On 2015-10-16 08:21, Christoph Hellwig wrote:
> On Fri, Oct 16, 2015 at 07:46:41AM -0400, Austin S Hemmelgarn wrote:
>> I should have been specific, what I meant was that some people will assume
>> that it actually creates a physical, on-disk byte-for-byte copy of the data.
>> There are many people out there (and sadly I have to deal with some at work)
>> who are absolutely terrified of the concept of data deduplication, and will
>> likely refuse to use this syscall for _anything_ if it reflinks by default
>> on filesystems that support it.
>
> If they use a file system that supports COW or dedup they are toast
> already.  It's not the system call that does the 'dedup', it's the file
> system or storage device, that's where they need to set their
> preferences.
BTRFS is COW and supports deduplication, it does _absolutely zero_ 
reflinking and/or deduplication unless you explicitly tell it to do so. 
  Likewise, ZFS is COW and supports deduplication, it also does 
_absolutely zero_ reflinking and/or deduplication unless you tell it to 
(note that in-band deduplication is off by default on ZFS).  AFAIK, XFS 
will not automatically reflink instead of copying either (and if it does 
decide to do it automatically, that will just be something else to add 
to the list of why I will never use it on any of my systems). OCFS2 
supports reflinks (although not many people know this, and I think it 
implements them slightly differently from BTRFS/ZFS/XFS) and yet again, 
does _absolutely zero_ reflinking unless you tell it to.  Based on this, 
it is in no way the filesystem that does the deduplication and 
reflinking, it only handles the implementation and provides the option 
to the user to do it.

Certain parts of userspace do try to reflink things instead of copying 
(for example, coreutils recently started doing so in mv and has had the 
option to do so with cp for a while now), but a properly designed 
general purpose filesystem does not and should not do this without the 
user telling it to do so.