Hi Tetsuo, On Wed, Apr 13, 2022 at 12:18 AM Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx> wrote: > > On 2022/04/12 22:46, Tetsuo Handa wrote: > > Flushing system-wide workqueues is dangerous and will be forbidden. > > > > Since system_long_wq is used only inside rnbd_destroy_sessions(), > > let's use list_head than creating a local workqueue for tracking > > work_struct to flush. > > > > Link: https://lkml.kernel.org/r/49925af7-78a8-a3dd-bce6-cfc02e1a9236@xxxxxxxxxxxxxxxxxxx > > > > Signed-off-by: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx> > > --- > > Notice: This patch is only compile tested. Please test before applying. > > > > drivers/block/rnbd/rnbd-clt.c | 5 ++++- > > drivers/block/rnbd/rnbd-clt.h | 1 + > > 2 files changed, 5 insertions(+), 1 deletion(-) > > > > diff --git a/drivers/block/rnbd/rnbd-clt.c b/drivers/block/rnbd/rnbd-clt.c > > index b66e8840b94b..b14e7c15133e 100644 > > --- a/drivers/block/rnbd/rnbd-clt.c > > +++ b/drivers/block/rnbd/rnbd-clt.c > > @@ -1730,6 +1730,7 @@ static void rnbd_destroy_sessions(void) > > { > > struct rnbd_clt_session *sess, *sn; > > struct rnbd_clt_dev *dev, *tn; > > + LIST_HEAD(list); > > > > /* Firstly forbid access through sysfs interface */ > > rnbd_clt_destroy_sysfs_files(); > > @@ -1762,11 +1763,13 @@ static void rnbd_destroy_sessions(void) > > */ > > INIT_WORK(&dev->unmap_on_rmmod_work, unmap_device_work); > > queue_work(system_long_wq, &dev->unmap_on_rmmod_work); > > + list_add_tail(&dev->unmap_on_rmmod_list, &list); > > } > > rnbd_clt_put_sess(sess); > > } > > /* Wait for all scheduled unmap works */ > > - flush_workqueue(system_long_wq); > > + list_for_each_entry(dev, &list, unmap_on_rmmod_list) > > + flush_work(&dev->unmap_on_rmmod_work); > > Since kfree(dev) might be called from unmap_device_work(), this seems unsafe. > We need rnbd_clt_get_dev() before queue_work() and rnbd_clt_put_dev() after flush_work() > in order to make dev->unmap_on_rmmod_list traversal safe... you are right, unmap_device_work could free the dev structure, lead to UAF. simply add rnbd_clt_get_dev and put_dev will lead to following WARN_ON to be triggered. > > > WARN_ON(!list_empty(&sess_list)); > > } I feel the simple approach is to replace the gloable system_long_wq with a local workqueue for rnbd_client. Thanks! > > > > diff --git a/drivers/block/rnbd/rnbd-clt.h b/drivers/block/rnbd/rnbd-clt.h > > index 2e2e8c4a85c1..a6d704abda61 100644 > > --- a/drivers/block/rnbd/rnbd-clt.h > > +++ b/drivers/block/rnbd/rnbd-clt.h > > @@ -136,6 +136,7 @@ struct rnbd_clt_dev { > > char *blk_symlink_name; > > refcount_t refcount; > > struct work_struct unmap_on_rmmod_work; > > + struct list_head unmap_on_rmmod_list; > > }; > > > > /* rnbd-clt.c */ >
