KSCrash enableSwapCxaThrow crashes app when a C++ exception is handled

[generalpf]

Stack trace

N/A

Reproduction steps

It's hard to explain, we use a C++ library called Linphone, and sometimes it throws C++ exceptions which we handle. When the exception is caught by KSCrash, and enableSwapCxaThrow is enabled, the crash handler itself crashes.

Volume

100

Affected SDK versions

3.5.0

Latest working SDK version

3.1.0

Does the crash manifest in the latest SDK version?

Yes

Deployment Target

iOS 15+

Device Information

Any iOS device.

Other relevant information

I know this is the reason because we used to use Embrace's SDK, and when they updated to the version of KSCrash with the experimental enableSwapCxaThrow feature, it started causing problems, and they actually turned it off for everyone.

Please turn this off, or allow the person calling CrashReporting.enable() to supply an argument.

[simaoseica-dd]

Hi @generalpf 👋

Thanks for the report.

Could you please share a crash log so we can get more details about the point of failure? Even if KSCrash crashes, iOS should record the crash and it may be available on the device or App Store Connect.

• On the device: Settings -> Privacy & Security -> Analytics & Improvements -> Analytics Data, then look for a file named like AppName-YYYY-MM-DD-xxxx.ips.
• TestFlight or App Store builds: the crash should appear automatically in the App Store Connect.

In the meantime, we are considering defaulting this experimental option to false in upcoming versions to avoid instability.

[generalpf]

Hi @simaoseica-dd there is no .ips file for my app. The exception handling chain is being interrupted.

I tracked it down to this line: https://github.com/kstenerud/KSCrash/blob/72e742c81d4ba03fab137e2651a1de342cdd8b3a/Sources/KSCrashRecordingCore/KSCxaThrowSwapper.c#L127

As soon as we call original() the process just flat-out terminates, nothing in the logs or anywhere else.

There has been an update to KSCrash in the past couple of days and a lot of activity around this file.

[simaoseica-dd]

Thanks for pointing out the exact line.

Could you please give a try with the latest KSCrash developments? You can point directly to my simaoseica/kscrash branch.

This feature has seen some recent improvements and is now automatically disabled when sanitizers (ASan, TSan, etc.) are detected, as they also intercept __cxa_throw and can conflict with this mechanism.

[generalpf]

I still have the same problem on your branch. The app just dies without an exception when original() is called:

if (function != 0) {
    KSLOG_TRACE("Calling original __cxa_throw function at %p", (void *)function);
    cxa_throw_type original = (cxa_throw_type)function;
    original(thrown_exception, tinfo, dest);
}

If I make this change (diff is for your branch, but the same change can be made to main), everything is fine:

diff --git a/DatadogCrashReporting/Sources/KSCrashIntegration/KSCrashPlugin.swift b/DatadogCrashReporting/Sources/KSCrashIntegration/KSCrashPlugin.swift
index 1a7dfdf08..a6a14bbda 100644
--- a/DatadogCrashReporting/Sources/KSCrashIntegration/KSCrashPlugin.swift
+++ b/DatadogCrashReporting/Sources/KSCrashIntegration/KSCrashPlugin.swift
@@ -111,6 +111,7 @@ extension KSCrashConfiguration {
         config.monitors = [.signal, .cppException, .nsException, .system]
         config.reportStoreConfiguration.maxReportCount = 1
         config.reportStoreConfiguration.reportCleanupPolicy = .never
+        config.enableSwapCxaThrow = false
         return config
     }
 }

I should point out that there's no actual crash happening at this time. Something that the swap CXA throw code is doing is making code of mine crash that doesn't normally crash. (My C++ exceptions are already being handled.)

[simaoseica-dd]

Thanks for all the detailed investigation @generalpf.

We're proceeding with a hotfix early next week that disables swapping of the __cxa_throw function.

We'll follow up once the fix is released.

[simaoseica-dd]

Hi @generalpf 👋

The fix is now released in version 3.6.1.

Thanks again for pointing out the issue and for all the shared details.

[generalpf]

Thank you @simaoseica-dd! The fix works.