RUSTSEC-2022-0041: Unsoundness of AtomicCell<*64> arithmetics on 32-bit targets that support Atomic*64

[github-actions]

Details
Package	crossbeam-utils
Version	0.7.2
Warning	unsound
URL	crossbeam-rs/crossbeam#781
Patched Versions	>=0.8.7
Aliases	GHSA-qc84-gqf4-9926, CVE-2022-23639

Impact

Affected versions of this crate incorrectly assumed that the alignment of {i,u}64 was always the same as Atomic{I,U}64.

However, the alignment of {i,u}64 on a 32-bit target can be smaller than Atomic{I,U}64.

This can cause the following problems:

• Unaligned memory accesses
• Data race

Crates using fetch_* methods with AtomicCell<{i,u}64> are affected by this issue.

32-bit targets without Atomic{I,U}64 and 64-bit targets are not affected by this issue.

32-bit targets with Atomic{I,U}64 and {i,u}64 have the same alignment are also not affected by this issue.

The following is a complete list of the builtin targets that may be affected. (last update: nightly-2022-02-11)

• armv7-apple-ios (tier 3)
• armv7s-apple-ios (tier 3)
• i386-apple-ios (tier 3)
• i586-unknown-linux-gnu
• i586-unknown-linux-musl
• i686-apple-darwin (tier 3)
• i686-linux-android
• i686-unknown-freebsd
• i686-unknown-haiku (tier 3)
• i686-unknown-linux-gnu
• i686-unknown-linux-musl
• i686-unknown-netbsd (tier 3)
• i686-unknown-openbsd (tier 3)
• i686-wrs-vxworks (tier 3)

(script to get list)

Patches

This has been fixed in crossbeam-utils 0.8.7.

Affected 0.8.x releases have been yanked.

Thanks to @taiki-e

[radhermit]

This is due to ipc-channel lagging on porting to a newer mio version.