From: Eric Sandeen <sandeen@redhat.com>
Subject: Re: [PATCH 5/7] mke2fs: proceed if the user doesn't type anything
 after 5 seconds
Date: Mon, 28 Apr 2014 19:32:23 -0500
Message-ID: <535EF317.7050102@redhat.com>
References: <1398556834-31913-1-git-send-email-tytso@mit.edu> <1398556834-31913-5-git-send-email-tytso@mit.edu> <535E74D4.4050805@redhat.com> <20140428232634.GD7857@thunk.org>
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Cc: Ext4 Developers List <linux-ext4@vger.kernel.org>
To: "Theodore Ts'o" <tytso@mit.edu>
In-Reply-To: <20140428232634.GD7857@thunk.org>
Sender: linux-ext4-owner@vger.kernel.org

On 4/28/14, 6:26 PM, Theodore Ts'o wrote:
> On Mon, Apr 28, 2014 at 10:33:40AM -0500, Eric Sandeen wrote:
>> On 4/26/14, 7:00 PM, Theodore Ts'o wrote:
>>> If mke2fs needs to ask the user for permission, and the user doesn't
>>> type anything for five seconds, proceed as if the user had said yes.
>>>
>>> This will allow us to add more stringent checks without breaking
>>> existing scripts (much).
>>
>> Hm, this sounds a little dangerous - "-F" overrides a lot.
> 
> Actually, if you take a look at what we use proceed_question() for, it
> doesn't actually override anything (up until now) that might lead to
> data loss.  It's for things like trying to create an file system with
> a block size greater than 4k on an x86 platform, creating a file
> system larger than the apparent block size, etc.  The main goal was to
> make sure the user actually *sees* the darned message.
> 
> Perhaps the only case where proceed_question() can prevent data loss
> is the one where the user typo's /dev/sda3 as /dev/sda.  Everything
> else is in the category of "we want to make sure the user sees the
> warning".
> 
> The motivation behind this is adding this safety check:
> 
> % ./misc/mke2fs -t ext4 -L test-filesystem /dev/sdc3 8M
> mke2fs 1.42.9 (4-Feb-2014)
> /dev/sdc3 contains a ext4 file system labelled 'test-filesystem'
> Proceed anyway (or wait 5 seconds) ? (y,n) 
> 
> Previously, we would blithely blow away /dev/sdc3 without even giving
> a warning.  So if stdin (fd 0) is not a tty, we skip this test
> entirely --- otherwise existing scripts would fail.  However, if a
> script is attached to a tty, we would end up stalling the script
> waiting for the user to answer yes/no where previously no question
> would be asked at all.  This is the case where it's important that
> proceed_question() will now pause five seconds, and then continue.

I guess it's up to you, but it gives me the heebie-jeebies.  xfs
and btrfs already stop on an existing fs (or a partition table) unless
the script adds the force option.  Stopping to make sure about an
irreversible action - but proceeding after 5s anyway - seems to me
like the worst of both worlds.  If it doesn't matter, don't ask.
If it matters, wait for a response, however long it might take.

At least that's my take on it.  :)

-Eric

> 		   		   	    	       - Ted
>