[OpenAFS] console messages: "all buffers locked"

Simon Wilkinson sxw@inf.ed.ac.uk
Mon, 26 Oct 2009 13:57:39 +0000

On 26 Oct 2009, at 09:34, Rainer Toebbicke wrote:

> On several nodes I got plenty of "all buffers locked" messages on  
> the console. In certain cases subshells of that shell-script segfault.
> I haven't yet had time to dig deeper into this, after finding that  
> the message comes from afs_buffer.c (and deserves an "afs:" and a  
> "\n"!) and google for once "not being my friend" I thought I'd ask  
> first. Has anybody run into this?

I've submitted gerrit 732 to address this part of your problem (minor  
cosmetics, but hopefully
it will make things easier for anyone encountering this in the future)

> (slabinfo shows e.g. 1579993 for afs_inode_cache. Is that big? About  
> same size as dentry_cache.
> My plan is to try add a traceback to afs_buffer.c, and eventually  
> try with 1.4.8+security patch which we still run on a big scale  
> here. Any other ideas?)

This suggests that something's locking, but not release buffers in the  
directory handling code. Are the directories you're writing these  
files into particularly large?