blk_insert_bs() requires that the caller holds the AioContext lock for
the node to be inserted. Since commit c066e808e11, neglecting to do so
causes a crash when the child has to be moved to a different AioContext
to attach it to the BlockBackend.

This fixes qmp_blockdev_insert_anon_medium(), which is called for the
QMP commands 'blockdev-insert-medium' and 'blockdev-change-medium', to
correctly take the lock.

Cc: qemu-stable@nongnu.org
Fixes: https://issues.redhat.com/browse/RHEL-3922
Fixes: c066e808e11a5c181b625537b6c78e0de27a4801
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
 block/qapi-sysemu.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/block/qapi-sysemu.c b/block/qapi-sysemu.c
index 3f614cbc04..1618cd225a 100644
--- a/block/qapi-sysemu.c
+++ b/block/qapi-sysemu.c
@@ -237,6 +237,7 @@ static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
                                             BlockDriverState *bs, Error **errp)
 {
     Error *local_err = NULL;
+    AioContext *ctx;
     bool has_device;
     int ret;
 
@@ -258,7 +259,11 @@ static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
         return;
     }
 
+    ctx = bdrv_get_aio_context(bs);
+    aio_context_acquire(ctx);
     ret = blk_insert_bs(blk, bs, errp);
+    aio_context_release(ctx);
+
     if (ret < 0) {
         return;
     }
-- 
2.41.0

On 13.10.23 17:33, Kevin Wolf wrote:
> blk_insert_bs() requires that the caller holds the AioContext lock for
> the node to be inserted. Since commit c066e808e11, neglecting to do so
> causes a crash when the child has to be moved to a different AioContext
> to attach it to the BlockBackend.
>
> This fixes qmp_blockdev_insert_anon_medium(), which is called for the
> QMP commands 'blockdev-insert-medium' and 'blockdev-change-medium', to
> correctly take the lock.
>
> Cc: qemu-stable@nongnu.org
> Fixes: https://issues.redhat.com/browse/RHEL-3922
> Fixes: c066e808e11a5c181b625537b6c78e0de27a4801
> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
> ---
>   block/qapi-sysemu.c | 5 +++++
>   1 file changed, 5 insertions(+)

Do we need to take the lock for the dev_ops tray callbacks, too?  I 
suppose not, and it also wouldn’t really matter in light of the lock 
being supposed to go away anyway, but still thought I should ask.

In any case, this change here is necessary, so:

Reviewed-by: Hanna Czenczek <hreitz@redhat.com>

Am 31.10.2023 um 12:54 hat Hanna Czenczek geschrieben:
> On 13.10.23 17:33, Kevin Wolf wrote:
> > blk_insert_bs() requires that the caller holds the AioContext lock for
> > the node to be inserted. Since commit c066e808e11, neglecting to do so
> > causes a crash when the child has to be moved to a different AioContext
> > to attach it to the BlockBackend.
> > 
> > This fixes qmp_blockdev_insert_anon_medium(), which is called for the
> > QMP commands 'blockdev-insert-medium' and 'blockdev-change-medium', to
> > correctly take the lock.
> > 
> > Cc: qemu-stable@nongnu.org
> > Fixes: https://issues.redhat.com/browse/RHEL-3922
> > Fixes: c066e808e11a5c181b625537b6c78e0de27a4801
> > Signed-off-by: Kevin Wolf <kwolf@redhat.com>
> > ---
> >   block/qapi-sysemu.c | 5 +++++
> >   1 file changed, 5 insertions(+)
> 
> Do we need to take the lock for the dev_ops tray callbacks, too?  I suppose
> not, and it also wouldn’t really matter in light of the lock being supposed
> to go away anyway, but still thought I should ask.

Seems nobody ever bothered to define what the callbacks expects, and I
don't know either. Not taking the lock can obviously be a problem, but
taking it can also be a problem if the callback then locks a second time
and calls a synchronous function that polls.

What I do see is that callers disagree about this, so no matter what the
correct answer is, I'm almost sure there is a bug hiding somewhere.

Kevin