lock_kernel twice possible ?

lock_kernel twice possible ?

[li nux]

Hi,

I was going thru the NFS v3 code for SMP kernel 2.6.11
to see how an inode gets revalidated. I found that
there is a possibility that there may be an attempt to
do lock_kernel() twice.

Is this possible ? If yes then how this deadlock
condition is/can be avoided.

-lnxluv

Below is the code flow (please see ** for
lock_kernel):

nfs_revalidate_inode
 - __nfs_revalidate_inode
   - ** lock_kernel() **
   - nfs_wait_on_inode
   - Call getattr() (which is nfs3_proc_getattr()) to
     get the attributes from the server and 
     refresh the inode with the new values
   - IF the cached data is invalid for the inode
     - Writeback (If dirty) and sync the 
       inode, call nfs_wb_all
     - nfs_wb_all
       - nfs_sync_inode 
           - call nfs_wait_on_requests to wait for
             the requests associated with the pages
             to get complete
           - nfs_flush_inode
             - nfs_scan_dirty
             - nfs_flush_list
               - nfs_flush_one
                 - nfs_write_rpcsetup
                   - nfs3_proc_write_setup
                     - rpc_init_task
                     - rpc_call_setup
           - nfs_execute_write
             - ** lock_kernel() **
             - rpc_execute
             - ** unlock_kernel() **
   - ** unlock_kernel() **     




		
__________________________________ 
Yahoo! Music Unlimited 
Access over 1 million songs. Try it free.
http://music.yahoo.com/unlimited/

Re: lock_kernel twice possible ?

[Coywolf Qi Hunt]

On 10/15/05, li nux <lnxluv@yahoo.com> wrote:
> Hi,
>
> I was going thru the NFS v3 code for SMP kernel 2.6.11
> to see how an inode gets revalidated. I found that
> there is a possibility that there may be an attempt to
> do lock_kernel() twice.
>
> Is this possible ? If yes then how this deadlock
> condition is/can be avoided.

The BKL is recursive!
--
Coywolf Qi Hunt
http://sosdg.org/~coywolf/

Re: lock_kernel twice possible ?

[li nux]

Sorry, couldn't get what you want to say.
Can you please elaborate.

--- Coywolf Qi Hunt <coywolf@gmail.com> wrote:

> On 10/15/05, li nux <lnxluv@yahoo.com> wrote:
> > Hi,
> >
> > I was going thru the NFS v3 code for SMP kernel
> 2.6.11
> > to see how an inode gets revalidated. I found that
> > there is a possibility that there may be an
> attempt to
> > do lock_kernel() twice.
> >
> > Is this possible ? If yes then how this deadlock
> > condition is/can be avoided.
> 
> The BKL is recursive!
> --
> Coywolf Qi Hunt
> http://sosdg.org/~coywolf/
> 



		
__________________________________ 
Yahoo! Music Unlimited 
Access over 1 million songs. Try it free.
http://music.yahoo.com/unlimited/

Re: lock_kernel twice possible ?

[Coywolf Qi Hunt]

On 10/15/05, li nux <lnxluv@yahoo.com> wrote:
> Sorry, couldn't get what you want to say.
> Can you please elaborate.

http://www.zipworld.com.au/~akpm/linux/patches/stuff/top-posting.txt
http://www.catb.org/~esr/jargon/html/T/top-post.html
--
Coywolf Qi Hunt
http://sosdg.org/~coywolf/

Re: lock_kernel twice possible ?

[Bill Davidsen]

Coywolf Qi Hunt wrote:
> On 10/15/05, li nux <lnxluv@yahoo.com> wrote:
> 
>>Sorry, couldn't get what you want to say.
>>Can you please elaborate.
> 
> 
> http://www.zipworld.com.au/~akpm/linux/patches/stuff/top-posting.txt
> http://www.catb.org/~esr/jargon/html/T/top-post.html

The whole post fit on one screen and he was not continuing the 
conversation, he was commenting on your post as a whole. I think you're 
being a bit pedantic here.

I don't think his reply failed your criteria, it wasn't overlong, it 
didn't break a flowing discussion, and no backward reading was required.

~~~~~~~

Q:  Why is top-posting evil?

A1: Because top-posters tend to leave the entirety of the email to which
     they are replying intact, so the email gets too long.

A2: Because we prefer to use bottom-posting, and it is *extremely* hard to
     maintain a coherently flowing email conversation when some of the
     participants are bottom-posting and some are top-posting.

A3: backwards read don't humans because

-- 
    -bill davidsen (davidsen@tmr.com)
"The secret to procrastination is to put things off until the
  last possible moment - but no longer"  -me

Re: lock_kernel twice possible ?

[Steven Rostedt]

On Wed, 19 Oct 2005, Bill Davidsen wrote:

>
> ~~~~~~~
>
> Q:  Why is top-posting evil?
>
> A1: Because top-posters tend to leave the entirety of the email to which
>      they are replying intact, so the email gets too long.
>
> A2: Because we prefer to use bottom-posting, and it is *extremely* hard to
>      maintain a coherently flowing email conversation when some of the
>      participants are bottom-posting and some are top-posting.
>
> A3: backwards read don't humans because
>

A4: top-posting is for pointy haired managers.

Sorry! ;)

-- Steve

Re: lock_kernel twice possible ?

[Mitchell Blank Jr]

li nux wrote:
> Sorry, couldn't get what you want to say.
> Can you please elaborate.

A "recursive lock" is one that can be taken multiple times by the same
owner.  So:

	lock_kernel();
	lock_kernel();
	unlock_kernel();
	unlock_kernel();

is perfectly ok.  The lock_kernel() code detects that the calling thread
already owns the lock and just increments current->lock_depth.

Under linux locks are non-recursive EXCEPT for lock_kernel() (aka the BLK
or "big kernel lock")

See lib/kernel_lock.c for more details.

-Mitch