smbd blocks for 120 secs and then hangs in D state

[Jeff Layton]

On Mon, 23 Aug 2010 10:40:34 +0530
Vijai Baskar <cristalmaze at gmail.com> wrote:

> Hi all,
> 
> I am running a samba copy of a 4GB file from my local hard disk to a
> remote hard disk and then find md5sum of the same file in the remote
> hard disk through samba read. I do the following to accomplish this:
> 
> 1. mount -t cifs //ip-addr/data /mnt
> 2. cp /home/4g /mnt
> 3. md5sum /mnt/4g
> 
> After a few iterations of the above operation I get the following
> warning on the board:
> 
> echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
> [ 6122.450000] smbd          D c0204908     0  9712    372 0x00000000
> [ 6122.480000] [<c0204908>] (schedule+0x2dc/0x328) from [<c0204980>]
> (io_schedule+0x2c/0x48)
> [ 6122.500000] [<c0204980>] (io_schedule+0x2c/0x48) from [<c0072064>]
> (sync_page+0x44/0x50)
> [ 6122.530000] [<c0072064>] (sync_page+0x44/0x50) from [<c0204dd4>]
> (__wait_on_bit_lock+0x5c/0xa8)
> [ 6122.560000] [<c0204dd4>] (__wait_on_bit_lock+0x5c/0xa8) from
> [<c0071ff0>] (__lock_page+0x88/0xa0)
> [ 6122.590000] [<c0071ff0>] (__lock_page+0x88/0xa0) from [<c007b530>]
> (truncate_inode_pages_range+0x2e4/0x38c)
> [ 6122.620000] [<c007b530>] (truncate_inode_pages_range+0x2e4/0x38c)
> from [<c007b5f0>] (truncate_inode_pages+0x18/0x20)
> [ 6122.640000] [<c007b5f0>] (truncate_inode_pages+0x18/0x20) from
> [<c0085acc>] (vmtruncate+0xe4/0x14c)
> [ 6122.670000] [<c0085acc>] (vmtruncate+0xe4/0x14c) from [<c00a43a8>]
> (inode_setattr+0x48/0x148)
> [ 6122.700000] [<c00a43a8>] (inode_setattr+0x48/0x148) from
> [<c00a4630>] (notify_change+0x188/0x1dc)
> [ 6122.730000] [<c00a4630>] (notify_change+0x188/0x1dc) from
> [<c0091440>] (do_truncate+0x6c/0x88)
> [ 6122.760000] [<c0091440>] (do_truncate+0x6c/0x88) from [<c00915c0>]
> (do_sys_ftruncate+0x164/0x170)
> [ 6122.800000] [<c00915c0>] (do_sys_ftruncate+0x164/0x170) from
> [<c00915e0>] (sys_ftruncate64+0x14/0x1c)
> [ 6122.830000] [<c00915e0>] (sys_ftruncate64+0x14/0x1c) from
> [<c0027fa0>] (ret_fast_syscall+0x0/0x2c)
> 
> After this samba daemon hangs in the D state. I changed the values of
> /proc/sys/vm/dirty_writeback_centisecs to 250 (default 500) and
> /proc/sys/vm/dirty_expire_centisecs to 1000 (default 3000). But this
> only delays the problem. ps-ax shows smbd in D state. This problem
> seems to occur only during samba read.
> 
> Can someone please provide me a solution for the above problem?
> 
> regards,
> vijai
> 

(cc'ing linux-fsdevel and samba-technical as this problem is probably
better reported there)

The stack trace above is for smbd, so I doubt this has much to do with
cifs per-se. Just to make sure though -- are the server and client the
same host? Hint: if so, that configuration is prone to deadlock under
heavy I/O.

When reporting kernel bugs, it's also a good idea to mention the kernel
version. It might also be helpful to know what the underlying
filesystem is that's being served out.

It looks like the process is hung trying to lock a page. Most likely
that means that something else is holding that lock and not releasing
it for some reason. Debugging this will probably mean figuring out
what's holding that lock and why it's not releasing it.

-- 
Jeff Layton <jlayton at redhat.com>