Bug 90662

Summary:	cryptsetup hangs in initramfs
Product:	systemd	Reporter:	Evangelos Foutras <evangelos>
Component:	general	Assignee:	systemd-bugs
Status:	RESOLVED FIXED	QA Contact:	systemd-bugs
Severity:	normal
Priority:	medium	CC:	ht990332, wshuman3
Version:	unspecified
Hardware:	Other
OS:	All
Whiteboard:
i915 platform:		i915 features:
Attachments:	strace of the cryptsetup command that hangs "maximum number (20) of children reached" errors

Description Evangelos Foutras 2015-05-26 14:18:03 UTC

Created attachment 116046 [details]
strace of the cryptsetup command that hangs

As part of the process of decrypting the root volume during early boot, 'cryptsetup open --type luks /path/to/dev root' is executed while we're still in the initramfs and before switch_root is called.

Starting with systemd 220 this call hangs and boot never continues. Attached is a screen shot of the end of the strace output.

Git bisect shows the *parent* commit of https://github.com/systemd/systemd/commit/cb49a4f2dd53 as the last working revision. While testing commit cb49a4f2dd53, 'udevadm settle' would hang, though that's not the case with v220 or git master. I might check the range cb49a4f2dd53..master to see if there's a point where the both the 'udevadm settle' command and cryptsetup work.

This is possibly related to bug 90656.

Comment 1 Evangelos Foutras 2015-05-26 14:19:21 UTC

Forgot to mention this is on Arch Linux, running Linux 4.0.4 and cryptsetup 1.6.6.

Comment 2 Lennart Poettering 2015-05-26 14:23:33 UTC

Is this dracut or some other initrd?

Comment 3 Evangelos Foutras 2015-05-26 14:26:09 UTC

It's not dracut; the initramfs is generated using mkinitcpio which I believe is specific to Arch. [1]

[1] https://wiki.archlinux.org/index.php/Mkinitcpio

Comment 4 Evangelos Foutras 2015-05-26 17:00:18 UTC

(In reply to Evangelos Foutras from comment #0)
> Git bisect shows the *parent* commit of
> https://github.com/systemd/systemd/commit/cb49a4f2dd53 as the last working
> revision. While testing commit cb49a4f2dd53, 'udevadm settle' would hang,
> though that's not the case with v220 or git master. I might check the range
> cb49a4f2dd53..master to see if there's a point where the both the 'udevadm
> settle' command and cryptsetup work.

Actually, ignore this paragraph; I was removing the 'assert(manager->pid == getpid());' assertion from all my test builds (in src/udev/udevd.c). This might have messed with the results and possibly caused the 'udevadm settle' issue.

The cryptsetup hang is still valid though.

Comment 5 Evangelos Foutras 2015-05-27 03:21:10 UTC

Created attachment 116064 [details]
"maximum number (20) of children reached" errors

I ran systemd-udevd with --debug (in the initramfs) and a lot of "maximum number (20) of children reached" errors get repeated after I execute 'cryptsetup open'.

Comment 6 Evangelos Foutras 2015-05-27 07:28:59 UTC

I did some more careful bisecting and the commit that introduced this bug is:

  commit e237d8cb0e5488eef311ca1eafe48e8a2248b04c
  Author: Tom Gundersen <teg@jklm.no>
  Date:   Tue May 12 21:16:47 2015 +0200

      udevd: move file descriptors to Manager

      No functional change.

Link: https://github.com/systemd/systemd/commit/e237d8cb0e54

I haven't looked deeper into this, but it appears that workers aren't terminated properly (or maybe the manager loses track of them).

Comment 7 Tom Gundersen 2015-05-27 17:08:46 UTC

This should be fixed by 86c3bece38bcf55da6387d20c6f01da9ad0284dc.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.