xfs
[Top] [All Lists]

Re: [fuse-devel] utimensat fails to update ctime

To: Jean-Pierre André <jean-pierre.andre@xxxxxxxxxx>
Subject: Re: [fuse-devel] utimensat fails to update ctime
From: OGAWA Hirofumi <hirofumi@xxxxxxxxxxxxxxxxxx>
Date: Tue, 22 Dec 2009 22:30:24 +0900
Cc: Eric Blake <ebb9@xxxxxxx>, fuse-devel@xxxxxxxxxxxxxxxxxxxxx, Miklos Szeredi <miklos@xxxxxxxxxx>, Christoph Hellwig <hch@xxxxxx>, Linux Kernel Mailing List <linux-kernel@xxxxxxxxxxxxxxx>, xfs@xxxxxxxxxxx
In-reply-to: <4B30B67A.7080703@xxxxxxxxxx> ("Jean-Pierre André"'s message of "Tue, 22 Dec 2009 13:07:22 +0100")
References: <4B2B156D.9040604@xxxxxxx> <87aaxclr4q.fsf@xxxxxxxxxxxxxxxxxxx> <4B2F7421.10005@xxxxxxx> <4B2F7A95.3010708@xxxxxxx> <87hbrkjrk8.fsf@xxxxxxxxxxxxxxxxxxx> <4B304D04.6040501@xxxxxxx> <87d427jscr.fsf@xxxxxxxxxxxxxxxxxxx> <4B3097C4.3060803@xxxxxxxxxx> <874onjjnln.fsf@xxxxxxxxxxxxxxxxxxx> <4B30B67A.7080703@xxxxxxxxxx>
User-agent: Gnus/5.13 (Gnus v5.13) Emacs/23.1.90 (gnu/linux)
Jean-Pierre André <jean-pierre.andre@xxxxxxxxxx> writes:

> Hi again,

Hi,

>> Well, the problem seems in fuse_lib_setattr() and ntfs_fuse_setattr()
>> (lowlevel op too).
>>
>> The both functions is requiring "ATIME | MTIME". Doesn't it mean the
>> ntfs-3g can't set only MTIME like above utimensat()?
>>    
>
> With ntfs-3g this is not directly possible, because
> the interface does not provide flags telling which
> timestamps should be updated. The only way would
> be fuse feeding both values (even though unchanged)
> before calling ntfs-3g. This is true for all versions of
> ntfs-3g.

Yes, with fuse_operations. It is why I'm saying the issue is libfuse or
ntfs-3g.

But I noticed ntfs-3g is including libfuse-lite sources and use it with
static link (I might be wrong here, because I just looked ntfs-3g source
slightly). AFAIK, the fuse of kernel part is passing the flags of some
sort of detail always.

[BTW, the code of that part in kernel may be the following,

fs/fuse/dir.c:iattr_to_fattr(),

        if (ivalid & ATTR_ATIME) {
                arg->valid |= FATTR_ATIME;
                arg->atime = iattr->ia_atime.tv_sec;
                arg->atimensec = iattr->ia_atime.tv_nsec;
                if (!(ivalid & ATTR_ATIME_SET))
                        arg->valid |= FATTR_ATIME_NOW;
        }
        if ((ivalid & ATTR_MTIME) && update_mtime(ivalid)) {
                arg->valid |= FATTR_MTIME;
                arg->mtime = iattr->ia_mtime.tv_sec;
                arg->mtimensec = iattr->ia_mtime.tv_nsec;
                if (!(ivalid & ATTR_MTIME_SET))
                        arg->valid |= FATTR_MTIME_NOW;
        }
]

So, if libfuse-lite was fixed to supported that update request, it would
be able to do even if fuse_operations.  I.e. in libfuse-lite, emulate
"ATIME | MTIME" request if "MTIME" only (pass unchanged original atime),
then call ->utime() callback. (or adds new utime2 callback with flags, or
something other solutions)

Or, if that request is known limitation of fuse_operations, I think it
would be clear state and ok. The fs needs to use lowlevel op to support
that.

> With lowntfs-3g (release candidate only), this could
> be possible.... but this is not implemented, as the
> case was never found up to now. I can provide you
> with a patch,... if fuse can feed in the flags selectively.

Yes.  AFAIK, fuse of kernel part is passing FATTR_MTIME without
FATTR_ATIME to userland (i.e. FUSE_SET_ATTR_ATIME and
FUSE_SET_ATTR_MTIME in libfuse).

I think it's good to implement if it's not design decision of ntfs-3g.

[BTW, just my guess though, it would be good to use "if (vaild &
ATTR_XXX)" style, not "switch()" to support various combinations of
flags]

Thanks.
-- 
OGAWA Hirofumi <hirofumi@xxxxxxxxxxxxxxxxxx>

<Prev in Thread] Current Thread [Next in Thread>