Subject: Re: LFS frailty vs. datestamping [Was Re: /dev/clock pseudodevice]
To: Manuel Bouyer <>
From: Bill Studenmund <>
List: tech-kern
Date: 07/30/2001 14:14:37
On Mon, 30 Jul 2001, Manuel Bouyer wrote:

> On Mon, Jul 30, 2001 at 01:12:47PM -0700, Bill Studenmund wrote:
> > For general databases, the db usually takes care of journalizing. For your
> > mailspool, a well-behaved MTA will take steps to force the mailspool to be
> > written synchronously.
> Yes, but when you're appending a 10M mail you can't guarantee that you
> won't be interrupted in the middle of the write, leaving a corrupted
> (corrupted because partially written) mailbox on disk.

True. :-)

But if you're writing more to a disk than one full segment's worth of data
and you're doing it in bits smaller than all at once (you aren't appending
the whole 10 MB mail message in one write() call), I think LFS will
capture the partial updates to the inode in the written segments, so a
crash will still give you a corrupted mailbox (since there are good
segments on disk which contain part of the write). :-)

Take care,