Re: write alignment matters?

[Paul Koning <paul_koning%dell.com@localhost>]

On Jun 28, 2011, at 12:53 PM, David Laight wrote:

> On Mon, Jun 27, 2011 at 05:55:50PM +0000, David Holland wrote:
>> RT-11 did not require explictly unmounting floppies before changing
>> them.
> 
> IIRC RT-11 fs could go wrong (docs suggested using TECO to read the raw disk
> and recover the file data from the adjacent sectors that contained it)
> I don't think you wanted to yank a disk with an open (for extending) file on 
> it.

We're going a bit off topic, but what the heck.  This does tie a bit into 
earlier comments.

If you do that to RT-11, the file system would be intact but data written on a 
file not yet closed would be invisible since the file length hadn't been 
updated yet.

Moving back on topic, it would be good for the OS to be able to survive loss of 
a device even if the file system on the device ends up broken.  And, one step 
further, it would be good for the file system to be structurally intact in such 
an event, though of course any buffered data wouldn't be there, and possibly 
additional data whose metadata hasn't been written would also be inaccessible.

A journaled file system does the latter without help, a non-journaled file 
system needs fsck.  Either way, just because one file system has gone away 
doesn't mean the whole OS should be stuck.

        paul