Re: [PATCH] docs/power: note Btrfs limitation with FIBMAP for swap file offset

Re: [PATCH] docs/power: note Btrfs limitation with FIBMAP for swap file offset

[Christoph Hellwig]

On Tue, May 12, 2026 at 12:07:13PM +0530, Arun Rao Balappa wrote:
> On Btrfs, the FIBMAP ioctl does not return physical block addresses.

On btrfs, and on anything that writes out of place or uses multiple
devices for that matter, FIBMAP is not implemented.

> Tools such as filefrag therefore cannot determine the correct swap file
> offset for use as resume_offset. Document the correct btrfs-progs
> command to use instead.

All of this is inherently unsafe.  File systems and do move file data
without notifying users.  This document is a really bad idea and
should be removed as we should not encourage users to rely on these
kinds of hacks.

Re: [PATCH] docs/power: note Btrfs limitation with FIBMAP for swap file offset

[Arun Rao]

Understood. My intent was mainly to document the current behaviour
observed by users on Btrfs systems where filefrag/FIBMAP-based
guidance fails, and to point users toward the existing btrfs-progs
tooling.

That said, I understand the concern about encouraging users to rely on
resume_offset-based workflows in general given the underlying
fragility of file block mappings on moving or CoW filesystems.

Re: [PATCH] docs/power: note Btrfs limitation with FIBMAP for swap file offset

[Christoph Hellwig]

On Tue, May 12, 2026 at 03:12:44PM +0530, Arun Rao wrote:
> Understood. My intent was mainly to document the current behaviour
> observed by users on Btrfs systems where filefrag/FIBMAP-based
> guidance fails, and to point users toward the existing btrfs-progs
> tooling.
> 
> That said, I understand the concern about encouraging users to rely on
> resume_offset-based workflows in general given the underlying
> fragility of file block mappings on moving or CoW filesystems.

Yes.  Can you maybe extend the text a bit to generally warn about this?

I also realized that filefrag uses FIEMAP by default these days and not
just FIBMAP, which is what makes it so dangerous on btrfs.  btrfs
smartly does not implement ->bmap, but it does implement FIEMAP in
this weird way mapping to the logical address space.