unfork(2) is the inverse of fork(2). sort of.
Running code in the process without running code in the process.
A: The inverse of fork(2).
A: Not really a question, but sure. fork(2) splits one process (really, address space) into two. unfork(2) joins two address spaces into one.
A: More of a comment again. Think of this technique as taking a cheap snapshot of another application that you can read, write, execute, discard, repeat, all while leaving no ptrace and sending no signal. Heh.
A: The magic of Linux! By combining userfaultfd with process_vm_readv, any userspace application can obtain a copy-on-write mapping (with some limitations) of memory it never owned. All it needs is ptrace privileges, which is to say, having the same uid usually works.
A: Dynamic binary analysis and instrumentation of applications with built-in integrity checks. As far as I know process_vm_readv isn’t even detectable if the agent process is more privileged than the examinee process—so you’re free to manipulate your private copy of the application in the comfort of your own address space.
A: It’s true that meshing address spaces is much harder than copying them. Especially on 32-bit systems, the likelihood of a collision is high, and that’s why everything is built using musl, which excels at producing compact, straightforward binaries that fit into the areas the kernel doesn’t place anything into; 64-bit systems with ASLR are far more forgiving. Nevertheless, I think that with some effort two allocators or even dynamic linkers could survive together.
In terms of functionality, you get virtual memory and TLS (TLS is everywhere!) but not, say, files or shared memory; at some point a kernel module would be necessary. On the platform support side, x86_64 and i386 seem to work well, and those two are also the most challenging ones.
A: You’re welcome!
A: tf2_healslut
Keep in mind that this repository is nothing more than a proof of concept.
You’ll need patched musl-gcc. (Using stock musl mostly works by accident on x86_64, but will usually crash on i386 without the patch.) After that, check the paths in the *.specs files, run make and it’s done.
The demo code uses puts as an entry point and prints two messages; the first one using a clean snapshot of another process, and the second one reusing the same snapshot. Any process that dynamically loads libc.so works, e.g. /bin/cat. Run ./unfork[32|64].elf $(pidof cat) and enjoy.
Although puts may seem trivial, it involves a surprisingly deep call stack (well… in glibc), syscalls, TLS, vDSOs, and is probably more complicated than the kind of problems I originally set out to solve.