Re: 2.6.23-rc7-mm1

[Reuben Farrelly <reuben-linuxkernel@reub.net>]

On 25/09/2007 3:12 AM, J. Bruce Fields wrote:
> On Mon, Sep 24, 2007 at 09:59:29AM -0700, Andrew Morton wrote:
>> On Tue, 25 Sep 2007 00:52:30 +1000 Reuben Farrelly <reuben-linuxkernel@reub.net> wrote:
>>
>>>
>>> On 24/09/2007 7:17 PM, Andrew Morton wrote:
>>>> ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.23-rc7/2.6.23-rc7-mm1/
>>>>
>>>> - New git tree git-powerpc-galak.patch added to the -mm lineup: ppc32
>>>>   things, mainly (Kumar Gala <galak@gate.crashing.org>)
>>> I'm observing a problem with this kernel (as well as 2.6.23-rc6-mm1) which 
>>> manifests itself only in my Postfix/application mail.logs:
>>>
>>> Sep 25 00:25:40 tornado postfix/smtp[12520]: fatal: select lock: Cannot allocate 
>>> memory
>>> Sep 25 00:25:41 tornado postfix/master[8002]: warning: process 
>>> /usr/lib64/postfix/smtp pid 12520 exit status 1
>>>
>>> This is happening frequently with processes started via 'master' (smtp, smtpd 
>>> and cleanup), but it does not appear to have any noticeable operational impact 
>>> apart from logging a lot of copies of this message.
>>>
>>> The corresponding code in Postfix which triggers this is (choice of 3 files in 
>>> src/master are all possibilities which all have much the same code)
> 
> Oog.  Looks like it's the "Memory shortage can result in inconsistent
> flocks state" patch--the error variable is being set in some cases when
> it shouldn't be.  Does the following fix it?
> 
> That's in my git tree, not in mainline.  I'll fix up my copy.
> 
> And I'll spend some time today figuring out what to do about regression
> testing for the posix lock, flock, and lease code.
> 
> Thanks for the bug report!
> 
> --b.
> 
> diff --git a/fs/locks.c b/fs/locks.c
> index a6c5917..3e8bfd2 100644
> --- a/fs/locks.c
> +++ b/fs/locks.c
> @@ -740,6 +740,7 @@ static int flock_lock_file(struct file *filp, struct file_lock *request)
>  		new_fl = locks_alloc_lock();
>  		if (new_fl == NULL)
>  			goto out;
> +		error = 0;
>  	}
>  
>  	for_each_lock(inode, before) {

Yes that has fixed it, thanks!

Reuben