Bug 92531

Summary:	[BISECTED]soft lockup while removing directory trees under 3.2.66
Product:	File System	Reporter:	NUXI (nuxi)
Component:	VFS	Assignee:	fs_vfs
Status:	NEW ---
Severity:	normal	CC:	alan, ben, szg00000
Priority:	P1
Hardware:	IA-32
OS:	Linux
Kernel Version:	3.2.66	Subsystem:
Regression:	No	Bisected commit-id:
Attachments:	soft lockup trace vfs: Fix vfsmount_lock imbalance in path_init()

Description NUXI 2015-02-02 18:54:03 UTC

Created attachment 165621 [details]
soft lockup trace

If you accidently start multiple instance of "rm -rf" on the same large directory tree, the 3.2.x kernels starting with 3.2.64 seem to lockup and die while firing soft lockup warnings on the way down. See attachment for dmesg output from the event.

How hard this is seems to depend on system performance. I ended up bisecting it from an older piece of hardware just to make it easier to reproduce. So the system the backtrace is from is a 32-bit only system with ext3 filesystem on an adaptec raid controller. But I've also seen it on 32-bit systems running on LSI controllers, 32bit systems running under vmware fake LSI controllers, and once on a 64-bit sytems with an adaptec controller. (All of those other systems also ran ext3.)

I bisected it down to this commit:
https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/commit/?h=linux-3.2.y&id=1a4ba51a1d7def5f36b7dec2defd3c4ab27e9fb0

Comment 1 Ben Hutchings 2015-02-16 03:26:05 UTC

Created attachment 167001 [details]
vfs: Fix vfsmount_lock imbalance in path_init()

Does this patch fix the issue?

Comment 2 NUXI 2015-02-16 21:06:47 UTC

Yes, that seems to fix it. Thanks Ben.