From: Tvrtko Ursulin <tvrtko.ursulin@onelan.co.uk>
To: Pavel Emelyanov <xemul@parallels.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>,
        David Rientjes <rientjes@google.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org,
        Al Viro <viro@zeniv.linux.org.uk>,
        Alexey Dobriyan <adobriyan@gmail.com>,
        James Bottomley <jbottomley@parallels.com>,
        Matthew Helsley <matt.helsley@gmail.com>,
        aneesh.kumar@linux.vnet.ibm.com, bfields@fieldses.org
Subject: Re: [patch 3/7] fs, notify: Add file handle entry into inotify_inode_mark
Date: Wed, 14 Nov 2012 10:38:16 +0000
Message-ID: <2092535.8S9EcgmZCZ@deuteros>
User-Agent: KMail/4.8.4 (Linux/3.4.6-2.fc17.x86_64; KDE/4.8.4; x86_64; ; )
In-Reply-To: <50A36ED5.4080505@parallels.com>
References: <20121112101440.665694060@openvz.org> <2105540.yeyMVrW4mH@deuteros> <50A36ED5.4080505@parallels.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 7Bit
Content-Type: text/plain; charset="us-ascii"
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2916
Lines: 61

On Wednesday 14 November 2012 14:13:41 Pavel Emelyanov wrote:
> On 11/14/2012 02:08 PM, Tvrtko Ursulin wrote:
> > On Wednesday 14 November 2012 13:58:12 Cyrill Gorcunov wrote:
> >> On Wed, Nov 14, 2012 at 09:50:55AM +0000, Tvrtko Ursulin wrote:
> >>>>> You could not use a pointer and then allocate your buffers on the
> >>>>> check
> >>>>> point operation, freeing on restore?
> >>>> 
> >>>> The problem is not allocating the memory itself but rather the time
> >>>> when
> >>>> the information needed (ie the dentry) is available. The only moment
> >>>> when we can use dentry of the target file/directory is at
> >>>> inotify_new_watch, that's why i need to compose fhandle that early. At
> >>>> any later point we simply have no dentry to use.
> >>> 
> >>> But you do not fundamentally need the dentry to restore a watch, right?
> >> 
> >> dentry only needed to encode the file handle.
> >> 
> >>> Couldn't you restore, creating a new restore path if needed, using the
> >>> inode which is pinned anyway while the watch exists?
> >> 
> >> plain inode is not enough as far as i can tell, iow i don't see the way
> >> to restore path from inode solely. or there something i miss?
> > 
> > I don't know, as I said I was not following this at all until now. Just
> > throwing in ideas.
> > 
> > I thought, since inotify does not use the path or dentry outside the
> > system
> > call at all, perhaps you need a different entry point allowing you to
> > restore the watch using the inode or something. Assuming life time of
> > objects and stuff in C&R world would allow you that. Since you don't need
> > the full path, just something 64 bytes long, I assumed that could be the
> > case.
> 
> Well, the kernel already has all the API we need but one -- it shows us
> _nothing_ about the inode being watched. And we'd appreciate any
> information about it. Even the ino:dev pair would work. We propose to show
> the handle because we believe, that such API is better that ino:dev. You
> can get the handle, call the open_by_handle_at right at once and get much
> much more information about the inode with any other API (e.g. calling
> fstat() will give you the ino:dev pair). Having just ino:dev pair at hands
> is not that flexible.

How much space does a typical file system need to encode a handle? Am I right 
that for must it is just a few bytes? (I just glanced at the code so I might 
be wrong.) In which case, could the handle buffer be allocated dynamically 
depending on the underlying filesystem? Perhaps adding a facility to query a 
filesystem about its maximum handle buffer needs? Do you think the saving 
would justify this extra work?

Regards,

Tvrtko

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/