You are asking for something that Linux is not capable of doing.
There is no means to send and error back from the disk flush to the
fs/appilcation period. The current 2.4 can not even find the partition
when send an error back up to the block layer. 2.5 has a chance, but
currently none have the ablitity to notify or flush disk cache and recover
of there is a flushcache error.
Therefore it is potential a preferred model to preserve the entire request
for a retry than to do a partial validation of an incomplete attempt.
On Mon, 17 Dec 2001, Joel Becker wrote:
> On Mon, Dec 17, 2001 at 11:59:56AM -0800, Linus Torvalds wrote:
> > On Mon, 17 Dec 2001, Joel Becker wrote:
> > > /* Smart program handles partial writes */
> > > write(100k); = 50k
> > > write(remaining 50k); = -1/ENOSPC|EIO|etc
> >
> > We do this, if the error is "hard". And "fatal" implies hardness, so we're
> > ok here.
>
> Right. "hard" is also synonymous with "non-transient".
>
> > > /* Dumb program doesn't handle partial write */
> > > write(100k); = 50k
> > > close(fd); = -1/EIO
> >
> > But we're not doing this.
>
> IMHO we should be, and not just to comply with the letter of
> SUS/Unix98. SUS specifies this behavior because a synchronous write()
> can return after copying data to the buffer cache. However, the EIO can
> happen later when the buffer cache is trying to flush to disk. The only
> way for an application to see this error is to either run O_SYNC or
> receive it upon close().
Andre Hedrick
CEO/President, LAD Storage Consulting Group
Linux ATA Development
Linux Disk Certification Project
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sun Dec 23 2001 - 21:00:14 EST