clone syscall fails for bedup-deduplicated files

clone syscall fails for bedup-deduplicated files

[Pascal Germroth]

Hi,

Running a 3.8 kernel I've been using bedup to deduplicate some files, for example two identical 500MB files A & B now share data chunks.
Running a 3.11 kernel, I wanted to clone some more files, but those previously deduplicated by bedup don't work, for example: `cp --reflink A C`:

open("A", O_WRONLY|O_TRUNC) = 4
fstat(4, {st_mode=S_IFREG|0775, st_size=0, ...}) = 0
ioctl(4, BTRFS_IOC_CLONE, 0x3) = -1 EINVAL (Invalid argument)

But I can clone a duplicate of the file just fine: `cp A D; cp --reflink D E` copies A to D and then clones D to E.
So I guess the way bedup deduplicated A broke the btrfs structure in a way that forbids further cloning, but a newly generated file like D can still be cloned.

How could bedup break this, did it manipulate the structures directly in 3.8 without the syscall and corrupt something? How can I repair this? Defragmenting A had no effect.

(If you answer, please CC me directly, I'm not currently subscribed to the list)

Cheers,

-- 
Pascal Germroth

Re: clone syscall fails for bedup-deduplicated files

[David Sterba]

On Sat, Dec 07, 2013 at 10:31:30PM +0100, Pascal Germroth wrote:
> Running a 3.8 kernel I've been using bedup to deduplicate some files,
> for example two identical 500MB files A & B now share data chunks.
> Running a 3.11 kernel, I wanted to clone some more files, but those
> previously deduplicated by bedup don't work, for example: `cp
> --reflink A C`:
> 
> open("A", O_WRONLY|O_TRUNC) = 4
> fstat(4, {st_mode=S_IFREG|0775, st_size=0, ...}) = 0
> ioctl(4, BTRFS_IOC_CLONE, 0x3) = -1 EINVAL (Invalid argument)

This can fail eg when the files have different status wrt checksums,
when one of the files is nocow or created under nodatasum mount (there's
no way to create nodatasum file otherwise, unlike nodatacow).

Look at lsattr output if the files have the C attribute.

david