
qemu-user. - It also supports
Intel Pin. - It also supports
Intel SDE. - It is redirected to
kvmmapwhen connecting to gdb stub ofqemu-system. - It supports detection and coloring of
Writable,ReadOnly,NoneandRWXregions. - It shows the area each register points to.
- It displays the memory map information even when connecting to gdb stub of
xinfo(improved)
pid, tid, filename, auxv, argv, envp, dumpargs, vdso, vvar, errno
pid/tid: prints pid and tid.filename: prints filename.auxv: pretty prints ELF auxiliary vector.argv/envp: pretty prints argv and envp.dumpargs: dumps arguments of current function.vdso: disassembles the text area of vdso smartly.vvar: dumps the area of vvar.errno: displays errno list or specified errno.
got, got-all, dwarf-exception-handler, dynamic, link-map, dtor-dump, iouring-dump, stdio-dump, fpchain, unwind-info, cpp-vtable
got(improved)got-all: shows got entries for all libraries.dwarf-exception-handler: dumps the DWARF exception handler information.dynamic: dumps the_DYNAMICarea.link-map: dumps useful members oflink_mapwith iterating.dtor-dump: dumps some destructor functions list.iouring-dump: dumps the area of iouring (x64 only).stdio-dump: dumps members of stdin/stdout/stderr.fpchain: dumps chains from__IO_list_all.unwind-info: shows the effective DWARF unwind rules at the current or specified PC.cpp-vtable: dumps C++ virtual tables and RTTI from an object or base subobject.
checksec, canary, capability, ptr-mangle, ptr-demangle, search-mangled-ptr, mte-tags, sighands
checksec(improved)canary(improved)capability: shows the capabilities of the debugging process.ptr-mangle: shows the mangled value that will be mangled byPTR_MANGLE.ptr-demangle: shows the demangled value of the value mangled byPTR_MANGLE.search-mangled-ptr: searches for the mangled value from RW memory.mte-tags: displays the MTE tags for the specified address.sighands: displays the signal dispositions of the current process.
magic, symbols
types, dt
types: lists all types with compaction.dt: makes it easier to useptype /ox TYPEandp ((TYPE*) ADDRESS)[0].
- This command is designed for several purposes.
- When displaying very large struct, you may want to go through a pager because the results will not fit on one screen. However, using a pager, the color information disappears. This command calls the pager with preserving colors.
- When
ptype /ox TYPE, interpreting member type recursively often result is too long and difficult to read. This command keeps result compact by displaying only top-level members. - When
p ((TYPE*) ADDRESS)[0]for large struct, the setting ofmax-value-sizeis too small to display. This command adjusts it automatically. - When debugging a binary written in the Golang, the offset information of the type is not displayed. This command also displays the offset.
- When debugging a binary written in the Golang, the
p ((TYPE*) ADDRESS)[0]command will be broken. This is because the Golang helper script is automatically loaded and overwrites the behavior ofpcommand. This command creates the display results on the Python side, so we can display it without any problems.
search-pattern, strings, search-cfi-gadgets, find-syscall
search-pattern(improved)strings: searches for ASCII string from specific location.search-cfi-gadgets: searches for CFI-valid (for CET IBT) and controllable generally gadgets in the executable area.find-syscall: searches the syscall gadget.
hexdump, hexdump-flexible, json, xs, xc, sigreturn, ucontext, jmpbuf
hexdump(improved)hexdump-flexible: displays the hexdump with user defined format.json: pretty prints json.xs: dumps string likex/scommand, but with hex-string style.xc: dumps address likex/xcommand, but with coloring at some intervals.sigreturn: displays stack values for sigreturn syscall.ucontext: displays the register state saved in aucontext_tormcontext_t.jmpbuf: displays the registers saved in ajmp_buforsigjmp_bufof glibc, withPTR_MANGLEdecoded.
patch, memset, memcpy, memswap, meminsert
patch(improved)memset: sets the value to the memory range, whether virtual or physical.memcpy: copies the contents from the address A to B, whether virtual or physical.memswap: swaps the contents of the address A and B, whether virtual or physical.meminsert: inserts the contents of the address A to B, whether virtual or physical.
hash, crc, base-n-decode, base-n-encode, morse-decode, morse-encode, strlen, is-mem-zero, seq-length
hash: calculates various (450+) hashes, or show known-collisions.crc: calculates various CRCs.base-n-decode/base-n-encode: decodes/encodes various baseN.morse-decode/morse-encode: decodes/encodes morse code.strlen: detects the length of the string.is-mem-zero: checks the contents of address range are all 0x00 or 0xff.seq-length: detects consecutive length of the same sequence.
smart-memory-dump, load-file, load-file-mmap
binwalk-memory, filetype-memory, lsix, peek-pageframe, peek-pageflags, vdump, freq-analysis, stack-recover
binwalk-memory: scans memory bybinwalk.filetype-memory: scans memory byfileandmagika.lsix: shows images in the terminal byimagemagick.peek-pageframe: reads page frame data.peek-pageflags: reads page flags of a page frame.vdump: visualizes memory data like an image.freq-analysis: visualizes the frequency of occurrence of each byte.stack-recover: heuristically recovers the call chain from the stack whenbtis unusable.
sysreg, pac-keys, fpu, mmx, sse, avx, avx512, cpuid
edit-flags, xmmset, mmxset
u2d, unsigned, convert, addressify, cpp-mangle, cpp-demangle
ret2dl-hint, srop-hint
distance, crc32rev
qemu-system-memory-region-dump
theme, reset-cache, tmux-setup, gef version, gef status, gef reset-breakpoint, gef arch-list, gef pyobj-list, gef avail-comm-list, gef set-arch, gef check-update, gef dump-commands, history
theme(improved)reset-cache(improved)- It has been integrated into
gefas sub-command. - The cache structure within GEF has changed significantly. This command corresponds to them.
- It has been integrated into
tmux-setup(improved)- It has been integrated into
gefas sub-command. screenis no longer supported.tmuxsettings are predefined and cannot be customized.- If you want to customize it, edit dev/tmux/tmux_setup.py and run
source /path/to/tmux_setup.py. - If you're a
zellijuser, you can use dev/zellij/zellij-wrapper.py. Run this script before startinggdborzellij.
- It has been integrated into
gef version: shows software versions that GEF uses.gef status: shows architecture information used in GEF.gef reset-breakpoint: shows and resets all breakpoints.gef arch-list: displays defined architecture information.gef pyobj-list: displays defined global Python objects.gef avail-comm-list: displays a list of commands which are available or not for the current architecture and gdb execution mode.gef set-arch: sets a specific architecture to GEF.gef check-update: checks for GEF updates.gef dump-commands: dumps GEF command documentation as Markdown.history: shows gdb command history easily.
gef help, etc.
- The category is introduced in
gef help. - Combined into one file (from
gef-extras). The following are moved fromgef-extras.current-stack-frame,xref-telescope,bytearray, andbincompare.- This is because a single file is more attractive for me than ease of maintenance.
- The system-call table used by
syscall-argsis moved fromgef-extras.- It was updated up to Linux kernel 7.x for each architecture.
- Removed some features that I don't use:
$,ida-interact,gef-remote,pie,pcustom,ksymaddr,trace-run,shellcode. - Many bug fixes / formatting improvements / usability enhancements (made it easy for me to use).
- See docs/FAQ.md.
- Why I decided to make this
- The story behind each command, etc.
- The story behind each command, etc. 2024 Edition
- The story behind each command, etc. 2025 Edition
- Orynth project page
- https://www.orynth.dev/projects/bata24-gef-8901
- Fan token symbol:
B24G(details in docs/FAQ.md)




















































































































































































































































