Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Mon, 15 Mar 2021 12:46:44 +0000
From:   Matthew Wilcox <willy@infradead.org>
To:     David Howells <dhowells@redhat.com>
Cc:     Alexander Viro <viro@zeniv.linux.org.uk>,
        Miklos Szeredi <miklos@szeredi.hu>,
        Ian Kent <raven@themaw.net>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [RFC][PATCH 0/3] vfs: Use an xarray instead of inserted
 bookmarks to scan mount list
Message-ID: <20210315124644.GU2577561@casper.infradead.org>
References: <161581005972.2850696.12854461380574304411.stgit@warthog.procyon.org.uk>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <161581005972.2850696.12854461380574304411.stgit@warthog.procyon.org.uk>
Precedence: bulk

On Mon, Mar 15, 2021 at 12:07:39PM +0000, David Howells wrote:
> 
> Hi Al, Mikl?s,
> 
> Can we consider replacing the "insert cursor" approach we're currently
> using for proc files to scan the current namespace's mount list[1] with
> something that uses an xarray of mounts indexed by mnt_id?
> 
> This has some advantages:
> 
>  (1) It's simpler.  We don't need to insert dummy mount objects as
>      bookmarks into the mount list and code that's walking the list doesn't
>      have to carefully step over them.
> 
>  (2) We can use the file position to represent the mnt_id and can jump to
>      it directly - ie. using seek() to jump to a mount object by its ID.
> 
>  (3) It might make it easier to use RCU in future to dump mount entries
>      rather than having to take namespace_sem.  xarray provides for the
>      possibility of tagging entries to say that they're viewable to avoid
>      dumping incomplete mount objects.

Usually one fully constructs the object, then inserts it into the XArray.

> But there are a number of disadvantages:
> 
>  (1) We have to allocate memory to maintain the xarray, which becomes more
>      of a problem as mnt_id values get scattered.

mnt_id values don't seem to get particularly scattered.  They're allocated
using an IDA, so they stay small (unlike someone using idr_alloc_cyclic
;-).