Re: [RFC][PATCH 00/14] VFS: Make all filesystems implement ->show_options()

From: "Theodore Ts'o" <tytso@mit.edu>
To: David Howells <dhowells@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>,
	viro@zeniv.linux.org.uk, linux-fsdevel@vger.kernel.org,
	linux-kernel@vger.kernel.org
Subject: Re: [RFC][PATCH 00/14] VFS: Make all filesystems implement ->show_options()
Date: Sat, 8 Jul 2017 11:33:41 -0400	[thread overview]
Message-ID: <20170708153341.i7h6wma6hykhz5jc@thunk.org> (raw)
In-Reply-To: <24033.1499443136@warthog.procyon.org.uk>

On Fri, Jul 07, 2017 at 04:58:56PM +0100, David Howells wrote:
> Yeah.  We actually currently get this *wrong* in both ext4 and btrfs - and
> probably other disk filesystems too.  During ext4 remount and btrfs remount,
> the options are parsed directly into the live xxx_fs_info struct, but if
> there's a parse error mid-way, we only partially apply the options and have no
> idea where it went wrong and what the current state is.
> 
> What I'm looking it is breaking it down into a number of steps:
> 
>  (1) Parse the options one at a time into a context struct.
> 
>  (2) Once we've got all the options, validate the options we've been given
>      with respect to themselves.
> 
>  (3) Under lock:
> 
>      (a) Check the coherency of the options we've been given with respect to
>      	 the superblock (if new mount) or the current live state (if remount).
> 
>      (b) Apply the options.  This is not permitted to fail.
> 
> This gets more complicated under ext4 as you've got an extra option string
> stored on disk that you also have to apply and combine with the options
> presented to the mount interface.

Remounts aren't really performance critical, so it might be simpler
just to do one pass which just checks the options, applying the
options to a scratch xx_fs_info struct, and if it seeded, redo the
parse from scratch, this time applying to the live xxx_fs_info struct.

      	   	    	      	       	  - Ted