[PATCH v2 0/3] mmc: sdio: fixes some leaks

[PATCH v2 0/3] mmc: sdio: fixes some leaks

[Yang Yingliang]

This patch a refcount leak and two memory leaks about
SDIO function.

v1 -> v2:
  Fix compile error in patch #2.

Yang Yingliang (3):
  mmc: sdio: fix of node refcount leak in sdio_add_func()
  mmc: sdio: fix possible memory leak in sdio_init_func()
  mmc: sdio: fix possible memory leak in mmc_attach_sdio()

 drivers/mmc/core/sdio.c     |  7 ++-----
 drivers/mmc/core/sdio_bus.c | 11 ++++++-----
 2 files changed, 8 insertions(+), 10 deletions(-)

-- 
2.25.1

[PATCH v2 1/3] mmc: sdio: fix of node refcount leak in sdio_add_func()

[Yang Yingliang]

If device_add() returns error in sdio_add_func(), sdio function is not
presented, so the node refcount that hold in sdio_set_of_node() can not
be put in sdio_remove_func() which is called from error path. Fix this
by calling of_node_put() if device_add() fails.

Fixes: 25185f3f31c9 ("mmc: Add SDIO function devicetree subnode parsing")
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
---
 drivers/mmc/core/sdio_bus.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/mmc/core/sdio_bus.c b/drivers/mmc/core/sdio_bus.c
index babf21a0adeb..b9308813a226 100644
--- a/drivers/mmc/core/sdio_bus.c
+++ b/drivers/mmc/core/sdio_bus.c
@@ -365,6 +365,8 @@ int sdio_add_func(struct sdio_func *func)
 	ret = device_add(&func->dev);
 	if (ret == 0)
 		sdio_func_set_present(func);
+	else
+		of_node_put(func->dev.of_node);
 
 	return ret;
 }
-- 
2.25.1

Re: [PATCH v2 1/3] mmc: sdio: fix of node refcount leak in sdio_add_func()

[Ulf Hansson]

On Wed, 2 Nov 2022 at 02:27, Yang Yingliang <yangyingliang@huawei.com> wrote:
>
> If device_add() returns error in sdio_add_func(), sdio function is not
> presented, so the node refcount that hold in sdio_set_of_node() can not
> be put in sdio_remove_func() which is called from error path. Fix this
> by calling of_node_put() if device_add() fails.
>
> Fixes: 25185f3f31c9 ("mmc: Add SDIO function devicetree subnode parsing")
> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
> ---
>  drivers/mmc/core/sdio_bus.c | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/drivers/mmc/core/sdio_bus.c b/drivers/mmc/core/sdio_bus.c
> index babf21a0adeb..b9308813a226 100644
> --- a/drivers/mmc/core/sdio_bus.c
> +++ b/drivers/mmc/core/sdio_bus.c
> @@ -365,6 +365,8 @@ int sdio_add_func(struct sdio_func *func)
>         ret = device_add(&func->dev);
>         if (ret == 0)
>                 sdio_func_set_present(func);
> +       else
> +               of_node_put(func->dev.of_node);

I think it would be better to change the sdio_remove_func() to manage
this. In principle sdio_func_present() should not make
sdio_remove_func() to return immediately, but rather just control
whether device_del() needs to be called or not.

>
>         return ret;
>  }


Kind regards
Uffe

[PATCH v2 2/3] mmc: sdio: fix possible memory leak in sdio_init_func()

[Yang Yingliang]

If it fails in sdio_init_func(), sdio_remove_func() can not
free the memory that allocated in sdio_alloc_func(), because
sdio_add_func() is not called yet, the sdio function is not
presented and sdio_remove_func() will return directly.

In this error path, we can not call put_device() to free the
memory in sdio_release_func(), because sdio_read_func_cis()
may fail, then sdio_free_func_cis() is called in release()
funtion which could cause put the reference that has not
been got.

So fix these leaks with calling kfree() instead of sdio_remove_func()
in error path.

Fixes: 3d10a1ba0d37 ("sdio: fix reference counting in sdio_remove_func()")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
---
 drivers/mmc/core/sdio.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/drivers/mmc/core/sdio.c b/drivers/mmc/core/sdio.c
index f64b9ac76a5c..f314224b362b 100644
--- a/drivers/mmc/core/sdio.c
+++ b/drivers/mmc/core/sdio.c
@@ -133,11 +133,8 @@ static int sdio_init_func(struct mmc_card *card, unsigned int fn)
 	return 0;
 
 fail:
-	/*
-	 * It is okay to remove the function here even though we hold
-	 * the host lock as we haven't registered the device yet.
-	 */
-	sdio_remove_func(func);
+	kfree(func->tmpbuf);
+	kfree(func);
 	return ret;
 }
 
-- 
2.25.1

[PATCH v2 3/3] mmc: sdio: fix possible memory leak in mmc_attach_sdio()

[Yang Yingliang]

If sdio_add_func() returns error in mmc_attach_sdio(),
sdio_remove_func() can not free the memory that allocated
in sdio_init_func(), because the sdio function is not
presented and sdio_remove_func() will return directly.

To fix these leaks, we can call put_device() to give up
the reference which was set in device_initialize(), then
the memory can be freed in sdio_release_func().

Fixes: 3d10a1ba0d37 ("sdio: fix reference counting in sdio_remove_func()")
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
---
 drivers/mmc/core/sdio_bus.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/drivers/mmc/core/sdio_bus.c b/drivers/mmc/core/sdio_bus.c
index b9308813a226..c4d3f721567e 100644
--- a/drivers/mmc/core/sdio_bus.c
+++ b/drivers/mmc/core/sdio_bus.c
@@ -379,11 +379,10 @@ int sdio_add_func(struct sdio_func *func)
  */
 void sdio_remove_func(struct sdio_func *func)
 {
-	if (!sdio_func_present(func))
-		return;
-
-	device_del(&func->dev);
-	of_node_put(func->dev.of_node);
+	if (sdio_func_present(func)) {
+		device_del(&func->dev);
+		of_node_put(func->dev.of_node);
+	}
 	put_device(&func->dev);
 }
 
-- 
2.25.1