[OpenAFS] heavy troubles with openafs 1.3.79 and kernel 2.6.11

Guillaume Rousse Guillaume.Rousse@inria.fr
Fri, 18 Mar 2005 16:32:56 +0100


chas williams - CONTRACTOR wrote:
> In message <4239A851.6070608@inria.fr>,Guillaume Rousse writes:
> 
>>EIP is at __inode_queue_event+0x18/0x60
>>eax: f33c1000   ebx: fffffff8   ecx: f33c1000   edx: 00000000
>>esi: 00000004   edi: 00000020   ebp: e8d3bf8c   esp: e8d3bf74
>>ds: 007b   es: 007b   ss: 0068
>>Process ls (pid: 7025, threadinfo=e8d3a000 task=ea79c530)
> 
> 
> this is pretty telling.  __inode_queue_event doesnt exist in any
> of the 'standard' kernels so it seems like a new thing.  a google
> search shows one hit from the lkml:
> 
> http://lkml.org/lkml/2005/3/8/187
> 
> it seems to add some members to the inode struct, a list and a spinlock.
> both of these are going to need to be inited since the auther apparenty
> didnt put them in init_inode_once().  the changes openafs needs would be
> similar to any of the ones in the past.  detect that member in at  
> configure time and initialize it in afs_vcache.c -- see the list 
> archives about how i_sb_list was handled.
I'm far from a kernel guru, but if it is really a kernel structure 
initialisation problem, wouldn't a kernel patch fix it for everyone, 
instead of having all fs implementations to get changed so as to take 
care of it ?
-- 
To study an application best, understand it thoroughly before you start
		-- Educational Considerations n°1