Re: [PATCH v6] IB/mlx4: Fix refcount leak in add_port() error path
From: Guangshuo Li
Date: Sun May 17 2026 - 22:19:58 EST
Hi Leon,
Thanks for reviewing.
On Sun, 17 May 2026 at 22:27, Leon Romanovsky <leon@xxxxxxxxxx> wrote:
>
> On Thu, May 14, 2026 at 07:01:39PM +0800, Guangshuo Li wrote:
> > After kobject_init_and_add(), the lifetime of the embedded struct
> > kobject is expected to be managed through the kobject core reference
> > counting.
> >
> > In add_port(), several failure paths after kobject_init_and_add() free
> > struct mlx4_port directly instead of releasing the embedded kobject with
> > kobject_put(). This leaves the kobject reference count unbalanced and can
> > lead to incorrect lifetime handling.
> >
> > Allocate the pkey and gid attribute arrays before kobject_init_and_add(),
> > so failures before kobject initialization can be handled by directly
> > freeing the allocated memory. Once kobject_init_and_add() has been
> > called, route failures through kobject_put(), and call kobject_del()
> > before kobject_put() on later failure paths after the kobject has been
> > successfully added.
> >
> > Fixes: c1e7e466120b ("IB/mlx4: Add iov directory in sysfs under the ib device")
> > Signed-off-by: Guangshuo Li <lgs201920130244@xxxxxxxxx>
> > ---
> > v6:
> > - drop the Cc stable tag
> > - allocate pkey and gid attribute arrays before kobject_init_and_add()
> > - keep the release callback unchanged by ensuring the attribute arrays
> > are initialized before kobject_init_and_add()
> >
> > v5:
> > - split the add_port() error paths after kobject_init_and_add()
> > - call kobject_del() before kobject_put() for failures after
> > kobject_init_and_add() succeeds
> >
> > v4:
> > - route all add_port() failures after kobject_init_and_add() through
> > a single kobject_put() based error path
> > - remove duplicated attribute array frees from add_port()
> > - keep mlx4_port_release() tolerant of partially initialized objects
> >
> > v3:
> > - make mlx4_port_release() tolerate NULL attribute arrays
> > - drop the parent kobject reference on the kobject_init_and_add()
> > failure path before putting the embedded kobject
> >
> > v2:
> > - note that the issue was identified by my static analysis tool
> > - and confirmed by manual review
> >
> > drivers/infiniband/hw/mlx4/sysfs.c | 39 ++++++++++++++++--------------
> > 1 file changed, 21 insertions(+), 18 deletions(-)
> >
> > diff --git a/drivers/infiniband/hw/mlx4/sysfs.c b/drivers/infiniband/hw/mlx4/sysfs.c
> > index b8fa4ecfc961..e4c822c96ee6 100644
> > --- a/drivers/infiniband/hw/mlx4/sysfs.c
> > +++ b/drivers/infiniband/hw/mlx4/sysfs.c
> > @@ -636,12 +636,6 @@ static int add_port(struct mlx4_ib_dev *dev, int port_num, int slave)
> > p->port_num = port_num;
> > p->slave = slave;
> >
> > - ret = kobject_init_and_add(&p->kobj, &port_type,
> > - kobject_get(dev->dev_ports_parent[slave]),
> > - "%d", port_num);
> > - if (ret)
> > - goto err_alloc;
> > -
> > p->pkey_group.name = "pkey_idx";
> > p->pkey_group.attrs =
> > alloc_group_attrs(show_port_pkey,
> > @@ -649,13 +643,9 @@ static int add_port(struct mlx4_ib_dev *dev, int port_num, int slave)
> > dev->dev->caps.pkey_table_len[port_num]);
> > if (!p->pkey_group.attrs) {
> > ret = -ENOMEM;
> > - goto err_alloc;
> > + goto err_free_port;
> > }
> >
> > - ret = sysfs_create_group(&p->kobj, &p->pkey_group);
> > - if (ret)
> > - goto err_free_pkey;
> > -
> > p->gid_group.name = "gid_idx";
> > p->gid_group.attrs = alloc_group_attrs(show_port_gid_idx, NULL, 1);
> > if (!p->gid_group.attrs) {
> > @@ -663,28 +653,41 @@ static int add_port(struct mlx4_ib_dev *dev, int port_num, int slave)
> > goto err_free_pkey;
> > }
> >
> > + ret = kobject_init_and_add(&p->kobj, &port_type,
> > + kobject_get(dev->dev_ports_parent[slave]),
> > + "%d", port_num);
> > + if (ret)
> > + goto err_put;
> > +
> > + ret = sysfs_create_group(&p->kobj, &p->pkey_group);
> > + if (ret)
> > + goto err_del;
> > +
> > ret = sysfs_create_group(&p->kobj, &p->gid_group);
> > if (ret)
> > - goto err_free_gid;
> > + goto err_del;
>
> You should call to sysfs_remove_group() too.
>
> Thanks
>
I added sysfs_remove_group() for the successfully created
pkey/gid groups before kobject_del() and sent v7.