All of lore.kernel.org
 help / color / mirror / Atom feed
* [Linux-kernel-mentees] [PATCH] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
@ 2020-07-28  1:43 ` Peilin Ye
  0 siblings, 0 replies; 18+ messages in thread
From: Peilin Ye @ 2020-07-28  1:43 UTC (permalink / raw)
  To: Maarten Lankhorst, Maxime Ripard, Thomas Zimmermann
  Cc: Peilin Ye, David Airlie, Daniel Vetter, Dan Carpenter,
	Arnd Bergmann, Greg Kroah-Hartman, linux-kernel-mentees,
	dri-devel, linux-kernel

copy_one_buf() is copying uninitialized stack memory to userspace due to
the compiler not initializing holes in statically allocated structures.
Fix it by initializing `v` with memset().

Cc: stable@vger.kernel.org
Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
---
 drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
index a0735fbc144b..f99cd4a3f951 100644
--- a/drivers/gpu/drm/drm_bufs.c
+++ b/drivers/gpu/drm/drm_bufs.c
@@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
 {
 	struct drm_buf_info *request = data;
 	struct drm_buf_desc __user *to = &request->list[count];
-	struct drm_buf_desc v = {.count = from->buf_count,
-				 .size = from->buf_size,
-				 .low_mark = from->low_mark,
-				 .high_mark = from->high_mark};
+	struct drm_buf_desc v;
+
+	memset(&v, 0, sizeof(v));
+
+	v.count = from->buf_count;
+	v.size = from->buf_size;
+	v.low_mark = from->low_mark;
+	v.high_mark = from->high_mark;
 
 	if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
 		return -EFAULT;
-- 
2.25.1


^ permalink raw reply related	[flat|nested] 18+ messages in thread

* [Linux-kernel-mentees] [PATCH] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
@ 2020-07-28  1:43 ` Peilin Ye
  0 siblings, 0 replies; 18+ messages in thread
From: Peilin Ye @ 2020-07-28  1:43 UTC (permalink / raw)
  To: Maarten Lankhorst, Maxime Ripard, Thomas Zimmermann
  Cc: Arnd Bergmann, David Airlie, linux-kernel, dri-devel,
	Daniel Vetter, linux-kernel-mentees, Peilin Ye, Dan Carpenter

copy_one_buf() is copying uninitialized stack memory to userspace due to
the compiler not initializing holes in statically allocated structures.
Fix it by initializing `v` with memset().

Cc: stable@vger.kernel.org
Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
---
 drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
index a0735fbc144b..f99cd4a3f951 100644
--- a/drivers/gpu/drm/drm_bufs.c
+++ b/drivers/gpu/drm/drm_bufs.c
@@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
 {
 	struct drm_buf_info *request = data;
 	struct drm_buf_desc __user *to = &request->list[count];
-	struct drm_buf_desc v = {.count = from->buf_count,
-				 .size = from->buf_size,
-				 .low_mark = from->low_mark,
-				 .high_mark = from->high_mark};
+	struct drm_buf_desc v;
+
+	memset(&v, 0, sizeof(v));
+
+	v.count = from->buf_count;
+	v.size = from->buf_size;
+	v.low_mark = from->low_mark;
+	v.high_mark = from->high_mark;
 
 	if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
 		return -EFAULT;
-- 
2.25.1

_______________________________________________
Linux-kernel-mentees mailing list
Linux-kernel-mentees@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees

^ permalink raw reply related	[flat|nested] 18+ messages in thread

* [Linux-kernel-mentees] [PATCH] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
@ 2020-07-28  1:43 ` Peilin Ye
  0 siblings, 0 replies; 18+ messages in thread
From: Peilin Ye @ 2020-07-28  1:43 UTC (permalink / raw)
  To: Maarten Lankhorst, Maxime Ripard, Thomas Zimmermann
  Cc: Arnd Bergmann, David Airlie, Greg Kroah-Hartman, linux-kernel,
	dri-devel, linux-kernel-mentees, Peilin Ye, Dan Carpenter

copy_one_buf() is copying uninitialized stack memory to userspace due to
the compiler not initializing holes in statically allocated structures.
Fix it by initializing `v` with memset().

Cc: stable@vger.kernel.org
Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
---
 drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
index a0735fbc144b..f99cd4a3f951 100644
--- a/drivers/gpu/drm/drm_bufs.c
+++ b/drivers/gpu/drm/drm_bufs.c
@@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
 {
 	struct drm_buf_info *request = data;
 	struct drm_buf_desc __user *to = &request->list[count];
-	struct drm_buf_desc v = {.count = from->buf_count,
-				 .size = from->buf_size,
-				 .low_mark = from->low_mark,
-				 .high_mark = from->high_mark};
+	struct drm_buf_desc v;
+
+	memset(&v, 0, sizeof(v));
+
+	v.count = from->buf_count;
+	v.size = from->buf_size;
+	v.low_mark = from->low_mark;
+	v.high_mark = from->high_mark;
 
 	if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
 		return -EFAULT;
-- 
2.25.1

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

^ permalink raw reply related	[flat|nested] 18+ messages in thread

* Re: [Linux-kernel-mentees] [PATCH] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
  2020-07-28  1:43 ` Peilin Ye
  (?)
@ 2020-07-28  8:11   ` Arnd Bergmann
  -1 siblings, 0 replies; 18+ messages in thread
From: Arnd Bergmann @ 2020-07-28  8:11 UTC (permalink / raw)
  To: Peilin Ye
  Cc: Maarten Lankhorst, Maxime Ripard, Thomas Zimmermann,
	David Airlie, Daniel Vetter, Dan Carpenter, Greg Kroah-Hartman,
	linux-kernel-mentees, dri-devel, linux-kernel

On Tue, Jul 28, 2020 at 3:45 AM Peilin Ye <yepeilin.cs@gmail.com> wrote:
>
> copy_one_buf() is copying uninitialized stack memory to userspace due to
> the compiler not initializing holes in statically allocated structures.
> Fix it by initializing `v` with memset().

I would add 'potentially' somewhere in that description: it is architecture
dependent whether there are holes in this structure as 'enum' types
and 'long' are both dependent on the ABI, and even if there is a hole,
it is undefined behavior whether the hold gets initialized.

Other than that, the patch looks good.

> Cc: stable@vger.kernel.org
> Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
> Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
> Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>

Acked-by: Arnd Bergmann <arnd@arndb.de>

> ---
>  drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
> index a0735fbc144b..f99cd4a3f951 100644
> --- a/drivers/gpu/drm/drm_bufs.c
> +++ b/drivers/gpu/drm/drm_bufs.c
> @@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
>  {
>         struct drm_buf_info *request = data;
>         struct drm_buf_desc __user *to = &request->list[count];
> -       struct drm_buf_desc v = {.count = from->buf_count,
> -                                .size = from->buf_size,
> -                                .low_mark = from->low_mark,
> -                                .high_mark = from->high_mark};
> +       struct drm_buf_desc v;
> +
> +       memset(&v, 0, sizeof(v));
> +
> +       v.count = from->buf_count;
> +       v.size = from->buf_size;
> +       v.low_mark = from->low_mark;
> +       v.high_mark = from->high_mark;
>
>         if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
>                 return -EFAULT;
> --
> 2.25.1
>

^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [Linux-kernel-mentees] [PATCH] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
@ 2020-07-28  8:11   ` Arnd Bergmann
  0 siblings, 0 replies; 18+ messages in thread
From: Arnd Bergmann @ 2020-07-28  8:11 UTC (permalink / raw)
  To: Peilin Ye
  Cc: Thomas Zimmermann, David Airlie, Maarten Lankhorst, linux-kernel,
	Maxime Ripard, dri-devel, Daniel Vetter, linux-kernel-mentees,
	Dan Carpenter

On Tue, Jul 28, 2020 at 3:45 AM Peilin Ye <yepeilin.cs@gmail.com> wrote:
>
> copy_one_buf() is copying uninitialized stack memory to userspace due to
> the compiler not initializing holes in statically allocated structures.
> Fix it by initializing `v` with memset().

I would add 'potentially' somewhere in that description: it is architecture
dependent whether there are holes in this structure as 'enum' types
and 'long' are both dependent on the ABI, and even if there is a hole,
it is undefined behavior whether the hold gets initialized.

Other than that, the patch looks good.

> Cc: stable@vger.kernel.org
> Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
> Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
> Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>

Acked-by: Arnd Bergmann <arnd@arndb.de>

> ---
>  drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
> index a0735fbc144b..f99cd4a3f951 100644
> --- a/drivers/gpu/drm/drm_bufs.c
> +++ b/drivers/gpu/drm/drm_bufs.c
> @@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
>  {
>         struct drm_buf_info *request = data;
>         struct drm_buf_desc __user *to = &request->list[count];
> -       struct drm_buf_desc v = {.count = from->buf_count,
> -                                .size = from->buf_size,
> -                                .low_mark = from->low_mark,
> -                                .high_mark = from->high_mark};
> +       struct drm_buf_desc v;
> +
> +       memset(&v, 0, sizeof(v));
> +
> +       v.count = from->buf_count;
> +       v.size = from->buf_size;
> +       v.low_mark = from->low_mark;
> +       v.high_mark = from->high_mark;
>
>         if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
>                 return -EFAULT;
> --
> 2.25.1
>
_______________________________________________
Linux-kernel-mentees mailing list
Linux-kernel-mentees@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees

^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [Linux-kernel-mentees] [PATCH] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
@ 2020-07-28  8:11   ` Arnd Bergmann
  0 siblings, 0 replies; 18+ messages in thread
From: Arnd Bergmann @ 2020-07-28  8:11 UTC (permalink / raw)
  To: Peilin Ye
  Cc: Thomas Zimmermann, David Airlie, Greg Kroah-Hartman,
	linux-kernel, dri-devel, linux-kernel-mentees, Dan Carpenter

On Tue, Jul 28, 2020 at 3:45 AM Peilin Ye <yepeilin.cs@gmail.com> wrote:
>
> copy_one_buf() is copying uninitialized stack memory to userspace due to
> the compiler not initializing holes in statically allocated structures.
> Fix it by initializing `v` with memset().

I would add 'potentially' somewhere in that description: it is architecture
dependent whether there are holes in this structure as 'enum' types
and 'long' are both dependent on the ABI, and even if there is a hole,
it is undefined behavior whether the hold gets initialized.

Other than that, the patch looks good.

> Cc: stable@vger.kernel.org
> Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
> Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
> Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>

Acked-by: Arnd Bergmann <arnd@arndb.de>

> ---
>  drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
> index a0735fbc144b..f99cd4a3f951 100644
> --- a/drivers/gpu/drm/drm_bufs.c
> +++ b/drivers/gpu/drm/drm_bufs.c
> @@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
>  {
>         struct drm_buf_info *request = data;
>         struct drm_buf_desc __user *to = &request->list[count];
> -       struct drm_buf_desc v = {.count = from->buf_count,
> -                                .size = from->buf_size,
> -                                .low_mark = from->low_mark,
> -                                .high_mark = from->high_mark};
> +       struct drm_buf_desc v;
> +
> +       memset(&v, 0, sizeof(v));
> +
> +       v.count = from->buf_count;
> +       v.size = from->buf_size;
> +       v.low_mark = from->low_mark;
> +       v.high_mark = from->high_mark;
>
>         if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
>                 return -EFAULT;
> --
> 2.25.1
>
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [Linux-kernel-mentees] [PATCH] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
  2020-07-28  8:11   ` Arnd Bergmann
  (?)
@ 2020-07-28 11:15     ` Peilin Ye
  -1 siblings, 0 replies; 18+ messages in thread
From: Peilin Ye @ 2020-07-28 11:15 UTC (permalink / raw)
  To: Arnd Bergmann
  Cc: Maarten Lankhorst, Maxime Ripard, Thomas Zimmermann,
	David Airlie, Daniel Vetter, Dan Carpenter, Greg Kroah-Hartman,
	linux-kernel-mentees, dri-devel, linux-kernel

On Tue, Jul 28, 2020 at 10:11:09AM +0200, Arnd Bergmann wrote:
> On Tue, Jul 28, 2020 at 3:45 AM Peilin Ye <yepeilin.cs@gmail.com> wrote:
> >
> > copy_one_buf() is copying uninitialized stack memory to userspace due to
> > the compiler not initializing holes in statically allocated structures.
> > Fix it by initializing `v` with memset().
> 
> I would add 'potentially' somewhere in that description: it is architecture
> dependent whether there are holes in this structure as 'enum' types
> and 'long' are both dependent on the ABI, and even if there is a hole,
> it is undefined behavior whether the hold gets initialized.

I see. I will fix that up. Thank you for the advice!

Peilin Ye

> Other than that, the patch looks good.
> 
> > Cc: stable@vger.kernel.org
> > Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
> > Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
> > Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
> 
> Acked-by: Arnd Bergmann <arnd@arndb.de>

^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [Linux-kernel-mentees] [PATCH] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
@ 2020-07-28 11:15     ` Peilin Ye
  0 siblings, 0 replies; 18+ messages in thread
From: Peilin Ye @ 2020-07-28 11:15 UTC (permalink / raw)
  To: Arnd Bergmann
  Cc: Thomas Zimmermann, David Airlie, Maarten Lankhorst, linux-kernel,
	Maxime Ripard, dri-devel, Daniel Vetter, linux-kernel-mentees,
	Dan Carpenter

On Tue, Jul 28, 2020 at 10:11:09AM +0200, Arnd Bergmann wrote:
> On Tue, Jul 28, 2020 at 3:45 AM Peilin Ye <yepeilin.cs@gmail.com> wrote:
> >
> > copy_one_buf() is copying uninitialized stack memory to userspace due to
> > the compiler not initializing holes in statically allocated structures.
> > Fix it by initializing `v` with memset().
> 
> I would add 'potentially' somewhere in that description: it is architecture
> dependent whether there are holes in this structure as 'enum' types
> and 'long' are both dependent on the ABI, and even if there is a hole,
> it is undefined behavior whether the hold gets initialized.

I see. I will fix that up. Thank you for the advice!

Peilin Ye

> Other than that, the patch looks good.
> 
> > Cc: stable@vger.kernel.org
> > Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
> > Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
> > Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
> 
> Acked-by: Arnd Bergmann <arnd@arndb.de>
_______________________________________________
Linux-kernel-mentees mailing list
Linux-kernel-mentees@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees

^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [Linux-kernel-mentees] [PATCH] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
@ 2020-07-28 11:15     ` Peilin Ye
  0 siblings, 0 replies; 18+ messages in thread
From: Peilin Ye @ 2020-07-28 11:15 UTC (permalink / raw)
  To: Arnd Bergmann
  Cc: Thomas Zimmermann, David Airlie, Greg Kroah-Hartman,
	linux-kernel, dri-devel, linux-kernel-mentees, Dan Carpenter

On Tue, Jul 28, 2020 at 10:11:09AM +0200, Arnd Bergmann wrote:
> On Tue, Jul 28, 2020 at 3:45 AM Peilin Ye <yepeilin.cs@gmail.com> wrote:
> >
> > copy_one_buf() is copying uninitialized stack memory to userspace due to
> > the compiler not initializing holes in statically allocated structures.
> > Fix it by initializing `v` with memset().
> 
> I would add 'potentially' somewhere in that description: it is architecture
> dependent whether there are holes in this structure as 'enum' types
> and 'long' are both dependent on the ABI, and even if there is a hole,
> it is undefined behavior whether the hold gets initialized.

I see. I will fix that up. Thank you for the advice!

Peilin Ye

> Other than that, the patch looks good.
> 
> > Cc: stable@vger.kernel.org
> > Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
> > Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
> > Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
> 
> Acked-by: Arnd Bergmann <arnd@arndb.de>
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

^ permalink raw reply	[flat|nested] 18+ messages in thread

* [Linux-kernel-mentees] [PATCH v2] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
  2020-07-28  1:43 ` Peilin Ye
  (?)
@ 2020-07-28 11:52   ` Peilin Ye
  -1 siblings, 0 replies; 18+ messages in thread
From: Peilin Ye @ 2020-07-28 11:52 UTC (permalink / raw)
  To: Maarten Lankhorst, Maxime Ripard, Thomas Zimmermann
  Cc: Peilin Ye, David Airlie, Daniel Vetter, Dan Carpenter,
	Arnd Bergmann, Greg Kroah-Hartman, linux-kernel-mentees,
	dri-devel, linux-kernel

Currently `struct drm_buf_desc` is defined as follows:

struct drm_buf_desc {
	int count;
	int size;
	int low_mark;
	int high_mark;
	enum {
		_DRM_PAGE_ALIGN = 0x01,
		_DRM_AGP_BUFFER = 0x02,
		_DRM_SG_BUFFER = 0x04,
		_DRM_FB_BUFFER = 0x08,
		_DRM_PCI_BUFFER_RO = 0x10
	} flags;
	unsigned long agp_start;
};

copy_one_buf() is potentially copying uninitialized kernel stack memory
to userspace, since the compiler may leave such "holes" (around `.flags`
and `.agp_start` fields) in this statically allocated structure. Prevent
it by initializing `v` with memset().

Cc: stable@vger.kernel.org
Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
---
Change in v2:
    - Improve commit description. (Suggested by Arnd Bergmann
      <arnd@arndb.de>)

 drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
index a0735fbc144b..f99cd4a3f951 100644
--- a/drivers/gpu/drm/drm_bufs.c
+++ b/drivers/gpu/drm/drm_bufs.c
@@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
 {
 	struct drm_buf_info *request = data;
 	struct drm_buf_desc __user *to = &request->list[count];
-	struct drm_buf_desc v = {.count = from->buf_count,
-				 .size = from->buf_size,
-				 .low_mark = from->low_mark,
-				 .high_mark = from->high_mark};
+	struct drm_buf_desc v;
+
+	memset(&v, 0, sizeof(v));
+
+	v.count = from->buf_count;
+	v.size = from->buf_size;
+	v.low_mark = from->low_mark;
+	v.high_mark = from->high_mark;
 
 	if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
 		return -EFAULT;
-- 
2.25.1


^ permalink raw reply related	[flat|nested] 18+ messages in thread

* [Linux-kernel-mentees] [PATCH v2] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
@ 2020-07-28 11:52   ` Peilin Ye
  0 siblings, 0 replies; 18+ messages in thread
From: Peilin Ye @ 2020-07-28 11:52 UTC (permalink / raw)
  To: Maarten Lankhorst, Maxime Ripard, Thomas Zimmermann
  Cc: Arnd Bergmann, David Airlie, linux-kernel, dri-devel,
	Daniel Vetter, linux-kernel-mentees, Peilin Ye, Dan Carpenter

Currently `struct drm_buf_desc` is defined as follows:

struct drm_buf_desc {
	int count;
	int size;
	int low_mark;
	int high_mark;
	enum {
		_DRM_PAGE_ALIGN = 0x01,
		_DRM_AGP_BUFFER = 0x02,
		_DRM_SG_BUFFER = 0x04,
		_DRM_FB_BUFFER = 0x08,
		_DRM_PCI_BUFFER_RO = 0x10
	} flags;
	unsigned long agp_start;
};

copy_one_buf() is potentially copying uninitialized kernel stack memory
to userspace, since the compiler may leave such "holes" (around `.flags`
and `.agp_start` fields) in this statically allocated structure. Prevent
it by initializing `v` with memset().

Cc: stable@vger.kernel.org
Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
---
Change in v2:
    - Improve commit description. (Suggested by Arnd Bergmann
      <arnd@arndb.de>)

 drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
index a0735fbc144b..f99cd4a3f951 100644
--- a/drivers/gpu/drm/drm_bufs.c
+++ b/drivers/gpu/drm/drm_bufs.c
@@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
 {
 	struct drm_buf_info *request = data;
 	struct drm_buf_desc __user *to = &request->list[count];
-	struct drm_buf_desc v = {.count = from->buf_count,
-				 .size = from->buf_size,
-				 .low_mark = from->low_mark,
-				 .high_mark = from->high_mark};
+	struct drm_buf_desc v;
+
+	memset(&v, 0, sizeof(v));
+
+	v.count = from->buf_count;
+	v.size = from->buf_size;
+	v.low_mark = from->low_mark;
+	v.high_mark = from->high_mark;
 
 	if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
 		return -EFAULT;
-- 
2.25.1

_______________________________________________
Linux-kernel-mentees mailing list
Linux-kernel-mentees@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees

^ permalink raw reply related	[flat|nested] 18+ messages in thread

* [Linux-kernel-mentees] [PATCH v2] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
@ 2020-07-28 11:52   ` Peilin Ye
  0 siblings, 0 replies; 18+ messages in thread
From: Peilin Ye @ 2020-07-28 11:52 UTC (permalink / raw)
  To: Maarten Lankhorst, Maxime Ripard, Thomas Zimmermann
  Cc: Arnd Bergmann, David Airlie, Greg Kroah-Hartman, linux-kernel,
	dri-devel, linux-kernel-mentees, Peilin Ye, Dan Carpenter

Currently `struct drm_buf_desc` is defined as follows:

struct drm_buf_desc {
	int count;
	int size;
	int low_mark;
	int high_mark;
	enum {
		_DRM_PAGE_ALIGN = 0x01,
		_DRM_AGP_BUFFER = 0x02,
		_DRM_SG_BUFFER = 0x04,
		_DRM_FB_BUFFER = 0x08,
		_DRM_PCI_BUFFER_RO = 0x10
	} flags;
	unsigned long agp_start;
};

copy_one_buf() is potentially copying uninitialized kernel stack memory
to userspace, since the compiler may leave such "holes" (around `.flags`
and `.agp_start` fields) in this statically allocated structure. Prevent
it by initializing `v` with memset().

Cc: stable@vger.kernel.org
Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
---
Change in v2:
    - Improve commit description. (Suggested by Arnd Bergmann
      <arnd@arndb.de>)

 drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
index a0735fbc144b..f99cd4a3f951 100644
--- a/drivers/gpu/drm/drm_bufs.c
+++ b/drivers/gpu/drm/drm_bufs.c
@@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
 {
 	struct drm_buf_info *request = data;
 	struct drm_buf_desc __user *to = &request->list[count];
-	struct drm_buf_desc v = {.count = from->buf_count,
-				 .size = from->buf_size,
-				 .low_mark = from->low_mark,
-				 .high_mark = from->high_mark};
+	struct drm_buf_desc v;
+
+	memset(&v, 0, sizeof(v));
+
+	v.count = from->buf_count;
+	v.size = from->buf_size;
+	v.low_mark = from->low_mark;
+	v.high_mark = from->high_mark;
 
 	if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
 		return -EFAULT;
-- 
2.25.1

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

^ permalink raw reply related	[flat|nested] 18+ messages in thread

* RE: [Linux-kernel-mentees] [PATCH v2] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
  2020-07-28 11:52   ` Peilin Ye
  (?)
@ 2020-07-28 12:35     ` David Laight
  -1 siblings, 0 replies; 18+ messages in thread
From: David Laight @ 2020-07-28 12:35 UTC (permalink / raw)
  To: 'Peilin Ye', Maarten Lankhorst, Maxime Ripard, Thomas Zimmermann
  Cc: David Airlie, Daniel Vetter, Dan Carpenter, Arnd Bergmann,
	Greg Kroah-Hartman, linux-kernel-mentees, dri-devel,
	linux-kernel

From: Peilin Ye
> Sent: 28 July 2020 12:52
> Currently `struct drm_buf_desc` is defined as follows:
> 
> struct drm_buf_desc {
> 	int count;
> 	int size;
> 	int low_mark;
> 	int high_mark;
> 	enum {
> 		_DRM_PAGE_ALIGN = 0x01,
> 		_DRM_AGP_BUFFER = 0x02,
> 		_DRM_SG_BUFFER = 0x04,
> 		_DRM_FB_BUFFER = 0x08,
> 		_DRM_PCI_BUFFER_RO = 0x10
> 	} flags;
> 	unsigned long agp_start;
> };
> 
> copy_one_buf() is potentially copying uninitialized kernel stack memory
> to userspace, since the compiler may leave such "holes" (around `.flags`
> and `.agp_start` fields) in this statically allocated structure. Prevent
> it by initializing `v` with memset().
> 
> Cc: stable@vger.kernel.org
> Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
> Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
> Acked-by: Arnd Bergmann <arnd@arndb.de>
> Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
> ---
> Change in v2:
>     - Improve commit description. (Suggested by Arnd Bergmann
>       <arnd@arndb.de>)
> 
>  drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
> index a0735fbc144b..f99cd4a3f951 100644
> --- a/drivers/gpu/drm/drm_bufs.c
> +++ b/drivers/gpu/drm/drm_bufs.c
> @@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
>  {
>  	struct drm_buf_info *request = data;
>  	struct drm_buf_desc __user *to = &request->list[count];
> -	struct drm_buf_desc v = {.count = from->buf_count,
> -				 .size = from->buf_size,
> -				 .low_mark = from->low_mark,
> -				 .high_mark = from->high_mark};
> +	struct drm_buf_desc v;
> +
> +	memset(&v, 0, sizeof(v));
> +
> +	v.count = from->buf_count;
> +	v.size = from->buf_size;
> +	v.low_mark = from->low_mark;
> +	v.high_mark = from->high_mark;
> 
>  	if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
>  		return -EFAULT;

The memset() isn't needed.
The copy_to_user() stops after the 4 'int' values so no 'random'
kernel stack can get copied.

Quite why it is 'right' to leave the remaining part of each
userspace structure unchanged is another matter.

	David.

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)


^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [Linux-kernel-mentees] [PATCH v2] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
@ 2020-07-28 12:35     ` David Laight
  0 siblings, 0 replies; 18+ messages in thread
From: David Laight @ 2020-07-28 12:35 UTC (permalink / raw)
  To: 'Peilin Ye', Maarten Lankhorst, Maxime Ripard, Thomas Zimmermann
  Cc: Arnd Bergmann, David Airlie, linux-kernel, dri-devel,
	Daniel Vetter, linux-kernel-mentees, Dan Carpenter

From: Peilin Ye
> Sent: 28 July 2020 12:52
> Currently `struct drm_buf_desc` is defined as follows:
> 
> struct drm_buf_desc {
> 	int count;
> 	int size;
> 	int low_mark;
> 	int high_mark;
> 	enum {
> 		_DRM_PAGE_ALIGN = 0x01,
> 		_DRM_AGP_BUFFER = 0x02,
> 		_DRM_SG_BUFFER = 0x04,
> 		_DRM_FB_BUFFER = 0x08,
> 		_DRM_PCI_BUFFER_RO = 0x10
> 	} flags;
> 	unsigned long agp_start;
> };
> 
> copy_one_buf() is potentially copying uninitialized kernel stack memory
> to userspace, since the compiler may leave such "holes" (around `.flags`
> and `.agp_start` fields) in this statically allocated structure. Prevent
> it by initializing `v` with memset().
> 
> Cc: stable@vger.kernel.org
> Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
> Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
> Acked-by: Arnd Bergmann <arnd@arndb.de>
> Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
> ---
> Change in v2:
>     - Improve commit description. (Suggested by Arnd Bergmann
>       <arnd@arndb.de>)
> 
>  drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
> index a0735fbc144b..f99cd4a3f951 100644
> --- a/drivers/gpu/drm/drm_bufs.c
> +++ b/drivers/gpu/drm/drm_bufs.c
> @@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
>  {
>  	struct drm_buf_info *request = data;
>  	struct drm_buf_desc __user *to = &request->list[count];
> -	struct drm_buf_desc v = {.count = from->buf_count,
> -				 .size = from->buf_size,
> -				 .low_mark = from->low_mark,
> -				 .high_mark = from->high_mark};
> +	struct drm_buf_desc v;
> +
> +	memset(&v, 0, sizeof(v));
> +
> +	v.count = from->buf_count;
> +	v.size = from->buf_size;
> +	v.low_mark = from->low_mark;
> +	v.high_mark = from->high_mark;
> 
>  	if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
>  		return -EFAULT;

The memset() isn't needed.
The copy_to_user() stops after the 4 'int' values so no 'random'
kernel stack can get copied.

Quite why it is 'right' to leave the remaining part of each
userspace structure unchanged is another matter.

	David.

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

_______________________________________________
Linux-kernel-mentees mailing list
Linux-kernel-mentees@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees

^ permalink raw reply	[flat|nested] 18+ messages in thread

* RE: [Linux-kernel-mentees] [PATCH v2] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
@ 2020-07-28 12:35     ` David Laight
  0 siblings, 0 replies; 18+ messages in thread
From: David Laight @ 2020-07-28 12:35 UTC (permalink / raw)
  To: 'Peilin Ye', Maarten Lankhorst, Maxime Ripard, Thomas Zimmermann
  Cc: Arnd Bergmann, David Airlie, Greg Kroah-Hartman, linux-kernel,
	dri-devel, linux-kernel-mentees, Dan Carpenter

From: Peilin Ye
> Sent: 28 July 2020 12:52
> Currently `struct drm_buf_desc` is defined as follows:
> 
> struct drm_buf_desc {
> 	int count;
> 	int size;
> 	int low_mark;
> 	int high_mark;
> 	enum {
> 		_DRM_PAGE_ALIGN = 0x01,
> 		_DRM_AGP_BUFFER = 0x02,
> 		_DRM_SG_BUFFER = 0x04,
> 		_DRM_FB_BUFFER = 0x08,
> 		_DRM_PCI_BUFFER_RO = 0x10
> 	} flags;
> 	unsigned long agp_start;
> };
> 
> copy_one_buf() is potentially copying uninitialized kernel stack memory
> to userspace, since the compiler may leave such "holes" (around `.flags`
> and `.agp_start` fields) in this statically allocated structure. Prevent
> it by initializing `v` with memset().
> 
> Cc: stable@vger.kernel.org
> Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
> Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
> Acked-by: Arnd Bergmann <arnd@arndb.de>
> Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
> ---
> Change in v2:
>     - Improve commit description. (Suggested by Arnd Bergmann
>       <arnd@arndb.de>)
> 
>  drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
> index a0735fbc144b..f99cd4a3f951 100644
> --- a/drivers/gpu/drm/drm_bufs.c
> +++ b/drivers/gpu/drm/drm_bufs.c
> @@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
>  {
>  	struct drm_buf_info *request = data;
>  	struct drm_buf_desc __user *to = &request->list[count];
> -	struct drm_buf_desc v = {.count = from->buf_count,
> -				 .size = from->buf_size,
> -				 .low_mark = from->low_mark,
> -				 .high_mark = from->high_mark};
> +	struct drm_buf_desc v;
> +
> +	memset(&v, 0, sizeof(v));
> +
> +	v.count = from->buf_count;
> +	v.size = from->buf_size;
> +	v.low_mark = from->low_mark;
> +	v.high_mark = from->high_mark;
> 
>  	if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
>  		return -EFAULT;

The memset() isn't needed.
The copy_to_user() stops after the 4 'int' values so no 'random'
kernel stack can get copied.

Quite why it is 'right' to leave the remaining part of each
userspace structure unchanged is another matter.

	David.

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [Linux-kernel-mentees] [PATCH v2] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
  2020-07-28 12:35     ` David Laight
  (?)
@ 2020-07-28 12:50       ` Peilin Ye
  -1 siblings, 0 replies; 18+ messages in thread
From: Peilin Ye @ 2020-07-28 12:50 UTC (permalink / raw)
  To: David Laight
  Cc: David Airlie, Daniel Vetter, Dan Carpenter, Arnd Bergmann,
	Greg Kroah-Hartman, linux-kernel-mentees, dri-devel,
	linux-kernel

On Tue, Jul 28, 2020 at 12:35:17PM +0000, David Laight wrote:
> From: Peilin Ye
> > Sent: 28 July 2020 12:52
> > Currently `struct drm_buf_desc` is defined as follows:
> > 
> > struct drm_buf_desc {
> > 	int count;
> > 	int size;
> > 	int low_mark;
> > 	int high_mark;
> > 	enum {
> > 		_DRM_PAGE_ALIGN = 0x01,
> > 		_DRM_AGP_BUFFER = 0x02,
> > 		_DRM_SG_BUFFER = 0x04,
> > 		_DRM_FB_BUFFER = 0x08,
> > 		_DRM_PCI_BUFFER_RO = 0x10
> > 	} flags;
> > 	unsigned long agp_start;
> > };
> > 
> > copy_one_buf() is potentially copying uninitialized kernel stack memory
> > to userspace, since the compiler may leave such "holes" (around `.flags`
> > and `.agp_start` fields) in this statically allocated structure. Prevent
> > it by initializing `v` with memset().
> > 
> > Cc: stable@vger.kernel.org
> > Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
> > Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
> > Acked-by: Arnd Bergmann <arnd@arndb.de>
> > Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
> > ---
> > Change in v2:
> >     - Improve commit description. (Suggested by Arnd Bergmann
> >       <arnd@arndb.de>)
> > 
> >  drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
> >  1 file changed, 8 insertions(+), 4 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
> > index a0735fbc144b..f99cd4a3f951 100644
> > --- a/drivers/gpu/drm/drm_bufs.c
> > +++ b/drivers/gpu/drm/drm_bufs.c
> > @@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
> >  {
> >  	struct drm_buf_info *request = data;
> >  	struct drm_buf_desc __user *to = &request->list[count];
> > -	struct drm_buf_desc v = {.count = from->buf_count,
> > -				 .size = from->buf_size,
> > -				 .low_mark = from->low_mark,
> > -				 .high_mark = from->high_mark};
> > +	struct drm_buf_desc v;
> > +
> > +	memset(&v, 0, sizeof(v));
> > +
> > +	v.count = from->buf_count;
> > +	v.size = from->buf_size;
> > +	v.low_mark = from->low_mark;
> > +	v.high_mark = from->high_mark;
> > 
> >  	if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
> >  		return -EFAULT;
> 
> The memset() isn't needed.
> The copy_to_user() stops after the 4 'int' values so no 'random'
> kernel stack can get copied.

You are right, I overlooked that. Thank you for pointing this out!

Peilin Ye

> Quite why it is 'right' to leave the remaining part of each
> userspace structure unchanged is another matter.
> 
> 	David.
> 
> -
> Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
> Registration No: 1397386 (Wales)
> 

^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [Linux-kernel-mentees] [PATCH v2] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
@ 2020-07-28 12:50       ` Peilin Ye
  0 siblings, 0 replies; 18+ messages in thread
From: Peilin Ye @ 2020-07-28 12:50 UTC (permalink / raw)
  To: David Laight
  Cc: Arnd Bergmann, David Airlie, linux-kernel, dri-devel,
	Daniel Vetter, linux-kernel-mentees, Dan Carpenter

On Tue, Jul 28, 2020 at 12:35:17PM +0000, David Laight wrote:
> From: Peilin Ye
> > Sent: 28 July 2020 12:52
> > Currently `struct drm_buf_desc` is defined as follows:
> > 
> > struct drm_buf_desc {
> > 	int count;
> > 	int size;
> > 	int low_mark;
> > 	int high_mark;
> > 	enum {
> > 		_DRM_PAGE_ALIGN = 0x01,
> > 		_DRM_AGP_BUFFER = 0x02,
> > 		_DRM_SG_BUFFER = 0x04,
> > 		_DRM_FB_BUFFER = 0x08,
> > 		_DRM_PCI_BUFFER_RO = 0x10
> > 	} flags;
> > 	unsigned long agp_start;
> > };
> > 
> > copy_one_buf() is potentially copying uninitialized kernel stack memory
> > to userspace, since the compiler may leave such "holes" (around `.flags`
> > and `.agp_start` fields) in this statically allocated structure. Prevent
> > it by initializing `v` with memset().
> > 
> > Cc: stable@vger.kernel.org
> > Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
> > Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
> > Acked-by: Arnd Bergmann <arnd@arndb.de>
> > Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
> > ---
> > Change in v2:
> >     - Improve commit description. (Suggested by Arnd Bergmann
> >       <arnd@arndb.de>)
> > 
> >  drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
> >  1 file changed, 8 insertions(+), 4 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
> > index a0735fbc144b..f99cd4a3f951 100644
> > --- a/drivers/gpu/drm/drm_bufs.c
> > +++ b/drivers/gpu/drm/drm_bufs.c
> > @@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
> >  {
> >  	struct drm_buf_info *request = data;
> >  	struct drm_buf_desc __user *to = &request->list[count];
> > -	struct drm_buf_desc v = {.count = from->buf_count,
> > -				 .size = from->buf_size,
> > -				 .low_mark = from->low_mark,
> > -				 .high_mark = from->high_mark};
> > +	struct drm_buf_desc v;
> > +
> > +	memset(&v, 0, sizeof(v));
> > +
> > +	v.count = from->buf_count;
> > +	v.size = from->buf_size;
> > +	v.low_mark = from->low_mark;
> > +	v.high_mark = from->high_mark;
> > 
> >  	if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
> >  		return -EFAULT;
> 
> The memset() isn't needed.
> The copy_to_user() stops after the 4 'int' values so no 'random'
> kernel stack can get copied.

You are right, I overlooked that. Thank you for pointing this out!

Peilin Ye

> Quite why it is 'right' to leave the remaining part of each
> userspace structure unchanged is another matter.
> 
> 	David.
> 
> -
> Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
> Registration No: 1397386 (Wales)
> 
_______________________________________________
Linux-kernel-mentees mailing list
Linux-kernel-mentees@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees

^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [Linux-kernel-mentees] [PATCH v2] drm/bufs: Prevent kernel-infoleak in copy_one_buf()
@ 2020-07-28 12:50       ` Peilin Ye
  0 siblings, 0 replies; 18+ messages in thread
From: Peilin Ye @ 2020-07-28 12:50 UTC (permalink / raw)
  To: David Laight
  Cc: Arnd Bergmann, David Airlie, Greg Kroah-Hartman, linux-kernel,
	dri-devel, linux-kernel-mentees, Dan Carpenter

On Tue, Jul 28, 2020 at 12:35:17PM +0000, David Laight wrote:
> From: Peilin Ye
> > Sent: 28 July 2020 12:52
> > Currently `struct drm_buf_desc` is defined as follows:
> > 
> > struct drm_buf_desc {
> > 	int count;
> > 	int size;
> > 	int low_mark;
> > 	int high_mark;
> > 	enum {
> > 		_DRM_PAGE_ALIGN = 0x01,
> > 		_DRM_AGP_BUFFER = 0x02,
> > 		_DRM_SG_BUFFER = 0x04,
> > 		_DRM_FB_BUFFER = 0x08,
> > 		_DRM_PCI_BUFFER_RO = 0x10
> > 	} flags;
> > 	unsigned long agp_start;
> > };
> > 
> > copy_one_buf() is potentially copying uninitialized kernel stack memory
> > to userspace, since the compiler may leave such "holes" (around `.flags`
> > and `.agp_start` fields) in this statically allocated structure. Prevent
> > it by initializing `v` with memset().
> > 
> > Cc: stable@vger.kernel.org
> > Fixes: 5c7640ab6258 ("switch compat_drm_infobufs() to drm_ioctl_kernel()")
> > Suggested-by: Dan Carpenter <dan.carpenter@oracle.com>
> > Acked-by: Arnd Bergmann <arnd@arndb.de>
> > Signed-off-by: Peilin Ye <yepeilin.cs@gmail.com>
> > ---
> > Change in v2:
> >     - Improve commit description. (Suggested by Arnd Bergmann
> >       <arnd@arndb.de>)
> > 
> >  drivers/gpu/drm/drm_bufs.c | 12 ++++++++----
> >  1 file changed, 8 insertions(+), 4 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
> > index a0735fbc144b..f99cd4a3f951 100644
> > --- a/drivers/gpu/drm/drm_bufs.c
> > +++ b/drivers/gpu/drm/drm_bufs.c
> > @@ -1349,10 +1349,14 @@ static int copy_one_buf(void *data, int count, struct drm_buf_entry *from)
> >  {
> >  	struct drm_buf_info *request = data;
> >  	struct drm_buf_desc __user *to = &request->list[count];
> > -	struct drm_buf_desc v = {.count = from->buf_count,
> > -				 .size = from->buf_size,
> > -				 .low_mark = from->low_mark,
> > -				 .high_mark = from->high_mark};
> > +	struct drm_buf_desc v;
> > +
> > +	memset(&v, 0, sizeof(v));
> > +
> > +	v.count = from->buf_count;
> > +	v.size = from->buf_size;
> > +	v.low_mark = from->low_mark;
> > +	v.high_mark = from->high_mark;
> > 
> >  	if (copy_to_user(to, &v, offsetof(struct drm_buf_desc, flags)))
> >  		return -EFAULT;
> 
> The memset() isn't needed.
> The copy_to_user() stops after the 4 'int' values so no 'random'
> kernel stack can get copied.

You are right, I overlooked that. Thank you for pointing this out!

Peilin Ye

> Quite why it is 'right' to leave the remaining part of each
> userspace structure unchanged is another matter.
> 
> 	David.
> 
> -
> Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
> Registration No: 1397386 (Wales)
> 
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

^ permalink raw reply	[flat|nested] 18+ messages in thread

end of thread, other threads:[~2020-07-28 19:31 UTC | newest]

Thread overview: 18+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-07-28  1:43 [Linux-kernel-mentees] [PATCH] drm/bufs: Prevent kernel-infoleak in copy_one_buf() Peilin Ye
2020-07-28  1:43 ` Peilin Ye
2020-07-28  1:43 ` Peilin Ye
2020-07-28  8:11 ` Arnd Bergmann
2020-07-28  8:11   ` Arnd Bergmann
2020-07-28  8:11   ` Arnd Bergmann
2020-07-28 11:15   ` Peilin Ye
2020-07-28 11:15     ` Peilin Ye
2020-07-28 11:15     ` Peilin Ye
2020-07-28 11:52 ` [Linux-kernel-mentees] [PATCH v2] " Peilin Ye
2020-07-28 11:52   ` Peilin Ye
2020-07-28 11:52   ` Peilin Ye
2020-07-28 12:35   ` David Laight
2020-07-28 12:35     ` David Laight
2020-07-28 12:35     ` David Laight
2020-07-28 12:50     ` Peilin Ye
2020-07-28 12:50       ` Peilin Ye
2020-07-28 12:50       ` Peilin Ye

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.