Uplink crash: "unaligned 64-bit atomic operation" on 32-bit systems

arrogantrabbit · October 5, 2023, 4:52am

I’ve just filed a bug for a regression in one of the recent versions of uplink causing fatal crash on 32-bit systems due to misaligned 64-bit read:

github.com/storj/uplink

Regression: "unaligned 64-bit atomic operation" on 32-bit systems

opened 04:34AM - 05 Oct 23 UTC

arrogantrabbit

Please look at this thread: https://forum.duplicacy.com/t/duplicacy-3-2-2-cra…shes-on-arm-32-bit/8024/7 To reproduce: build 32-bit arm version of the library and try to use. Regression: - v1.9.0: no issues - v1.12.0: fails. The proposed solution is to use `atomic.AlignedInt64` instead of `int64` in the `MemoryBackend` struct, or reorder `len` and `buf`, ideally -- do both, to reduce waste of space on padding. Quoting analysis from the linked post: > I believe this is an issue in `storj.io/uplink`, which was upgraded from v1.9.0 to v1.12.0 for CLI 3.2.1: https://github.com/gilbertchen/duplicacy/commit/7cc1b4222cb466f9707bd338245f7f22f608e5a7 > > This is the line caused the crash: https://github.com/storj/uplink/blob/c46ff1e3c01a1ef00000e5bf3ae48f4e887f8466/private/storage/streams/buffer/backend.go#L38C31-L38C31 > >This is the definition of `MemoryBackend`: > https://github.com/storj/uplink/blob/c46ff1e3c01a1ef00000e5bf3ae48f4e887f8466/private/storage/streams/buffer/backend.go#L27-L31 > >If they can simply swap `buf` and `len` like this then the error will go away: > >``` >type MemoryBackend struct { > len int64 > buf []byte > closed bool >} >```

Please prioritize…

bre · October 5, 2023, 3:05pm

Thank you for alerting us.
The team said a fix is on the way.

arrogantrabbit · October 5, 2023, 6:08pm

Huge kudos to Storj team – 6 hours from bug report to fix release. I have never seen such a rapid response and attentiveness in any other project!