Re: journaling filesystem

Hans Reiser (
Fri, 15 May 1998 00:34:30 -0700

Forgive my stupidity, but if I am depending on unlock_buffers() only
being called on a buffer after the buffer has reached disk (more
precisely, reached some state at which the disk won't lose it if power
fails), am I mistaken to depend on this, with the current SCSI drivers
and typical disk drives? This matters to me....
If I lose power, can I frequently lose a buffer write of a buffer that
has already passed through unlock_buffers?

Is IDE safe in this regard?

Less important to me:

With respect to needing an infinite journal, why not just let the FS
track whether there are buffers whose writes are preventing writing the
journal, and then when the journal gets large enough that we fear some
buffer is never getting to disk (which will rarely happen), stall the FS
until we get the report that the buffer was written?
Preserve lists work this way in reiserfs, when I run out of disk space
because I have preserved too many blocks without freeing them, I sync,
and then the preserved blocks are freed. I rarely sync, because usually
disk traffic has enough lulls in it that I don't need to.


Leonard N. Zubkoff wrote:
> Date: Thu, 14 May 1998 02:22:40 -0700
> From: Hans Reiser <>
> Do we need to order writes, or do we need to merely know when a write
> has completed, and let the FS order the writes?
> For reiserfs it is enough to know that the write has completed, where
> completed may be NVRAM on the RAID controller, I don't need to know.
> Precisely. A journalling file system should be happy if we know when the write
> has been committed and if we know that no write request will remain outstanding
> indefinitely so that we don't need an infinite journal.
> Leonard
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to