In imx_pd_bind, the duplicated memory for imxpd->edid via kmemdup should
be released in drm_of_find_panel_or_bridge or imx_pd_register fail.
Fixes: ebc944613567 ("drm: convert drivers to use drm_of_find_panel_or_bridge")
Fixes: 19022aaae677 ("staging: drm/imx: Add parallel display support")
Signed-off-by: Navid Emamdoost <[email protected]>
---
drivers/gpu/drm/imx/parallel-display.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/imx/parallel-display.c b/drivers/gpu/drm/imx/parallel-display.c
index e7ce17503ae1..9522d2cb0ad5 100644
--- a/drivers/gpu/drm/imx/parallel-display.c
+++ b/drivers/gpu/drm/imx/parallel-display.c
@@ -227,14 +227,18 @@ static int imx_pd_bind(struct device *dev, struct device *master, void *data)
/* port@1 is the output port */
ret = drm_of_find_panel_or_bridge(np, 1, 0, &imxpd->panel, &imxpd->bridge);
- if (ret && ret != -ENODEV)
+ if (ret && ret != -ENODEV) {
+ kfree(imxpd->edid);
return ret;
+ }
imxpd->dev = dev;
ret = imx_pd_register(drm, imxpd);
- if (ret)
+ if (ret) {
+ kfree(imxpd->edid);
return ret;
+ }
dev_set_drvdata(dev, imxpd);
--
2.17.1
> In imx_pd_bind, the duplicated memory for imxpd->edid via kmemdup should
> be released in drm_of_find_panel_or_bridge or imx_pd_register fail.
Please improve this change description.
> +++ b/drivers/gpu/drm/imx/parallel-display.c
> @@ -227,14 +227,18 @@ static int imx_pd_bind(struct device *dev, struct device *master, void *data)
>
> /* port@1 is the output port */
> ret = drm_of_find_panel_or_bridge(np, 1, 0, &imxpd->panel, &imxpd->bridge);
> - if (ret && ret != -ENODEV)
> + if (ret && ret != -ENODEV) {
> + kfree(imxpd->edid);
> return ret;
> + }
>
> imxpd->dev = dev;
Please use a jump target here instead of adding duplicate source code
for the completion of exception handling.
if (ret && ret != -ENODEV)
- return ret;
+ goto free_edid;
…
+free_edid:
+ kfree(imxpd->edid);
+ return ret;
Regards,
Markus
I have taken another look also at the implementation of the function “imx_pd_bind”.
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/gpu/drm/imx/parallel-display.c?id=43b815c6a8e7dbccb5b8bd9c4b099c24bc22d135#n197
https://elixir.bootlin.com/linux/v5.4-rc1/source/drivers/gpu/drm/imx/parallel-display.c#L197
Now I find an unchecked call of the function “kmemdup” suspicious.
Will this detail trigger further software development considerations?
Regards,
Markus
Hi Markus,
I agree with you, kmemdup may fail so a null check seems necessary there.
On Sun, Oct 6, 2019 at 4:33 AM Markus Elfring <[email protected]> wrote:
>
> I have taken another look also at the implementation of the function “imx_pd_bind”.
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/gpu/drm/imx/parallel-display.c?id=43b815c6a8e7dbccb5b8bd9c4b099c24bc22d135#n197
> https://elixir.bootlin.com/linux/v5.4-rc1/source/drivers/gpu/drm/imx/parallel-display.c#L197
>
> Now I find an unchecked call of the function “kmemdup” suspicious.
> Will this detail trigger further software development considerations?
>
> Regards,
> Markus
--
Navid.
> I agree with you, kmemdup may fail so a null check seems necessary there.
Would this place (and similar ones) be pointed out for further considerations
also by the source code analysis tool which your software research group
seems to be developing?
https://github.com/umnsec/cheq/
Regards,
Markus
From: Markus Elfring <[email protected]>
Date: Sat, 12 Oct 2019 10:45:45 +0200
Two update suggestions were taken into account
from static source code analysis.
Markus Elfring (2):
Fix error handling for a kmemdup() call in imx_pd_bind()
Fix error handling for a kmemdup() call in imx_ldb_panel_ddc()
drivers/gpu/drm/imx/imx-ldb.c | 2 ++
drivers/gpu/drm/imx/parallel-display.c | 7 ++++++-
2 files changed, 8 insertions(+), 1 deletion(-)
--
2.23.0
From: Markus Elfring <[email protected]>
Date: Sat, 12 Oct 2019 10:30:21 +0200
The return value from a call of the function “kmemdup” was not checked
in this function implementation. Thus add the corresponding error handling.
Fixes: 19022aaae677dfa171a719e9d1ff04823ce65a65 ("staging: drm/imx: Add parallel display support")
Signed-off-by: Markus Elfring <[email protected]>
---
drivers/gpu/drm/imx/parallel-display.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/imx/parallel-display.c b/drivers/gpu/drm/imx/parallel-display.c
index 35518e5de356..39c4798f56b6 100644
--- a/drivers/gpu/drm/imx/parallel-display.c
+++ b/drivers/gpu/drm/imx/parallel-display.c
@@ -210,8 +210,13 @@ static int imx_pd_bind(struct device *dev, struct device *master, void *data)
return -ENOMEM;
edidp = of_get_property(np, "edid", &imxpd->edid_len);
- if (edidp)
+ if (edidp) {
imxpd->edid = kmemdup(edidp, imxpd->edid_len, GFP_KERNEL);
+ if (!imxpd->edid) {
+ devm_kfree(dev, imxpd);
+ return -ENOMEM;
+ }
+ }
ret = of_property_read_string(np, "interface-pix-fmt", &fmt);
if (!ret) {
--
2.23.0
From: Markus Elfring <[email protected]>
Date: Sat, 12 Oct 2019 10:33:47 +0200
The return value from a call of the function “kmemdup” was not checked
in this function implementation. Thus add the corresponding error handling.
This issue was detected by using the Coccinelle software.
Fixes: dc80d7038883feca2abd08975165bc0d83c84762 ("drm/imx-ldb: Add support to drm-bridge")
Signed-off-by: Markus Elfring <[email protected]>
---
drivers/gpu/drm/imx/imx-ldb.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/drivers/gpu/drm/imx/imx-ldb.c b/drivers/gpu/drm/imx/imx-ldb.c
index 208069faf183..801a2265dd11 100644
--- a/drivers/gpu/drm/imx/imx-ldb.c
+++ b/drivers/gpu/drm/imx/imx-ldb.c
@@ -569,6 +569,8 @@ static int imx_ldb_panel_ddc(struct device *dev,
channel->edid = kmemdup(edidp,
channel->edid_len,
GFP_KERNEL);
+ if (!channel->edid)
+ return -ENOMEM;
} else if (!channel->panel) {
/* fallback to display-timings node */
ret = of_get_drm_display_mode(child,
--
2.23.0
> +free_edid:
> + kfree(imxpd->edid);
> + return ret;
I have taken another look at this change idea.
Can the function call “devm_kfree(dev, imxpd)” become relevant
also at this place?
Would you like to combine it with the update suggestion
“Fix error handling for a kmemdup() call in imx_pd_bind()”?
https://lore.kernel.org/r/[email protected]/
https://lore.kernel.org/patchwork/patch/1138912/
https://lkml.org/lkml/2019/10/12/87
Regards,
Markus
On Sat, Oct 12, 2019 at 4:07 AM Markus Elfring <[email protected]> wrote:
>
> From: Markus Elfring <[email protected]>
> Date: Sat, 12 Oct 2019 10:30:21 +0200
>
> The return value from a call of the function “kmemdup” was not checked
> in this function implementation. Thus add the corresponding error handling.
>
> Fixes: 19022aaae677dfa171a719e9d1ff04823ce65a65 ("staging: drm/imx: Add parallel display support")
> Signed-off-by: Markus Elfring <[email protected]>
> ---
> drivers/gpu/drm/imx/parallel-display.c | 7 ++++++-
> 1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/imx/parallel-display.c b/drivers/gpu/drm/imx/parallel-display.c
> index 35518e5de356..39c4798f56b6 100644
> --- a/drivers/gpu/drm/imx/parallel-display.c
> +++ b/drivers/gpu/drm/imx/parallel-display.c
> @@ -210,8 +210,13 @@ static int imx_pd_bind(struct device *dev, struct device *master, void *data)
> return -ENOMEM;
>
> edidp = of_get_property(np, "edid", &imxpd->edid_len);
> - if (edidp)
> + if (edidp) {
> imxpd->edid = kmemdup(edidp, imxpd->edid_len, GFP_KERNEL);
> + if (!imxpd->edid) {
> + devm_kfree(dev, imxpd);
You should not try to free imxpd here as it is a resource-managed
allocation via devm_kzalloc(). It means memory allocated with this
function is
automatically freed on driver detach. So, this patch introduces a double-free.
> + return -ENOMEM;
> + }
> + }
>
> ret = of_property_read_string(np, "interface-pix-fmt", &fmt);
> if (!ret) {
> --
> 2.23.0
>
--
Navid.
No, that is not correct! You should not try to free imxpd here as it
is a resource-managed allocation via devm_kzalloc(). It means memory
allocated with this function is automatically freed on driver detach.
So, this patch introduces a double-free.
On Sat, Oct 12, 2019 at 6:54 AM Markus Elfring <[email protected]> wrote:
>
> > +free_edid:
> > + kfree(imxpd->edid);
> > + return ret;
>
> I have taken another look at this change idea.
> Can the function call “devm_kfree(dev, imxpd)” become relevant
> also at this place?
>
> Would you like to combine it with the update suggestion
> “Fix error handling for a kmemdup() call in imx_pd_bind()”?
> https://lore.kernel.org/r/[email protected]/
> https://lore.kernel.org/patchwork/patch/1138912/
> https://lkml.org/lkml/2019/10/12/87
>
> Regards,
> Markus
--
Navid.
On Sat, 12 Oct 2019, Navid Emamdoost wrote:
> On Sat, Oct 12, 2019 at 4:07 AM Markus Elfring <[email protected]> wrote:
> >
> > From: Markus Elfring <[email protected]>
> > Date: Sat, 12 Oct 2019 10:30:21 +0200
> >
> > The return value from a call of the function “kmemdup” was not checked
> > in this function implementation. Thus add the corresponding error handling.
> >
> > Fixes: 19022aaae677dfa171a719e9d1ff04823ce65a65 ("staging: drm/imx: Add parallel display support")
> > Signed-off-by: Markus Elfring <[email protected]>
> > ---
> > drivers/gpu/drm/imx/parallel-display.c | 7 ++++++-
> > 1 file changed, 6 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/gpu/drm/imx/parallel-display.c b/drivers/gpu/drm/imx/parallel-display.c
> > index 35518e5de356..39c4798f56b6 100644
> > --- a/drivers/gpu/drm/imx/parallel-display.c
> > +++ b/drivers/gpu/drm/imx/parallel-display.c
> > @@ -210,8 +210,13 @@ static int imx_pd_bind(struct device *dev, struct device *master, void *data)
> > return -ENOMEM;
> >
> > edidp = of_get_property(np, "edid", &imxpd->edid_len);
> > - if (edidp)
> > + if (edidp) {
> > imxpd->edid = kmemdup(edidp, imxpd->edid_len, GFP_KERNEL);
> > + if (!imxpd->edid) {
> > + devm_kfree(dev, imxpd);
>
> You should not try to free imxpd here as it is a resource-managed
> allocation via devm_kzalloc(). It means memory allocated with this
> function is
> automatically freed on driver detach. So, this patch introduces a double-free.
No, it's not double freed since the proposed code frees it with a devm
function, removing it from the list of things to free later. One can
wonder why the free has to be made apparent, though.
julia
>
> > + return -ENOMEM;
> > + }
> > + }
> >
> > ret = of_property_read_string(np, "interface-pix-fmt", &fmt);
> > if (!ret) {
> > --
> > 2.23.0
> >
>
>
> --
> Navid.
>
On Fri, Oct 4, 2019 at 2:09 PM Navid Emamdoost
<[email protected]> wrote:
>
> In imx_pd_bind, the duplicated memory for imxpd->edid via kmemdup should
> be released in drm_of_find_panel_or_bridge or imx_pd_register fail.
>
> Fixes: ebc944613567 ("drm: convert drivers to use drm_of_find_panel_or_bridge")
> Fixes: 19022aaae677 ("staging: drm/imx: Add parallel display support")
> Signed-off-by: Navid Emamdoost <[email protected]>
> ---
Would you please review this patch?
Thanks,
> drivers/gpu/drm/imx/parallel-display.c | 8 ++++++--
> 1 file changed, 6 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/imx/parallel-display.c b/drivers/gpu/drm/imx/parallel-display.c
> index e7ce17503ae1..9522d2cb0ad5 100644
> --- a/drivers/gpu/drm/imx/parallel-display.c
> +++ b/drivers/gpu/drm/imx/parallel-display.c
> @@ -227,14 +227,18 @@ static int imx_pd_bind(struct device *dev, struct device *master, void *data)
>
> /* port@1 is the output port */
> ret = drm_of_find_panel_or_bridge(np, 1, 0, &imxpd->panel, &imxpd->bridge);
> - if (ret && ret != -ENODEV)
> + if (ret && ret != -ENODEV) {
> + kfree(imxpd->edid);
> return ret;
> + }
>
> imxpd->dev = dev;
>
> ret = imx_pd_register(drm, imxpd);
> - if (ret)
> + if (ret) {
> + kfree(imxpd->edid);
> return ret;
> + }
>
> dev_set_drvdata(dev, imxpd);
>
> --
> 2.17.1
>
--
Navid.
Hi Navid,
On 19-11-21 12:31, Navid Emamdoost wrote:
> On Fri, Oct 4, 2019 at 2:09 PM Navid Emamdoost
> <[email protected]> wrote:
> >
> > In imx_pd_bind, the duplicated memory for imxpd->edid via kmemdup should
> > be released in drm_of_find_panel_or_bridge or imx_pd_register fail.
> >
> > Fixes: ebc944613567 ("drm: convert drivers to use drm_of_find_panel_or_bridge")
> > Fixes: 19022aaae677 ("staging: drm/imx: Add parallel display support")
> > Signed-off-by: Navid Emamdoost <[email protected]>
> > ---
>
> Would you please review this patch?
>
> Thanks,
I currently work on the drm/imx driver(s) to avoid errors like [1].
Hopefully I have a working version till next week. There I fixed this
issue by using the devm_kmemdup() and dropped the explicit kfree()
within unbind().
[1] https://www.spinics.net/lists/dri-devel/msg189388.html
Regards,
Marco
>
> > drivers/gpu/drm/imx/parallel-display.c | 8 ++++++--
> > 1 file changed, 6 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/imx/parallel-display.c b/drivers/gpu/drm/imx/parallel-display.c
> > index e7ce17503ae1..9522d2cb0ad5 100644
> > --- a/drivers/gpu/drm/imx/parallel-display.c
> > +++ b/drivers/gpu/drm/imx/parallel-display.c
> > @@ -227,14 +227,18 @@ static int imx_pd_bind(struct device *dev, struct device *master, void *data)
> >
> > /* port@1 is the output port */
> > ret = drm_of_find_panel_or_bridge(np, 1, 0, &imxpd->panel, &imxpd->bridge);
> > - if (ret && ret != -ENODEV)
> > + if (ret && ret != -ENODEV) {
> > + kfree(imxpd->edid);
> > return ret;
> > + }
> >
> > imxpd->dev = dev;
> >
> > ret = imx_pd_register(drm, imxpd);
> > - if (ret)
> > + if (ret) {
> > + kfree(imxpd->edid);
> > return ret;
> > + }
> >
> > dev_set_drvdata(dev, imxpd);
> >
> > --
> > 2.17.1
> >
>
>
> --
> Navid.
>
>
--
Pengutronix e.K. | |
Steuerwalder Str. 21 | http://www.pengutronix.de/ |
31137 Hildesheim, Germany | Phone: +49-5121-206917-0 |
Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
Thanks for the update.
On Fri, Nov 22, 2019 at 1:22 AM Marco Felsch <[email protected]> wrote:
>
> Hi Navid,
>
> On 19-11-21 12:31, Navid Emamdoost wrote:
> > On Fri, Oct 4, 2019 at 2:09 PM Navid Emamdoost
> > <[email protected]> wrote:
> > >
> > > In imx_pd_bind, the duplicated memory for imxpd->edid via kmemdup should
> > > be released in drm_of_find_panel_or_bridge or imx_pd_register fail.
> > >
> > > Fixes: ebc944613567 ("drm: convert drivers to use drm_of_find_panel_or_bridge")
> > > Fixes: 19022aaae677 ("staging: drm/imx: Add parallel display support")
> > > Signed-off-by: Navid Emamdoost <[email protected]>
> > > ---
> >
> > Would you please review this patch?
> >
> > Thanks,
>
> I currently work on the drm/imx driver(s) to avoid errors like [1].
> Hopefully I have a working version till next week. There I fixed this
> issue by using the devm_kmemdup() and dropped the explicit kfree()
> within unbind().
>
> [1] https://www.spinics.net/lists/dri-devel/msg189388.html
>
> Regards,
> Marco
>
> >
> > > drivers/gpu/drm/imx/parallel-display.c | 8 ++++++--
> > > 1 file changed, 6 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/imx/parallel-display.c b/drivers/gpu/drm/imx/parallel-display.c
> > > index e7ce17503ae1..9522d2cb0ad5 100644
> > > --- a/drivers/gpu/drm/imx/parallel-display.c
> > > +++ b/drivers/gpu/drm/imx/parallel-display.c
> > > @@ -227,14 +227,18 @@ static int imx_pd_bind(struct device *dev, struct device *master, void *data)
> > >
> > > /* port@1 is the output port */
> > > ret = drm_of_find_panel_or_bridge(np, 1, 0, &imxpd->panel, &imxpd->bridge);
> > > - if (ret && ret != -ENODEV)
> > > + if (ret && ret != -ENODEV) {
> > > + kfree(imxpd->edid);
> > > return ret;
> > > + }
> > >
> > > imxpd->dev = dev;
> > >
> > > ret = imx_pd_register(drm, imxpd);
> > > - if (ret)
> > > + if (ret) {
> > > + kfree(imxpd->edid);
> > > return ret;
> > > + }
> > >
> > > dev_set_drvdata(dev, imxpd);
> > >
> > > --
> > > 2.17.1
> > >
> >
> >
> > --
> > Navid.
> >
> >
>
> --
> Pengutronix e.K. | |
> Steuerwalder Str. 21 | http://www.pengutronix.de/ |
> 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 |
> Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
--
Navid.