Questions about extended file attributes (xattrs)

[pjotrek-b]

Hello everyone 😄

It's about btrfs-documentation:
I'd like to ask about technical details on extended attributes (xattrs) on btrfs, and found these pages:

• https://docs.kernel.org/6.1/filesystems/btrfs.html
• https://btrfs.readthedocs.io/en/latest/

Which both don't really list or mention extended file attributes handling (xattrs).
I know btrfs can do xattrs, and ArchWiki mentions possible size-constraints, but I'm interested in the specific limits and settings for xattrs on btrfs 😄

I'm moving this as issue as question here, following advice from @kdave in btrfs/linux#1455

If anyone wants to talk about this or point me towards some code/docs, I'd be very grateful! 🍰

[adam900710]

XATTRS is stored with key (<inode number> XATTR_ITEM 0), its related data is btrfs_file_extent_item but only in the inline extent mode.

And xattrs are stored as inline data extents, meaning it's limited by two things:

• Sector size
  Inline data extent can not reach sector size. For most filesystems it's 4K.
  This is just a soft limit, but will make kernel implementation much easier.

• Inline data extent size
  This is limited by node size (metadata size), since we're storing everything inside metadata.
  This is a hard limit, but shouldn't cause any hassle because we're using 16KiB as node size for a long long time.

[pjotrek-b]

@adam900710 : This is great information, thank you!

Do I understand this correctly:

1. 4k sector size:
   Every xattr byte (all keys + all values) must stay below 4k?
   "Soft limit" means: Theoretically adjustable - but not really tested or really laid out for?

2. "16kiB as node size"
   I don't yet fully understand this.
   Do you mean: Internally btrfs could handle up-to-16kiB xattrs - if the sector size would be that big?

So if I'd like to use xattrs like BeFS or HFS/APFS where even binary-data (eg images) are stored: would btrfs' defaults be capped to 4k in total right now?

Thank you so much! 😄