GNU bug report logs - #52516
Avoiding race condition between partprobe and getting PARTUUID using lsblk?

Previous Next

To add a comment to this bug, you must first unarchive it, by sending
a message to control AT debbugs.gnu.org, with unarchive 52516 in the body.
You can then email your comments to 52516 AT debbugs.gnu.org in the normal way.

Toggle the display of automated, internal messages from the tracker.

Message #5 received at submit <at> debbugs.gnu.org (full text, mbox):

From: Anton Hvornum <anton <at> hvornum.se>
To: bug-parted <at> gnu.org
Subject: Avoiding race condition between partprobe and getting PARTUUID using
 lsblk?
Date: Wed, 15 Dec 2021 13:12:43 +0000

Hi.

So I've been struggling a bit with optimizing applications calling
partprobe right after lsblk to retrieve PARTUUID of newly generated
partitions using parted.

Assuming the following loop:
* partprobe
* lsblk
* if !result: sleep + loop

It feels as if every partprobe (indirectly or directly) flushes a
cache/struct of information somewhere, which makes lsblk "fail" to
retrieve certain information if executed too quickly in succession. A
normal approach to solving this is adding a delay between partprobe
and lsblk, something I wanted to avoid by putting the sleep at the
very end of the loop, and instead do many quick loops until the
information is available. I could also put partprobe before the loop,
and continue looping until the information becomes available. But that
poses an issue if the disk-operation (parted) completes after
partprobe performs it's tasks. Leaving the loop in a dead state until
a new partprobe is called (or that seems to be the issue at least).

My question boils down to:
 1. Is there a way to get an indication of when the partprobe (or
kernel) has completed its task of populating the partition
information?
 2. Is there a way to tell partprobe/kernel to not clear the old
struct before the new information is available (leaving "old
information" intact until new exists, rather than wiping it and then
populating)

If none of the two above is possible, I would like to consider those
being a feature as it would greatly help to verify when disk
operations are 100% complete. I don't mind optionally hanging
applications/scripts until parted is complete, as I would like to
avoid continuing based on parted exit code if the exit code is not a
guarantee of the process being completed in this case.

Best wishes:
Anton Hvornum

Message #10 received at 52516-close <at> debbugs.gnu.org (full text, mbox):

From: "Brian C. Lane" <bcl <at> redhat.com>
To: 52516-close <at> debbugs.gnu.org
Subject: Re: bug#52516: Avoiding race condition between partprobe and getting
 PARTUUID using lsblk?
Date: Wed, 15 Dec 2021 08:54:50 -0800

On Wed, Dec 15, 2021 at 01:12:43PM +0000, Anton Hvornum wrote:

[snip]

> My question boils down to:
>  1. Is there a way to get an indication of when the partprobe (or
> kernel) has completed its task of populating the partition
> information?

udevadm settle should help. The problem is that partprobe tells the
kernel about all the partitions, the kernel then tells udev about the
changes and then udev updates device nodes. So you cannot depend on
anything being stable until udev is finished.

>  2. Is there a way to tell partprobe/kernel to not clear the old
> struct before the new information is available (leaving "old
> information" intact until new exists, rather than wiping it and then
> populating)

No. It has to refresh everything, otherwise some parts of it might get
out of sync.

> If none of the two above is possible, I would like to consider those
> being a feature as it would greatly help to verify when disk
> operations are 100% complete. I don't mind optionally hanging
> applications/scripts until parted is complete, as I would like to
> avoid continuing based on parted exit code if the exit code is not a
> guarantee of the process being completed in this case.

There's not much you can do other than wait for udev, or check to make
sure device nodes you expect to be present are there. We used to hit
problems like this in the parted test suite all the time, until we added
loops to wait for the partitions to appear.

Good Luck!

Brian

-- 
Brian C. Lane (PST8PDT) - weldr.io - lorax - parted - pykickstart

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.