On 2015-10-16 08:24, Christoph Hellwig wrote:
> On Fri, Oct 16, 2015 at 01:02:23PM +0100, P??draig Brady wrote:
>> Right. reflinking is transparent to the user, though its consequences are not.
>> Consequences being the possible extra latency or ENOSPC on CoW.
>
> You can get all these consequences without doing the file system reflink
> by using a COW file system, any dedup scheme or thinly provisioned or
> COW storage devices.
>
>> Therefore reflinking should be an explicit action/flag IMHO.
>
> This still does not make any sense, as it only prevents one of many
> ways a file could do COW operations underneath.  If you don't want
> ENOSPC use fallocate, or the proposed COPY_FALLOC flag.  If you want
> care about latency you need to carefull benchmark your setup but in
> general falloc / COPY_FALLOC might be a good starting point.  But for
> 99% of the copies a reflink is exactly the right thing to do.
There is at least one reason other than avoiding ENOSPC and minimizing 
latency that people may want to avoid reflinking things: They actually 
_want_ multiple physically independent copies of the same file on the 
disk.  Usually people do go about this wrong (some people I know don't 
understand that having multiple copies of a file on the same filesystem 
provides no greater safety than one copy), but that doesn't mean that 
this isn't a perfectly valid use case for copying a file.