On Mon, Nov 08, 2010 at 12:32:00PM -0800, Mike Waychison wrote:
> The netconsole driver currently deadlocks if a NETDEV_UNREGISTER event
> is received while netconsole is in use, which in turn causes it to pin a
> reference to the network device. The first deadlock was dealt with in
> 3b410a31 so that we wouldn't recursively grab RTNL, but even calling
> __netpoll_cleanup isn't safe to do considering that we are in atomic
> context. __netpoll_cleanup assumes it can sleep and has several
> sleeping calls, such as synchronize_rcu_bh and
> cancel_rearming_delayed_work.
>
> Fix this by deferring netpoll_cleanup using scheduling work that
> operates in process context. We have to grab a reference to the
> config_item in this case as we need to pin the item in place until it is
> operated on.
>
> Signed-off-by: Mike Waychison <mikew@xxxxxxxxxx>
> ---
> drivers/net/netconsole.c | 55 ++++++++++++++++++++++++++++++++++++++++------
> 1 files changed, 48 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/net/netconsole.c b/drivers/net/netconsole.c
> index 288a025..02ba5c4 100644
> --- a/drivers/net/netconsole.c
> +++ b/drivers/net/netconsole.c
> @@ -106,6 +106,7 @@ struct netconsole_target {
> #endif
> int np_state;
> struct netpoll np;
> + struct work_struct cleanup_work;
> };
>
> #ifdef CONFIG_NETCONSOLE_DYNAMIC
> @@ -166,6 +167,22 @@ static void netconsole_target_put(struct netconsole_target *nt)
>
> #endif /* CONFIG_NETCONSOLE_DYNAMIC */
>
> +static void deferred_netpoll_cleanup(struct work_struct *work)
> +{
> + struct netconsole_target *nt;
> + unsigned long flags;
> +
> + nt = container_of(work, struct netconsole_target, cleanup_work);
> + netpoll_cleanup(&nt->np);
> +
> + spin_lock_irqsave(&target_list_lock, flags);
> + BUG_ON(nt->np_state != NETPOLL_CLEANING);
> + nt->np_state = NETPOLL_DISABLED;
> + spin_unlock_irqrestore(&target_list_lock, flags);
> +
> + netconsole_target_put(nt);
> +}
> +
Where is the synchronization on the new work queue when the module is getting
removed? The target get/put code does nothing to the module refcount, and
cleanup_netconsole just deletes targets, it doesn't block or fail on netconsole
refcounts, so you could run this work after the module has been removed and oops
the system.
Neil
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
