Re: multiple pending filesystem requests for the same block ?

In message <86mzo94qks.fsf_at_xps.des.no>, =?iso-8859-1?q?Dag-Erling_Sm=F8rgrav?= 
writes:
>Luigi Rizzo <rizzo_at_icir.org> writes:
>> can the filesystem code have multiple pending write requests for
>> the same disk block ?
>
>This was discussed on the lists a few weeks ago, and ISTR the
>conclusion was that it is the filesystem's responsibility to ensure
>this never happens.

Not quite.

It is legal, but the order is undefined (and indeterminable!).

I have a hard time figuring out a filesystem design that could use such
writes profitably on regular disks.

On raw flash devices however, writes are either "NOR" or "NAND"'ed
into the existing data and therefore the order may not be important
and therefore, such writes are explicitly legal.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk_at_FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.