mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* Rereading the Partition Table (ioctl BLKRRPART)
@ 2003-10-20 15:38 Calin A. Culianu
  2003-10-20 16:38 ` Andreas Dilger
  0 siblings, 1 reply; 3+ messages in thread
From: Calin A. Culianu @ 2003-10-20 15:38 UTC (permalink / raw)
  To: linux-kernel

Hi all,

I was adminning one of my IDE hard drives the other day, taking an
existing partition and slicing and dicing it a bit to break it out into a
few partitions, and was slightly annoyed by something that has always been
true in Linux (and in any OS really):  the need to completely unmount all
partitions on a drive (getting the usecount for the entire device to 0) in
order to successfully re-read the partition table.  In many cases if the
drive also has the rootfs on it, this effectively means rebooting.

Anyhow, in my quest to keep my uptime as high as possible, I was wondering
how feasible it would be to make the BLKRRPART ioctl a little more
flexible, so that in some cases a reboot wouldn't be required.

Of course, if you modify a parition that is currently in use (ie: has a
mounted fs), then a reboot would be necessary.  Also, if you insert new
partitions in such a way that existing partitions would be 'renumbered',
perhaps a reboot would still be necessary.

However, consider the following case:

1. You have a disk with some unused space at the end, say hda5.  Your
    Rootfs is hda1, and your swapfs is hda2. (hda5 is not mounted.)

2. You repartition hda5 into hda5 and hda6 (by splitting it in two).

3. You would like the changes to take effect (meaning hda6 now points to a
   real block device, and hda5 is now resized) immediately, without a
   reboot.

4. You issue the BLKRRPART ioctl. (well, not you, but, say, cfdisk or
   fdisk does)

5. The linux kernel groks the partition table, realizes that even though
   hda has a usecount > 0, the partition table changes are ok, since
   they don't cause any insanity.. so the kernel resizes hda5 in its data
   structures, and registers hda6.  It returns 0 for success.

6. Your uptime is not affected since you don't have to reboot.

7. PROFIT!!!! :)


Well, what do you guys think?  I am tempted to just hack the sources now
to get this working for myself, but before I start I am afraid maybe the
situation isn't quite so simple... or there might be something I am
overlooking.  Can someone with more experience in the kernel share their
thoughts on this?

Thanks!

-Calin





^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: Rereading the Partition Table (ioctl BLKRRPART)
  2003-10-20 15:38 Rereading the Partition Table (ioctl BLKRRPART) Calin A. Culianu
@ 2003-10-20 16:38 ` Andreas Dilger
  2003-10-20 17:26   ` Calin A. Culianu
  0 siblings, 1 reply; 3+ messages in thread
From: Andreas Dilger @ 2003-10-20 16:38 UTC (permalink / raw)
  To: Calin A. Culianu; +Cc: linux-kernel

On Oct 20, 2003  11:38 -0400, Calin A. Culianu wrote:
> Anyhow, in my quest to keep my uptime as high as possible, I was wondering
> how feasible it would be to make the BLKRRPART ioctl a little more
> flexible, so that in some cases a reboot wouldn't be required.
> 
> Well, what do you guys think?  I am tempted to just hack the sources now
> to get this working for myself, but before I start I am afraid maybe the
> situation isn't quite so simple... or there might be something I am
> overlooking.  Can someone with more experience in the kernel share their
> thoughts on this?

It already exists, and newer partition editors already support this.  I
think parted and partx will use the new partition ioctls to change the
kernel's partition table without a reboot.  Sadly, no documentation on
how to use partx.

Cheers, Andreas
--
Andreas Dilger
http://sourceforge.net/projects/ext2resize/
http://www-mddsp.enel.ucalgary.ca/People/adilger/


^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: Rereading the Partition Table (ioctl BLKRRPART)
  2003-10-20 16:38 ` Andreas Dilger
@ 2003-10-20 17:26   ` Calin A. Culianu
  0 siblings, 0 replies; 3+ messages in thread
From: Calin A. Culianu @ 2003-10-20 17:26 UTC (permalink / raw)
  To: Andreas Dilger; +Cc: linux-kernel

On Mon, 20 Oct 2003, Andreas Dilger wrote:

> On Oct 20, 2003  11:38 -0400, Calin A. Culianu wrote:
> > Anyhow, in my quest to keep my uptime as high as possible, I was wondering
> > how feasible it would be to make the BLKRRPART ioctl a little more
> > flexible, so that in some cases a reboot wouldn't be required.
> >
> > Well, what do you guys think?  I am tempted to just hack the sources now
> > to get this working for myself, but before I start I am afraid maybe the
> > situation isn't quite so simple... or there might be something I am
> > overlooking.  Can someone with more experience in the kernel share their
> > thoughts on this?
>
> It already exists, and newer partition editors already support this.  I
> think parted and partx will use the new partition ioctls to change the
> kernel's partition table without a reboot.  Sadly, no documentation on
> how to use partx.

Really? Coolness..  so the onus is on the userspace process to specify
which partitions are suspected of having changed?  Is that the reason for
a new ioctl?

Just out of curiousity: What was the motivation for adding a new ioctl?
What would have been wrong with also hacking BLKRRPART to behave this way,
as in, have it figure out if a re-read is possible based on the heuristics
I already suggested...

-Calin


^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2003-10-20 17:26 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2003-10-20 15:38 Rereading the Partition Table (ioctl BLKRRPART) Calin A. Culianu
2003-10-20 16:38 ` Andreas Dilger
2003-10-20 17:26   ` Calin A. Culianu

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®