Skip to contentSkip to conversation

ipu3-imgu: Fix NULL pointer dereference in active selection access

PATCH In review Sakari Ailus 4 participants 7 messages Aug 25, 2022 – Sep 7, 2022

Subsystems (3)
lore
Share & tools
lore

Sakari Ailus

author maintainer
#0 lore raw

PATCH 1/1 ipu3-imgu: Fix NULL pointer dereference in active selection access

What the IMGU driver did was that it first acquired the pointers to active and try V4L2 subdev state, and only then figured out which one to use.

The problem with that approach and a later patch (see Fixes: tag) is that as sd_state argument to v4l2_subdev_get_try_crop() et al is NULL, there is now an attempt to dereference that.

Fix this.

Also rewrap lines a little.

Fixes 0d346d2a6f54 ("media: v4l2-subdev: add subdev-wide state struct")
Cc [email protected] # for v5.14 and later
---
 drivers/staging/media/ipu3/ipu3-v4l2.c | 31 ++++++++++++--------------
 1 file changed, 14 insertions(+), 17 deletions(-)
drivers/staging/media/ipu3/ipu3-v4l2.c +14 -17
index d1c539cefba87..2234bb8d48b34 100644
@@ -192,33 +192,30 @@ static int imgu_subdev_get_selection(struct v4l2_subdev *sd,
 				     struct v4l2_subdev_state *sd_state,
 				     struct v4l2_subdev_selection *sel)
 {
-	struct v4l2_rect *try_sel, *r;
-	struct imgu_v4l2_subdev *imgu_sd = container_of(sd,
-							struct imgu_v4l2_subdev,
-							subdev);
+	struct imgu_v4l2_subdev *imgu_sd =
+		container_of(sd, struct imgu_v4l2_subdev, subdev);
 
 	if (sel->pad != IMGU_NODE_IN)
 		return -EINVAL;
 
 	switch (sel->target) {
 	case V4L2_SEL_TGT_CROP:
-		try_sel = v4l2_subdev_get_try_crop(sd, sd_state, sel->pad);
-		r = &imgu_sd->rect.eff;
-		break;
+		if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
+			sel->r = *v4l2_subdev_get_try_crop(sd, sd_state,
+							   sel->pad);
+		else
+			sel->r = imgu_sd->rect.eff;
+		return 0;
 	case V4L2_SEL_TGT_COMPOSE:
-		try_sel = v4l2_subdev_get_try_compose(sd, sd_state, sel->pad);
-		r = &imgu_sd->rect.bds;
-		break;
+		if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
+			sel->r = *v4l2_subdev_get_try_compose(sd, sd_state,
+							      sel->pad);
+		else
+			sel->r = imgu_sd->rect.bds;
+		return 0;
 	default:
 		return -EINVAL;
 	}
-
-	if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
-		sel->r = *try_sel;
-	else
-		sel->r = *r;
-
-	return 0;
 }
 
 static int imgu_subdev_set_selection(struct v4l2_subdev *sd,
Signature · 1 line
2.30.2
Reviewed-by on v1 1/1
#1 lore raw

Sakari,

Thanks for your patch, does it make sense that prevent the NULL deference and throw warning in v4l2-subdev API? Though it is the responsibility of driver to keep it safe.

Quoted 71 lines · Sakari Ailus

What the IMGU driver did was that it first acquired the pointers to active and try V4L2 subdev state, and only then figured out which one to use.

The problem with that approach and a later patch (see Fixes: tag) is that as sd_state argument to v4l2_subdev_get_try_crop() et al is NULL, there is now an attempt to dereference that.

Fix this.

Also rewrap lines a little.

Fixes 0d346d2a6f54 ("media: v4l2-subdev: add subdev-wide state struct")
Cc [email protected] # for v5.14 and later
---
 drivers/staging/media/ipu3/ipu3-v4l2.c | 31 ++++++++++++--------------
 1 file changed, 14 insertions(+), 17 deletions(-)
drivers/staging/media/ipu3/ipu3-v4l2.c +14 -17
index d1c539cefba87..2234bb8d48b34 100644
@@ -192,33 +192,30 @@ static int imgu_subdev_get_selection(struct v4l2_subdev *sd,
 				     struct v4l2_subdev_state *sd_state,
 				     struct v4l2_subdev_selection *sel)
 {
-	struct v4l2_rect *try_sel, *r;
-	struct imgu_v4l2_subdev *imgu_sd = container_of(sd,
-							struct imgu_v4l2_subdev,
-							subdev);
+	struct imgu_v4l2_subdev *imgu_sd =
+		container_of(sd, struct imgu_v4l2_subdev, subdev);
 
 	if (sel->pad != IMGU_NODE_IN)
 		return -EINVAL;
 
 	switch (sel->target) {
 	case V4L2_SEL_TGT_CROP:
-		try_sel = v4l2_subdev_get_try_crop(sd, sd_state, sel->pad);
-		r = &imgu_sd->rect.eff;
-		break;
+		if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
+			sel->r = *v4l2_subdev_get_try_crop(sd, sd_state,
+							   sel->pad);
+		else
+			sel->r = imgu_sd->rect.eff;
+		return 0;
 	case V4L2_SEL_TGT_COMPOSE:
-		try_sel = v4l2_subdev_get_try_compose(sd, sd_state, sel->pad);
-		r = &imgu_sd->rect.bds;
-		break;
+		if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
+			sel->r = *v4l2_subdev_get_try_compose(sd, sd_state,
+							      sel->pad);
+		else
+			sel->r = imgu_sd->rect.bds;
+		return 0;
 	default:
 		return -EINVAL;
 	}
-
-	if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
-		sel->r = *try_sel;
-	else
-		sel->r = *r;
-
-	return 0;
 }
 
 static int imgu_subdev_set_selection(struct v4l2_subdev *sd,
Signature · 2 lines
Best regards,
Bingbu Cao
#2 lore raw
In reply to Sakari Ailus

Hi,

Quoted 8 lines · Sakari Ailus

What the IMGU driver did was that it first acquired the pointers to active and try V4L2 subdev state, and only then figured out which one to use.

The problem with that approach and a later patch (see Fixes: tag) is that as sd_state argument to v4l2_subdev_get_try_crop() et al is NULL, there is now an attempt to dereference that.

Fix this.

Thank you for this fix, this however only addresses imgu_subdev_get_selection(), but the issue is also present in imgu_subdev_set_selection(). I assume that a similar fix is needed for that. I've added a diff for that below. Feel free to squash that into your patch or let me know if I should submit this separately.

Regards,
Max

---
 drivers/staging/media/ipu3/ipu3-v4l2.c | 25 ++++++++++++-------------
 1 file changed, 12 insertions(+), 13 deletions(-)
drivers/staging/media/ipu3/ipu3-v4l2.c +12 -13
index 2234bb8d48b3..079f2635c70d 100644
@@ -223,10 +223,9 @@ static int imgu_subdev_set_selection(struct v4l2_subdev *sd,
                                     struct v4l2_subdev_selection *sel)
 {
        struct imgu_device *imgu = v4l2_get_subdevdata(sd);
-       struct imgu_v4l2_subdev *imgu_sd = container_of(sd,
-                                                       struct imgu_v4l2_subdev,
-                                                       subdev);
-       struct v4l2_rect *rect, *try_sel;
+       struct imgu_v4l2_subdev *imgu_sd =
+               container_of(sd, struct imgu_v4l2_subdev, subdev);
+       struct v4l2_rect *rect;
 
        dev_dbg(&imgu->pci_dev->dev,
                 "set subdev %u sel which %u target 0x%4x rect [%ux%u]",
@@ -238,22 +237,22 @@ static int imgu_subdev_set_selection(struct v4l2_subdev *sd,
 
        switch (sel->target) {
        case V4L2_SEL_TGT_CROP:
-               try_sel = v4l2_subdev_get_try_crop(sd, sd_state, sel->pad);
-               rect = &imgu_sd->rect.eff;
+               if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
+                       rect = v4l2_subdev_get_try_crop(sd, sd_state, sel->pad);
+               else
+                       rect = &imgu_sd->rect.eff;
                break;
        case V4L2_SEL_TGT_COMPOSE:
-               try_sel = v4l2_subdev_get_try_compose(sd, sd_state, sel->pad);
-               rect = &imgu_sd->rect.bds;
+               if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
+                       rect = v4l2_subdev_get_try_compose(sd, sd_state, sel->pad);
+               else
+                       rect = &imgu_sd->rect.bds;
                break;
        default:
                return -EINVAL;
        }
 
-       if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
-               *try_sel = sel->r;
-       else
-               *rect = sel->r;
-
+       *rect = sel->r;
        return 0;
 }
 

--
2.37.3

#3 lore raw

Hi Max,

Quoted 8 lines · Maximilian Luz
Quoted 7 lines · Sakari Ailus

What the IMGU driver did was that it first acquired the pointers to active and try V4L2 subdev state, and only then figured out which one to use.

The problem with that approach and a later patch (see Fixes: tag) is that as sd_state argument to v4l2_subdev_get_try_crop() et al is NULL, there is now an attempt to dereference that.

Quoted 7 lines
Quoted 1 line

Fix this.

Thank you for this fix, this however only addresses imgu_subdev_get_selection(), but the issue is also present in imgu_subdev_set_selection(). I assume that a similar fix is needed for that. I've added a diff for that below. Feel free to squash that into your patch or let me know if I should submit this separately.

This looks like a problem indeed. I'll let Sakari decide what he wants to do. Adding wrappers along the lines of

static struct v4l2_rect *
imgu_subdev_get_crop(struct imgu_v4l2_subdev *sd,
struct v4l2_subdev_state *sd_state, unsigned int pad, enum v4l2_subdev_format_whence which)

{
	if (which == V4L2_SUBDEV_FORMAT_TRY)
		return v4l2_subdev_get_try_crop(&imgu_sd->subdev, sd_state, pad);
	else
		return &imgu_sd->rect.eff;
}

(same for the selection rectangle) and using them through the code may help.

Quoted 54 lines
---
  drivers/staging/media/ipu3/ipu3-v4l2.c | 25 ++++++++++++-------------
  1 file changed, 12 insertions(+), 13 deletions(-)
drivers/staging/media/ipu3/ipu3-v4l2.c +12 -13
index 2234bb8d48b3..079f2635c70d 100644
@@ -223,10 +223,9 @@ static int imgu_subdev_set_selection(struct v4l2_subdev *sd,
                                      struct v4l2_subdev_selection *sel)
  {
         struct imgu_device *imgu = v4l2_get_subdevdata(sd);
-       struct imgu_v4l2_subdev *imgu_sd = container_of(sd,
-                                                       struct imgu_v4l2_subdev,
-                                                       subdev);
-       struct v4l2_rect *rect, *try_sel;
+       struct imgu_v4l2_subdev *imgu_sd =
+               container_of(sd, struct imgu_v4l2_subdev, subdev);
+       struct v4l2_rect *rect;
 
         dev_dbg(&imgu->pci_dev->dev,
                  "set subdev %u sel which %u target 0x%4x rect [%ux%u]",
@@ -238,22 +237,22 @@ static int imgu_subdev_set_selection(struct v4l2_subdev *sd,
 
         switch (sel->target) {
         case V4L2_SEL_TGT_CROP:
-               try_sel = v4l2_subdev_get_try_crop(sd, sd_state, sel->pad);
-               rect = &imgu_sd->rect.eff;
+               if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
+                       rect = v4l2_subdev_get_try_crop(sd, sd_state, sel->pad);
+               else
+                       rect = &imgu_sd->rect.eff;
                 break;
         case V4L2_SEL_TGT_COMPOSE:
-               try_sel = v4l2_subdev_get_try_compose(sd, sd_state, sel->pad);
-               rect = &imgu_sd->rect.bds;
+               if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
+                       rect = v4l2_subdev_get_try_compose(sd, sd_state, sel->pad);
+               else
+                       rect = &imgu_sd->rect.bds;
                 break;
         default:
                 return -EINVAL;
         }
 
-       if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
-               *try_sel = sel->r;
-       else
-               *rect = sel->r;
-
+       *rect = sel->r;
         return 0;
  }
 
Signature · 3 lines
Regards,

Laurent Pinchart
#4 lore raw

Hi,

Quoted 33 lines · Laurent Pinchart

Hi Max,

Quoted 8 lines · Maximilian Luz
Quoted 7 lines · Sakari Ailus

What the IMGU driver did was that it first acquired the pointers to active and try V4L2 subdev state, and only then figured out which one to use.

The problem with that approach and a later patch (see Fixes: tag) is that as sd_state argument to v4l2_subdev_get_try_crop() et al is NULL, there is now an attempt to dereference that.

Quoted 7 lines
Quoted 1 line

Fix this.

Thank you for this fix, this however only addresses imgu_subdev_get_selection(), but the issue is also present in imgu_subdev_set_selection(). I assume that a similar fix is needed for that. I've added a diff for that below. Feel free to squash that into your patch or let me know if I should submit this separately.

This looks like a problem indeed. I'll let Sakari decide what he wants to do. Adding wrappers along the lines of

static struct v4l2_rect *
imgu_subdev_get_crop(struct imgu_v4l2_subdev *sd,
struct v4l2_subdev_state *sd_state, unsigned int pad, enum v4l2_subdev_format_whence which)

{
	if (which == V4L2_SUBDEV_FORMAT_TRY)
		return v4l2_subdev_get_try_crop(&imgu_sd->subdev, sd_state, pad);
	else
		return &imgu_sd->rect.eff;
}
Quoted 2 lines

(same for the selection rectangle) and using them through the code may help.

Thanks! That does seem a good idea. I'll include those in the patch.

Regards,
Max

Quoted 56 lines
Quoted 55 lines
---
   drivers/staging/media/ipu3/ipu3-v4l2.c | 25 ++++++++++++-------------
   1 file changed, 12 insertions(+), 13 deletions(-)
drivers/staging/media/ipu3/ipu3-v4l2.c +12 -13
index 2234bb8d48b3..079f2635c70d 100644
@@ -223,10 +223,9 @@ static int imgu_subdev_set_selection(struct v4l2_subdev *sd,
                                       struct v4l2_subdev_selection *sel)
   {
          struct imgu_device *imgu = v4l2_get_subdevdata(sd);
-       struct imgu_v4l2_subdev *imgu_sd = container_of(sd,
-                                                       struct imgu_v4l2_subdev,
-                                                       subdev);
-       struct v4l2_rect *rect, *try_sel;
+       struct imgu_v4l2_subdev *imgu_sd =
+               container_of(sd, struct imgu_v4l2_subdev, subdev);
+       struct v4l2_rect *rect;
 
          dev_dbg(&imgu->pci_dev->dev,
                   "set subdev %u sel which %u target 0x%4x rect [%ux%u]",
@@ -238,22 +237,22 @@ static int imgu_subdev_set_selection(struct v4l2_subdev *sd,
 
          switch (sel->target) {
          case V4L2_SEL_TGT_CROP:
-               try_sel = v4l2_subdev_get_try_crop(sd, sd_state, sel->pad);
-               rect = &imgu_sd->rect.eff;
+               if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
+                       rect = v4l2_subdev_get_try_crop(sd, sd_state, sel->pad);
+               else
+                       rect = &imgu_sd->rect.eff;
                  break;
          case V4L2_SEL_TGT_COMPOSE:
-               try_sel = v4l2_subdev_get_try_compose(sd, sd_state, sel->pad);
-               rect = &imgu_sd->rect.bds;
+               if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
+                       rect = v4l2_subdev_get_try_compose(sd, sd_state, sel->pad);
+               else
+                       rect = &imgu_sd->rect.bds;
                  break;
          default:
                  return -EINVAL;
          }
 
-       if (sel->which == V4L2_SUBDEV_FORMAT_TRY)
-               *try_sel = sel->r;
-       else
-               *rect = sel->r;
-
+       *rect = sel->r;
          return 0;
   }
 
author maintainer
#5 lore raw
In reply to Maximilian Luz

Hi Maximilian,

Quoted 10 lines · Maximilian Luz

Hi,

Quoted 7 lines · Sakari Ailus

What the IMGU driver did was that it first acquired the pointers to active and try V4L2 subdev state, and only then figured out which one to use.

The problem with that approach and a later patch (see Fixes: tag) is that as sd_state argument to v4l2_subdev_get_try_crop() et al is NULL, there is now an attempt to dereference that.

Quoted 7 lines
Quoted 1 line

Fix this.

Thank you for this fix, this however only addresses imgu_subdev_get_selection(), but the issue is also present in imgu_subdev_set_selection(). I assume that a similar fix is needed for that. I've added a diff for that below. Feel free to squash that into your patch or let me know if I should submit this separately.

I've already sent a PR that includes it. Could you send this one as a patch, please?

Signature · 1 line
Sakari Ailus
#6 lore raw
Quoted 15 lines · Sakari Ailus

Hi Maximilian,

Quoted 12 lines · Maximilian Luz

Hi,

Quoted 8 lines · Sakari Ailus

What the IMGU driver did was that it first acquired the pointers to active and try V4L2 subdev state, and only then figured out which one to use.

The problem with that approach and a later patch (see Fixes: tag) is that as sd_state argument to v4l2_subdev_get_try_crop() et al is NULL, there is now an attempt to dereference that.

Fix this.

Quoted 8 lines
Quoted 5 lines

Thank you for this fix, this however only addresses imgu_subdev_get_selection(), but the issue is also present in imgu_subdev_set_selection(). I assume that a similar fix is needed for that. I've added a diff for that below. Feel free to squash that into your patch or let me know if I should submit this separately.

I've already sent a PR that includes it. Could you send this one as a patch, please?

Sure, I'll do that. I'll also follow Laurent's suggestions and introduce get_crop()/get_compose() helpers as I think that might be a bit cleaner.

Regards,
Max