[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <aOi3u82R9pLLdr4I@osx.local>
Date: Fri, 10 Oct 2025 15:37:31 +0800
From: Shuhao Fu <sfual@....ust.hk>
To: Leon Romanovsky <leon@...nel.org>
Cc: Jason Gunthorpe <jgg@...dia.com>, linux-rdma@...r.kernel.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH] RDMA/uverbs: fix umem release in UVERBS_METHOD_CQ_CREATE
On Thu, Oct 09, 2025 at 09:48:54PM +0300, Leon Romanovsky wrote:
> On Fri, Oct 10, 2025 at 12:40:26AM +0800, Shuhao Fu wrote:
> > In `UVERBS_METHOD_CQ_CREATE`, umem should be released if anything goes
> > wrong. Currently, if `create_cq_umem` fails, umem would not be
> > released or referenced, causing a possible leak.
>
> It is only partially true. UMEM is released inside .create_cq_umem()
> implementation. However there is an issue there that it doesn't release
> in all flows. You need to change efa_create_cq_umem() too.
>
My apologies for overlooking the umem release in `.create_cq_umem`. I have
sent out patch v2 [1] for this issue. In v2, the driver should not release
umem on failure, avoiding possible double free here.
[1] https://lore.kernel.org/linux-rdma/aOh1le4YqtYwj-hH@osx.local/
> >
> > Fixes: 1a40c362ae26 ("RDMA/uverbs: Add a common way to create CQ with umem")
> > Signed-off-by: Shuhao Fu <sfual@....ust.hk>
> > ---
> > drivers/infiniband/core/uverbs_std_types_cq.c | 4 +++-
> > 1 file changed, 3 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/infiniband/core/uverbs_std_types_cq.c b/drivers/infiniband/core/uverbs_std_types_cq.c
> > index 37cd37556510..9344020dede1 100644
> > --- a/drivers/infiniband/core/uverbs_std_types_cq.c
> > +++ b/drivers/infiniband/core/uverbs_std_types_cq.c
> > @@ -193,8 +193,10 @@ static int UVERBS_HANDLER(UVERBS_METHOD_CQ_CREATE)(
> >
> > ret = umem ? ib_dev->ops.create_cq_umem(cq, &attr, umem, attrs) :
> > ib_dev->ops.create_cq(cq, &attr, attrs);
> > - if (ret)
> > + if (ret) {
> > + ib_umem_release(umem);
> > goto err_free;
> > + }
> >
> > obj->uevent.uobject.object = cq;
> > obj->uevent.uobject.user_handle = user_handle;
> > --
> > 2.39.5
> >
> >
Powered by blists - more mailing lists