Re: [PATCH v3] media: meson: vdec: Fix a NULL pointer dereference in amvdec_add_ts()

From: Neil Armstrong
Date: Wed Jan 12 2022 - 03:57:23 EST


Hi,

On 11/01/2022 14:03, Qingyang Zhou wrote:
> Hi Hans:
>
> Here I certify that this is an origin patch I wrote. I did not copy Neil's patch, and I did not know his patch.
>
> Neil, please post your original patch here, so everyone can see both patches.
>
> Yours 
> Zhou Qingyang.
>
> On Tue, Jan 11, 2022 at 5:16 PM Hans Verkuil <hverkuil-cisco@xxxxxxxxx <mailto:hverkuil-cisco@xxxxxxxxx>> wrote:
>
> Zhou Qingyang, this is exactly the patch Neil wrote, except you just stuck your
> name on it. Not nice.
>
> Neil, can you post your patch with your own Signed-off-by, then I'll take that one.
>
> Regards,
>
>         Hans
>
> On 15/12/2021 04:35, Zhou Qingyang wrote:
> > In amvdec_add_ts(), there is a dereference of kzalloc(), which could lead
> > to a NULL pointer dereference on failure of kzalloc().
> >
> > Fix this bug by adding a NULL check of new_ts.
> >
> > This bug was found by a static analyzer[1].
> >
> > Builds with CONFIG_VIDEO_MESON_VDEC=m show no new warnings,
> > and our static analyzer no longer warns about this code.
> >
> > Fixes: 876f123b8956 ("media: meson: vdec: bring up to compliance")
> > Signed-off-by: Zhou Qingyang <zhou1615@xxxxxxx <mailto:zhou1615@xxxxxxx>>
> > ---
> >
> > [1] The analysis employs differential checking to identify inconsistent
> > security operations (e.g., checks or kfrees) between two code paths and
> > confirms that the inconsistent operations are not recovered in the
> > current function or the callers, so they constitute bugs.
> >
> > Note that, as a bug found by static analysis, it can be a false
> > positive or hard to trigger. Multiple researchers have cross-reviewed
> > the bug.
> >
> > Changes in v3:
> >   -  Change the description of patch
> >   -  Turn the return type from 'void' to 'int'
> >   -  Check the return value in the caller 'esparser_queue()'
> >
> > Changes in v2:
> >   -  Delete dev_err() message
> >
> >  drivers/staging/media/meson/vdec/esparser.c     | 7 ++++++-
> >  drivers/staging/media/meson/vdec/vdec_helpers.c | 8 ++++++--
> >  drivers/staging/media/meson/vdec/vdec_helpers.h | 4 ++--
> >  3 files changed, 14 insertions(+), 5 deletions(-)
> >
> > diff --git a/drivers/staging/media/meson/vdec/esparser.c b/drivers/staging/media/meson/vdec/esparser.c
> > index db7022707ff8..095100a50da8 100644
> > --- a/drivers/staging/media/meson/vdec/esparser.c
> > +++ b/drivers/staging/media/meson/vdec/esparser.c
> > @@ -328,7 +328,12 @@ esparser_queue(struct amvdec_session *sess, struct vb2_v4l2_buffer *vbuf)
>
> >       offset = esparser_get_offset(sess);
>
> > -     amvdec_add_ts(sess, vb->timestamp, vbuf->timecode, offset, vbuf->flags);
> > +     ret = amvdec_add_ts(sess, vb->timestamp, vbuf->timecode, offset, vbuf->flags);
> > +     if (!ret) {
> > +             v4l2_m2m_buf_done(vbuf, VB2_BUF_STATE_ERROR);
> > +             return ret;
> > +     }
> > +
> >       dev_dbg(core->dev, "esparser: ts = %llu pld_size = %u offset = %08X flags = %08X\n",
> >               vb->timestamp, payload_size, offset, vbuf->flags);
>
> > diff --git a/drivers/staging/media/meson/vdec/vdec_helpers.c b/drivers/staging/media/meson/vdec/vdec_helpers.c
> > index b9125c295d1d..06fd66539797 100644
> > --- a/drivers/staging/media/meson/vdec/vdec_helpers.c
> > +++ b/drivers/staging/media/meson/vdec/vdec_helpers.c
> > @@ -227,13 +227,16 @@ int amvdec_set_canvases(struct amvdec_session *sess,
> >  }
> >  EXPORT_SYMBOL_GPL(amvdec_set_canvases);
>
> > -void amvdec_add_ts(struct amvdec_session *sess, u64 ts,
> > -                struct v4l2_timecode tc, u32 offset, u32 vbuf_flags)
> > +int amvdec_add_ts(struct amvdec_session *sess, u64 ts,
> > +               struct v4l2_timecode tc, u32 offset, u32 vbuf_flags)
> >  {
> >       struct amvdec_timestamp *new_ts;
> >       unsigned long flags;
>
> >       new_ts = kzalloc(sizeof(*new_ts), GFP_KERNEL);
> > +     if (!new_ts)
> > +             return -ENOMEM;
> > +
> >       new_ts->ts = ts;
> >       new_ts->tc = tc;
> >       new_ts->offset = offset;
> > @@ -242,6 +245,7 @@ void amvdec_add_ts(struct amvdec_session *sess, u64 ts,
> >       spin_lock_irqsave(&sess->ts_spinlock, flags);
> >       list_add_tail(&new_ts->list, &sess->timestamps);
> >       spin_unlock_irqrestore(&sess->ts_spinlock, flags);
> > +     return 0;
> >  }
> >  EXPORT_SYMBOL_GPL(amvdec_add_ts);
>
> > diff --git a/drivers/staging/media/meson/vdec/vdec_helpers.h b/drivers/staging/media/meson/vdec/vdec_helpers.h
> > index 88137d15aa3a..4bf3e61d081b 100644
> > --- a/drivers/staging/media/meson/vdec/vdec_helpers.h
> > +++ b/drivers/staging/media/meson/vdec/vdec_helpers.h
> > @@ -56,8 +56,8 @@ void amvdec_dst_buf_done_offset(struct amvdec_session *sess,
> >   * @offset: offset in the VIFIFO where the associated packet was written
> >   * @flags: the vb2_v4l2_buffer flags
> >   */
> > -void amvdec_add_ts(struct amvdec_session *sess, u64 ts,
> > -                struct v4l2_timecode tc, u32 offset, u32 flags);
> > +int amvdec_add_ts(struct amvdec_session *sess, u64 ts,
> > +               struct v4l2_timecode tc, u32 offset, u32 flags);
> >  void amvdec_remove_ts(struct amvdec_session *sess, u64 ts);
>
> >  /**
>

Let's inspect:

I posted a probable solution the 09/12/2021 in a reply to 20211209085840.2081024-1-jiasheng@xxxxxxxxxxx :

===================><=============================
diff --git a/drivers/staging/media/meson/vdec/esparser.c b/drivers/staging/media/meson/vdec/esparser.c
index db7022707ff8..df733eff9ac7 100644
--- a/drivers/staging/media/meson/vdec/esparser.c
+++ b/drivers/staging/media/meson/vdec/esparser.c
@@ -328,7 +328,11 @@ esparser_queue(struct amvdec_session *sess, struct vb2_v4l2_buffer *vbuf)

offset = esparser_get_offset(sess);

- amvdec_add_ts(sess, vb->timestamp, vbuf->timecode, offset, vbuf->flags);
+ ret = amvdec_add_ts(sess, vb->timestamp, vbuf->timecode, offset, vbuf->flags);
+ if (ret) {
+ v4l2_m2m_buf_done(vbuf, VB2_BUF_STATE_ERROR);
+ return ret;
+ }
dev_dbg(core->dev, "esparser: ts = %llu pld_size = %u offset = %08X flags = %08X\n",
vb->timestamp, payload_size, offset, vbuf->flags);

diff --git a/drivers/staging/media/meson/vdec/vdec_helpers.c b/drivers/staging/media/meson/vdec/vdec_helpers.c
index b9125c295d1d..593b2ccbece2 100644
--- a/drivers/staging/media/meson/vdec/vdec_helpers.c
+++ b/drivers/staging/media/meson/vdec/vdec_helpers.c
@@ -227,13 +227,15 @@ int amvdec_set_canvases(struct amvdec_session *sess,
}
EXPORT_SYMBOL_GPL(amvdec_set_canvases);

-void amvdec_add_ts(struct amvdec_session *sess, u64 ts,
- struct v4l2_timecode tc, u32 offset, u32 vbuf_flags)
+int amvdec_add_ts(struct amvdec_session *sess, u64 ts,
+ struct v4l2_timecode tc, u32 offset, u32 vbuf_flags)
{
struct amvdec_timestamp *new_ts;
unsigned long flags;

new_ts = kzalloc(sizeof(*new_ts), GFP_KERNEL);
+ if (!new_ts)
+ return -ENOMEM;
new_ts->ts = ts;
new_ts->tc = tc;
new_ts->offset = offset;
diff --git a/drivers/staging/media/meson/vdec/vdec_helpers.h b/drivers/staging/media/meson/vdec/vdec_helpers.h
index 88137d15aa3a..4bf3e61d081b 100644
--- a/drivers/staging/media/meson/vdec/vdec_helpers.h
+++ b/drivers/staging/media/meson/vdec/vdec_helpers.h
@@ -56,8 +56,8 @@ void amvdec_dst_buf_done_offset(struct amvdec_session *sess,
* @offset: offset in the VIFIFO where the associated packet was written
* @flags: the vb2_v4l2_buffer flags
*/
-void amvdec_add_ts(struct amvdec_session *sess, u64 ts,
- struct v4l2_timecode tc, u32 offset, u32 flags);
+int amvdec_add_ts(struct amvdec_session *sess, u64 ts,
+ struct v4l2_timecode tc, u32 offset, u32 flags);
void amvdec_remove_ts(struct amvdec_session *sess, u64 ts);
===================><=============================


Then the patches were sent:
- 20211210015620.2143555-1-jiasheng@xxxxxxxxxxx on 10/12/2021
- 20211215033535.40422-1-zhou1615@xxxxxxx on 15/12/2021

They are extremely close but not similar, mostly indenting differs.

Both patches have the missing final "return 0" in amvdec_add_ts which is missing in my proposal.

But only 20211210015620.2143555-1-jiasheng@xxxxxxxxxxx has the correct "if (ret)" in esparser_queue().
Patch 20211215033535.40422-1-zhou1615@xxxxxx has a wrong "if (!ret)".

But when comparing, 20211215033535.40422-1-zhou1615@xxxxxxx is an almost exact copy of my proposal, minus the fixes and the bogus return check.

To be honest, there is a limited way to fix this, it's probable 20211215033535.40422-1-zhou1615@xxxxxxx was written independently from
my proposal since there is a bug return check, and 20211210015620.2143555-1-jiasheng@xxxxxxxxxxx was rewritten from my proposal.

Since 20211215033535.40422-1-zhou1615@xxxxxx has a bogus return check, it should be naked.

I'll only ask 20211210015620.2143555-1-jiasheng@xxxxxxxxxxx to be resent with a "Suggested-by" and indentation fixed like my proposal.

Neil