Re: [sshfs] [issue904] record fails on sshfs (sshfs has no atomic_create)

[Trent W. B. <tre...@gm...>]

On Sun, Jan 25, 2009 at 08:32:00PM +0100, Miklos Szeredi wrote:
> On Sun, 25 Jan 2009, tre...@gm... (Trent W. Buck wrote:
>> Nathaniel W Filardo <nw...@cs...> writes:
>>>>> open(O_CREAT| O_EXCL) is inherently an atomic operation [...]
>>>>> AFAICT, sshfs understands open(O_CREAT|O_EXCL) just fine and
>>>>> carries the O_EXCL bit across the SFTP transport faithfully[...]
>>
>> Would a developer here please confirm the above?
>
> It has been true for a while now (since about 2006).  The
> requirements for atomic create+open are:
>
>   linux kernel>= 2.6.15
>   libfuse>= 2.5
>   sshfs>= 1.3

Thank you for confirming that, I'll get our people to chase it up now.

>> Such confirmation is necessary before Darcs developers will remove
>> a paranoia setting that prevents Darcs from working over sshfs:
>>
>>> Hello fuse-sshfs.  I'm attempting to get a patch included into
>>> darcs to make it fall back to open(O_CREAT|O_EXCL) locking
>>> mechanisms when link() returns ENOSYS as it does on sshfs.  While
>>> it certainly appears that this operation is safe, the darcs
>>> developers would like confirmation that this is the case.
>>
>> It seems Nathaniel didn't receive a reply.  Perhaps this list is
>> for users, and the developer list is elsewhere?
>
> It seems the previous posting didn't make it to the list.  It may
> have been discarded accidentally while moderating for spam.

Hmm, I found it when reading the list via gmane, but perhaps it was
filtered further downstream.  Oh well, doesn't matter.