Bug 11941

Summary:	losetup and partprobe fail when partitions exist
Product:	Busybox	Reporter:	Alkis Georgopoulos <alkisg>
Component:	Other	Assignee:	unassigned
Status:	RESOLVED FIXED
Severity:	normal	CC:	busybox-cvs
Priority:	P5
Version:	unspecified
Target Milestone:	---
Hardware:	All
OS:	Linux
Host:		Target:
Build:

Description Alkis Georgopoulos 2019-06-05 07:56:40 UTC

Hi! I have the following issue with busybox:
$ busybox losetup /dev/loop0 /dev/sda
$ busybox partprobe /dev/loop0
partprobe: /dev/loop0: Invalid argument

While partprobe from libparted works fine.
Additionally, `losetup -P /dev/loop0 /dev/sda` from util-linux also works fine; if `busybox losetup` supported -P, it would solve the problem.

I think the reason for this issue is stated in https://git.kernel.org/pub/scm/utils/util-linux/util-linux.git/tree/lib/loopdev.c#n899

> - kernels >= 3.2 enable PT scanner only if max_part= is non-zero or if the
>   LO_FLAGS_PARTSCAN flag is set for the device. The PT scanner is disabled
>   by default.

So, `losetup -P` could set LO_FLAGS_PARTSCAN and solve the issue.

As a workaround, I'm currently using `modprobe loop max_part=9`, when loop isn't built-in to the kernel, and `loop.max_part=9` in the kernel cmdline when it is.
This indirectly re-enables the PT scanner, but this is a bad workaround, and not always easy to implement.
Please allow busybox to handle loop devices with partitions without having to specify max_part.

Comment 1 Alkis Georgopoulos 2019-06-05 08:05:45 UTC

This happens for blockdev too:

$ busybox losetup /dev/loop0 /dev/sda
$ busybox blockdev --rereadpt /dev/loop0
blockdev: /dev/loop0: Invalid argument

And again the workaround works, i.e. it doesn't happen if `modprobe loop max_part=9` was used before those 2 commands.

Comment 2 Denys Vlasenko 2019-06-07 13:36:15 UTC

commit 726cbb1be8620496cdf05d23bd98bdeb0983b4f0
Author: Jack O'Sullivan <jackos1998@gmail.com>
Date:   Tue May 28 15:28:27 2019 +0100

    losetup: Add partition scanning option

    Add -P option from util-linux losetup to scan for partitions.

Comment 3 Denys Vlasenko 2019-06-08 06:42:39 UTC

losetup -P was implemented this May.

This does not solve partprobe / blockdev --rereadpt examples (in busybox, they both boil down to calling ioctl(BLKRRPART)) - from what I see in strace, partprobe from GNU parted 3.2 does not use this ioctl at all, it does some complicated magic with newer APIs designed to set up arbitrary pertitioning schemes from userspace without having kernel to know anything about partition tables.

"blockdev --rereadpt" does not work for util-linux's blockdev as well, since it also simply calls ioctl(BLKRRPART).

IOW: "losetup -P" issue is fixed, "extend partprobe" issue is a significantly taller order.

Comment 4 Alkis Georgopoulos 2019-06-08 07:10:24 UTC

When I saw your commit link in Comment #2, it took me a while to realize how you were able to solve the issue 10 days before the bug report! :)

`losetup -P` is more than enough, thank you very much.