Crash in SafeDeleteSslContext.WriteToConnection

[filipnavara]

We started getting large number of crashes in the field with a native crash in SafeDeleteSslContext.WriteToConnection.

Majority of the crashes happen on iOS with the following NativeAOT stack trace:

0   MailClient.Mobile.iOS           0x203eec3f8         _System_Net_Security_System_Net_SafeDeleteSslContext__WriteToConnection
1   Security                        0x316b6ea3c         <redacted>
2   Security                        0x316b6e600         SSLClose
3   Security                        0x316b6e3f8         SSLRead
4   MailClient.Mobile.iOS           0x202bd3ec0         AppleCryptoNative_SslRead (pal_ssl.c:389)
5   MailClient.Mobile.iOS           0x203ee3a74         _System_Net_Security_Interop_AppleCrypto__SslRead (LibraryImports.g.cs:2181)
6   MailClient.Mobile.iOS           0x203ef8b38         _System_Net_Security_System_Net_Security_SslStreamPal__DecryptMessage (SslStreamPal.OSX.cs:206)
7   MailClient.Mobile.iOS           0x203ef0ef4         _System_Net_Security_System_Net_Security_SslStream__DecryptData (SslStream.IO.cs:787)
8   MailClient.Mobile.iOS           0x204e080ac         _System_Net_Security_System_Net_Security_SslStream__ReadAsyncInternal_d__170_1<T>__MoveNex
...

However, we also caught a very similar issue on Android in a debugger:

In both cases it's .NET 9.0.12.

The issues look vaguely similar to #117045, #117950 and #117982, which was fixed in .NET 10. If there's a possibility that the root cause is the same perhaps a backport to servicing would be warranted.

[filipnavara]

/cc @kotlarmilos

[dotnet-policy-service]

Tagging subscribers to 'os-ios': @vitek-karas, @kotlarmilos, @steveisok, @akoeplinger
See info in area-owners.md if you want to be subscribed.

[dotnet-policy-service]

Tagging subscribers to this area: @karelz, @dotnet/ncl
See info in area-owners.md if you want to be subscribed.

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/ncl, @bartonjs, @vcsjones
See info in area-owners.md if you want to be subscribed.

[BrzVlad]

.net9 will soon go out of support so I'm not sure about backporting here. Do we have a way to validate that those PRs fix the issues you are encountering ? (ex a relatively deterministic repro). Is upgrading to .net10 a better option instead?

[filipnavara]

Unfortunately we don't have a local repro. It only happens very sporadically and most of the crashes are happing in the wild to the customers. We started seeing a sharp increase in these crashes around the beginning of February but we didn't track down whether it is related to some tooling upgrade, iOS update, or something else. Now we get around thousand of these crashes on weekdays reported through Sentry.

.net9 will soon go out of support

3 months may seem soon but it's still some way to go...

Is upgrading to .net10 a better option instead?

We would love to but due to breaking changes in dependencies (CommunityToolkit.Maui) we had to make a lot of changes that are risky and need proper testing. We struggle to keep up with the update cycles even if we basically had builds on top of .NET 10 pretty early after the official release.

[BrzVlad]

The iOS fix seems very low risk, just adding a private locking object. @kotlarmilos What are your thoughts about doing a backport for this for .net9? Even if we don't have a repro that we can validate.

[kotlarmilos]

Backporting is low risk, so we can try it. Since we can’t reproduce the issue, this backport may not completely fix the problem. I’ll initiate the backport. If you get a reliable repro, please let us know.