fail to load on Android 7.1 and lower

[trinity-1686a]

i don't think sysinfo can really solve this problem, so i'm opening this more as a documentation for a problem and its workaround.

Since #1543, sysinfo links to setpwent, getpwent and endpwent. On Android, these functions only exists starting with SDK 26 (Android 8.0). Trying to load a rust dynamic library depending on sysinfo on Android 7.1 fails with a UnsatisfiedLinkError exception.

Calls and extern declarations of the problematic function can be entirely removed by not building with the feature user (which is a default feature, but sounds like its rarely needed on android)

[GuillaumeGomez]

Thanks for the information! I don't really know how I could put this into the documentation so that people developing for older Android version would be aware of it. If you have suggestions? Maybe in the README.md file?

[trinity-1686a]

honestly i don't know, the error one gets doesn't even say what crate requested the symbol, so someone getting the error we got is unlikely to come here before having done a fair bit of troubleshooting already 😕

[robertkirkman]

out of interest, could you provide a reproduction script for the issue?

I have Android 7.1.2, and normal use of this repository through Cargo didn't produce any errors for me.

However, I understand that you have explained that this may only happen when "trying to load a rust dynamic library", I don't really understand what that means, so I would love to see it and test the resulting example on my device.

[trinity-1686a]

the original ticket on our side is here https://gitlab.torproject.org/tpo/applications/vpn/-/issues/374 , there more information there, but a summary is:

and this is the app that had the issue: https://play.google.com/store/apps/details?id=org.torproject.vpn
it loads a dynamic library (.so) through the JNI, and that dynamic library contains sysinfo (among a lot of other things)

it originally happened on a Xiaomi Redmi 5 (that's a phone from 2017) with android 7.1 (sdk 25), the crash being observed in the wild, we have no way to contact whoever had that phone.
we had a lot of trouble reproducing the issue (it doesn't reproduce on a Samsung Galaxy J6 from 2018, sdk 25 too), but managed to do it with the x86 android emulator with sdk 24, i don't know if emulator with sdk 25 was tested

Bionic claims it was introduced in sdk 26: https://android.googlesource.com/platform/bionic/+/b28d496/libc/include/pwd.h#89 , but it's possible it was already present in some (but not all) cases before.

[robertkirkman]

Thank you for explaining!

I would like to let you know that I have downloaded your app, and I have been able to reproduce the launch crash in it on my Android 7.1.2 device.

I would also like to let you know that, after inspecting the code of your app, I have discovered that it is possible to force the app to stop crashing successfully on my Android 7.1.2 device by performing this modification to your app's code:

--- a/scripts/reproducible_build.sh
+++ b/scripts/reproducible_build.sh
@@ -63,6 +63,12 @@ export RUSTFLAGS=\"\$RUSTFLAGS --remap-path-prefix $CARGO_HOME=/cargo/ --remap-p
 
 fi
 
+# work around https://github.com/GuillaumeGomez/sysinfo/issues/1582
+cargo vendor
+echo '[patch.crates-io]' >> Cargo.toml
+echo 'sysinfo = { path = "./vendor/sysinfo" }' >> Cargo.toml
+sed -i '/"user",/d' vendor/sysinfo/Cargo.toml
+
 cargo clean
 # https://gitlab.torproject.org/tpo/core/arti/-/issues/1993
 # Arti #1993: reproducible builds are not

Yes, I have actually compiled your app using the provided reproducible_build.sh both before and after this modification, and I can confirm that this exact modification results in the resulting build of your app not crashing on launch for me anymore on my Android 7.1.2 device.

However, understandably, this kind of change, taken verbatim as I have written it, might be considered too inconvenient, unreliable, DRY-violating (in the case if you would need to implement it multiple times in other build scripts that are not reproducible_build.sh) or otherwise unoptimal for directly upstreaming into your repository. I am just showing this so that you understand that I reproduced the error and also successfully worked around it locally by writing this change.

My Android 7.1.2 device is Samsung Galaxy S III SPH-L710 with LineageOS 14.1.

[trinity-1686a]

we are in a position where we can disable the feature without patching, sysinfo is imported by tor-memquota, which is also developed by us. We're just waiting on the next release cycle to have the fix bubble toward the app, but thanks for taking the time to try it!

[GuillaumeGomez]

What could work would be to load these functions dynamically and have a nice fallback in case it doesn't work. However, that would either mean I stop using libc for these functions and that libc needs to completely change paradigm (linking at compile time vs at run time). If someone sees an alternative, I don't think there is anything I can do here. :-/

[trinity-1686a]

i mean, dlsym(RTLD_DEFAULT, ..) is definitely a possibility, but also definitely not worth the hassle.
I've had to do terrible things because of Bionic and wanting to support old Android, i don't think it's on maintainers of general-purpose crates to support a version of Android that was superseded 8 years ago
As i said:

i don't think sysinfo can really solve this problem, so i'm opening this more as a documentation for a problem and its workaround.

i think being able to compile out the module that causes issues is the most that can reasonably be asked for, and that already existed before this ticket

[GuillaumeGomez]

Well if there is a non-terrible solution that can make old android developers life easier, I don't mind doing it.