[PATCH gfs2/for-next] gfs2: use delay during spinlock area

[PATCH gfs2/for-next] gfs2: use delay during spinlock area

[Alexander Aring]

In a rare case of gfs2 spectator mount the ls->ls_recover_spin is being
held. In this case we cannot call msleep_interruptible() as we a in a
non-sleepable context. Replace it with mdelay() to busy wait for 1
second.

Cc: stable@vger.kernel.org
Fixes: 4a7727725dc7 ("GFS2: Fix recovery issues for spectators")
Signed-off-by: Alexander Aring <aahringo@redhat.com>
---
 fs/gfs2/lock_dlm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/gfs2/lock_dlm.c b/fs/gfs2/lock_dlm.c
index 58aeeae7ed8c..ac0afedff49b 100644
--- a/fs/gfs2/lock_dlm.c
+++ b/fs/gfs2/lock_dlm.c
@@ -996,7 +996,7 @@ static int control_mount(struct gfs2_sbd *sdp)
 		if (sdp->sd_args.ar_spectator) {
 			fs_info(sdp, "Recovery is required. Waiting for a "
 				"non-spectator to mount.\n");
-			msleep_interruptible(1000);
+			mdelay(1000);
 		} else {
 			fs_info(sdp, "control_mount wait1 block %u start %u "
 				"mount %u lvb %u flags %lx\n", block_gen,
-- 
2.43.0

Re: [PATCH gfs2/for-next] gfs2: use delay during spinlock area

[David Laight]

On Mon, 31 Mar 2025 15:36:56 -0400
Alexander Aring <aahringo@redhat.com> wrote:

> In a rare case of gfs2 spectator mount the ls->ls_recover_spin is being
> held. In this case we cannot call msleep_interruptible() as we a in a
> non-sleepable context. Replace it with mdelay() to busy wait for 1
> second.

You can't busy wait like that.
You've just stopped any RT process that last ran on the cpu you are
on from running, as well as all any interrupts tied to the cpu.
Also consider a single cpu system.

	David

> 
> Cc: stable@vger.kernel.org
> Fixes: 4a7727725dc7 ("GFS2: Fix recovery issues for spectators")
> Signed-off-by: Alexander Aring <aahringo@redhat.com>
> ---
>  fs/gfs2/lock_dlm.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/gfs2/lock_dlm.c b/fs/gfs2/lock_dlm.c
> index 58aeeae7ed8c..ac0afedff49b 100644
> --- a/fs/gfs2/lock_dlm.c
> +++ b/fs/gfs2/lock_dlm.c
> @@ -996,7 +996,7 @@ static int control_mount(struct gfs2_sbd *sdp)
>  		if (sdp->sd_args.ar_spectator) {
>  			fs_info(sdp, "Recovery is required. Waiting for a "
>  				"non-spectator to mount.\n");
> -			msleep_interruptible(1000);
> +			mdelay(1000);
>  		} else {
>  			fs_info(sdp, "control_mount wait1 block %u start %u "
>  				"mount %u lvb %u flags %lx\n", block_gen,

Re: [PATCH gfs2/for-next] gfs2: use delay during spinlock area

[Andreas Gruenbacher]

On Mon, Mar 31, 2025 at 11:32 PM David Laight
<david.laight.linux@gmail.com> wrote:
> On Mon, 31 Mar 2025 15:36:56 -0400 Alexander Aring <aahringo@redhat.com> wrote:
>
> > In a rare case of gfs2 spectator mount the ls->ls_recover_spin is being
> > held. In this case we cannot call msleep_interruptible() as we a in a
> > non-sleepable context. Replace it with mdelay() to busy wait for 1
> > second.
>
> You can't busy wait like that.
> You've just stopped any RT process that last ran on the cpu you are
> on from running, as well as all any interrupts tied to the cpu.
> Also consider a single cpu system.

I agree. This is easy to fix by waiting after releasing the spin lock.

Thanks,
Andreas

>         David
>
> >
> > Cc: stable@vger.kernel.org
> > Fixes: 4a7727725dc7 ("GFS2: Fix recovery issues for spectators")
> > Signed-off-by: Alexander Aring <aahringo@redhat.com>
> > ---
> >  fs/gfs2/lock_dlm.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/fs/gfs2/lock_dlm.c b/fs/gfs2/lock_dlm.c
> > index 58aeeae7ed8c..ac0afedff49b 100644
> > --- a/fs/gfs2/lock_dlm.c
> > +++ b/fs/gfs2/lock_dlm.c
> > @@ -996,7 +996,7 @@ static int control_mount(struct gfs2_sbd *sdp)
> >               if (sdp->sd_args.ar_spectator) {
> >                       fs_info(sdp, "Recovery is required. Waiting for a "
> >                               "non-spectator to mount.\n");
> > -                     msleep_interruptible(1000);
> > +                     mdelay(1000);
> >               } else {
> >                       fs_info(sdp, "control_mount wait1 block %u start %u "
> >                               "mount %u lvb %u flags %lx\n", block_gen,
>