It may not unpack if the AppImage is on an NTFS partition

[Boria138]

I received a complaint from a user that after updating the kernel to 7.0.0-14-generic in Linux Mint 22.3, AppImage stopped launching, and memfd:dwarfs was showing up in the process manager. After reviewing the strace output, I found that it was simply getting stuck on appimage-extract -and-run. I suggested that the user move the AppImage to their home directory on an NTFS partition, for example, and that solved the problem. This might be a kernel regression, since rolling back to 6.19 also resolves the issue.

log.txt

[Samueru-sama]

appimage-extract-and-run should not be happening by default to start. That's the final fallback from the uruntime.

• It will first try to use FUSE via fusermount, which will mount it on TMPDIR.

• if FUSE is not possible, it will try to use mount namespaces.

• If that is not possible either it will finally fallback to extract and run.

I suggested that the user move the AppImage to their home directory on an NTFS partition, for example, and that solved the problem.

to a NTFS solves the issue?

This might be a kernel regression, since rolling back to 6.19 also resolves the issue.

It is.

Anyways; ping @VHSgunzo

[Samueru-sama]

Just asked AI to check the log.txt file, here is what I got:

Root Cause
This is a kernel regression in the NTFS driver (likely ntfs3) on kernel 7.0.0-14-generic.
The dwarfsextract tool (part of the dwarfs FUSE filesystem library) uses SEEK_DATA/SEEK_HOLE to scan for sparse file regions. 
The expected POSIX behavior at end-of-file is:
- lseek(fd, end_of_file, SEEK_DATA) → should return ENXIO (no more data regions)
- The caller then knows to stop.
On this kernel/NTFS combination, SEEK_DATA at EOF returns the file size instead of ENXIO, so the loop condition never triggers a break. 
The tool spins forever calling lseek() in a tight loop, consuming 100% CPU and never unpacking.

[Samueru-sama]

rip

[Boria138]

to a NTFS solves the issue?

Translation error: It doesn't work on NTFS, presumably due to a regression in the kernel; everything other than NTFS works.