Is there a way to replace emscripten stub functions?

[iliakan]

I'm trying to port a program that calls fork and execve on some occasions.

I'd like to show my error and stop the execution in cases of such calls.

Here's the stub for e.g. fork:

https://github.com/emscripten-core/emscripten/blob/main/system/lib/libc/emscripten_libc_stubs.c#L165

I'd like the build to use my fork instead.

That is, replace this symbol build-time OR after the build (or in any other way).
How ? ;)

[sbc100]

If you include your own definition of fork in your program (i.e. in one of your object files, or libraries) the linker will choose your definition over the stub one.

Alternatively if you set -sALLOW_UNIMPLEMENTED_SYSCALLS=0 the stub library will not be included at all, and you will be forced to implement any symbols your program depends on.

From settings.js:

// Include unimplemented JS syscalls to be included in the final output.  This   
// allows programs that depend on these syscalls at runtime to be compiled, even 
// though these syscalls will fail (or do nothing) at runtime.                      
var ALLOW_UNIMPLEMENTED_SYSCALLS = true;  

[iliakan]

Here's the error if I do that without any flags:

wasm-ld: error: duplicate symbol: fork
>>> defined in my.a(run-command.o)
>>> defined in /opt/projects/git/emsdk/upstream/emscripten/cache/sysroot/lib/wasm32-emscripten/libstubs-debug.a(emscripten_libc_stubs.o)

[sbc100]

Ah, this is because your program also uses some other function in the emscripten_libc_stub.s.c file. I think the simplest solution here would be to mark the stub function(s) as weak.

I think musl provides a weak macro which expands to __attribute__((__weak__)) to I think in this file you can just prefix the declaration of fork with the word weak: weak pid_t fork(void) {.

Whenever you make any changes to the source code for any of the system libraries you also need to make sure they get rebuilt. The simplest way to do that would be ./emcc --clear-cache which removes all libraries. But you can be more precise with ./embuilder build <name> --force where name is one of the library names, which you can find from ./embuilder --help.

[iliakan]

You mean I should change the file emscripten_libc_stub.s.c?

[sbc100]

The stubs are defined in system/lib/libc/emscripten_libc_stubs.c

[iliakan]

I'd just prefer not to patch emscripten itself for the purposes of a single project ;)

Or maybe we should update Emscripten to make all these functions weak?
To allow redefinition for other projects?

[iliakan]

Right now I'm building in two steps:

1. emmake make all ... CFLAGS="$LINK_AND_COMPILE_CFLAGS"
2. emcc ${=LINK_AND_COMPILE_CFLAGS} ${=COMPILE_CFLAGS} *.o -o my.js

Where should I put the -sALLOW_UNIMPLEMENTED_SYSCALLS=0, if I am to try it? ;)

Should it go to (1) or (2)? In CFLAGS or LDFLAGS?

P.S. I tried a couple of things, didn't work, so would like to ensure I do that right.

[sbc100]

ALLOW_UNIMPLEMENTED_SYSCALLS is a link time setting, so it would go in LDFLAGS. We even have a nice warning if you try to put it in CFLAGS:

$ ./emcc -sALLOW_UNIMPLEMENTED_SYSCALLS=0 test/hello_world.c -c
emcc: warning: linker setting ignored during compilation: 'ALLOW_UNIMPLEMENTED_SYSCALLS' [-Wunused-command-line-argument]

(But I think the marking of the stub functions as weak is the likely the way you want to go).

[iliakan]

Ok, I actually tried that. Got a bunch or errors for other symbols ;)

wasm-ld: error: daemon.o: undefined symbol: getpwnam
wasm-ld: error: daemon.o: undefined symbol: getgrnam
wasm-ld: error: libgit.a(ident.o): undefined symbol: getpwuid
wasm-ld: error: daemon.o: undefined symbol: getpwnam
wasm-ld: error: /opt/projects/git/emsdk/upstream/emscripten/cache/sysroot/lib/wasm32-emscripten/libc-debug.a(getpgid.o): undefined symbol: __syscall_getpgid
wasm-ld: error: /opt/projects/git/emsdk/upstream/emscripten/cache/sysroot/lib/wasm32-emscripten/libc-debug.a(getrlimit.o): undefined symbol: __syscall_prlimit64
wasm-ld: error: /opt/projects/git/emsdk/upstream/emscripten/cache/sysroot/lib/wasm32-emscripten/libc-debug.a(getrlimit.o): undefined symbol: __syscall_ugetrlimit
wasm-ld: error: /opt/projects/git/emsdk/upstream/emscripten/cache/sysroot/lib/wasm32-emscripten/libc-debug.a(getuid.o): undefined symbol: __syscall_getuid32
wasm-ld: error: /opt/projects/git/emsdk/upstream/emscripten/cache/sysroot/lib/wasm32-emscripten/libc-debug.a(link.o): undefined symbol: __syscall_link
wasm-ld: error: /opt/projects/git/emsdk/upstream/emscripten/cache/sysroot/lib/wasm32-emscripten/libc-debug.a(setsid.o): undefined symbol: __syscall_setsid
wasm-ld: error: /opt/projects/git/emsdk/upstream/emscripten/cache/sysroot/lib/wasm32-emscripten/libc-debug.a(uname.o): undefined symbol: __syscall_uname
wasm-ld: error: /opt/projects/git/emsdk/upstream/emscripten/cache/sysroot/lib/wasm32-emscripten/libc-debug.a(waitpid.o): undefined symbol: __syscall_wait4
wasm-ld: error: /opt/projects/git/emsdk/upstream/emscripten/cache/sysroot/lib/wasm32-emscripten/libsockets.a(setsockopt.o): undefined symbol: __syscall_setsockopt

So I am to implement all that stuff, right?

[sbc100]

I think the best approach, if you just want to impement/modify, fork is to mark is as weak and not to take the ALLOW_UNIMPLEMENTED_SYSCALLS appoach.

As you can see, ALLOW_UNIMPLEMENTED_SYSCALLS opts out of the emscrpiten stubs, which would mean you would need to fill all of them in again somehow, or modify your program not to depend on them.

[iliakan]

Why not label all functions as weak in stubs, in upstream Emscripten?

Looks like a good way to go. Stubs that can easily be replaced without patching :yammy!:

[sbc100]

Yup, that makes sense me.

[sbc100]

BTW, if you are going to try to implement fork for in emscripten you might have hard time, depending on what your program is doing. If you program is just using fork immediately followed by exec you might want to look into some perhaps easier solutions such as vfork or (even better) posix_spawn. These don't require the complete cloning the current process.

[iliakan]

Right now the program dies in a strange manner when it tries to use fork and few other not implemented syscalls.

Some stuff I rewrote using WASM module instantiation to allow nested execution.

For other, less important functionality that relies on fork, I'd like to make fork show a nice error message, explaining that the functionality is limited, because it's in the browser environment. And then kill the process, to avoid other odd side-effects ;)

That's all I plan to put into a new, rewritten fork ;)

[iliakan]

The emcc --clear-cache worked, thanks (in correct emsdk env).

[iliakan]

So I added __attribute__((__weak__)) to pid_t fork(void) in emscripten_libc_stubs.c, and it compiled!

My fork is working now.

Does it seem to be a good PR if I add this __attribute__((__weak__)) to all other functions in this file?

[sbc100]

Yes please, that would great. Perhaps we can use the musl macro weak instead though?

[iliakan]

Do we have to?

Actually, I tried include <features.h> and then just weak, but it didn't fly :/

[iliakan]

#include <features.h>
...
weak pid_t fork(void) {

unknown type name 'weak'

[sbc100]

Fair enough, looks like our stub library doesn't inherit the musl internal header path. The spelled out attribute sounds fine to me, or you could add #define weak ...` at the top of the file.

[iliakan]

Suggested as #18736