Bug 451738

Summary:	kcryptd I/O blockage using encrypted filesystem on LVM
Product:	[Fedora] Fedora	Reporter:	Michael Hampton <error>
Component:	kernel	Assignee:	Milan Broz <mbroz>
Status:	CLOSED RAWHIDE	QA Contact:	Fedora Extras Quality Assurance <extras-qa>
Severity:	medium	Docs Contact:
Priority:	low
Version:	9	CC:	pvrabec
Target Milestone:	---
Target Release:	---
Hardware:	x86_64
OS:	Linux
Whiteboard:
Fixed In Version:		Doc Type:	Bug Fix
Doc Text:		Story Points:	---
Clone Of:		Environment:
Last Closed:	2008-08-26 11:13:14 UTC	Type:	---
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:

Description Michael Hampton 2008-06-17 01:34:01 UTC

Description of problem:
When using an encrypted filesystem on LVM, whether an encrypted PV or LV, the
kernel frequently blocks for long periods of time waiting for I/O. top shows
anywhere from 20% to 100% CPU time waiting for I/O. Affected processes freeze
for several seconds, up to a minute or more, waiting for even small disk
reads/writes.

Version-Release number of selected component (if applicable):
kernel-2.6.25.6-55.fc9.x86_64

How reproducible:
Random

Steps to Reproduce:
1. Install Fedora 9 to encrypted PV.
2. Use moderately disk intensive processes such as firefox, thunderbird or liferea.
  
Actual results:
Some disk operations take dozens of seconds to complete.

Expected results:
Disk operations should complete in a reasonable time.

Additional info:
Issue did not occur with Fedora 8. Help tracking this down would be greatly
appreciated.

Comment 1 Dave Jones 2008-06-17 01:45:54 UTC

this is likely another case of the 'firefox fsyncs too much' bug.  (it calls
fsync every time it updates the history db - ie, each time you view a new page.
unfortunatly fsync on ext3 means 'write everything in memory out to disk, not
just the data from this process'.

Comment 2 Milan Broz 2008-07-04 16:32:29 UTC

There is commit in 2.6.26 kernel (adding cond_resched()) in dmcrypt code, which
should help in this situation.
(System still need wait for io to finish, but other running application should
be more responsible.)

Comment 3 Milan Broz 2008-08-26 11:13:14 UTC

cond_resched() call is in 2.6.26+ kernels, it should help in some situations.

There are still some situations when system will wait for IO - mainly if you have multiple LVs but only one encrypted PV...

closing this rahwide, there is already kernel with this patch.