Hi Mauro,
On Friday 19 April 2013 08:18:01 Mauro Carvalho Chehab wrote:
> Em Fri, 19 Apr 2013 15:16:56 +0530 Prabhakar lad escreveu:
> > From: Lad, Prabhakar <prabhakar.csengg@xxxxxxxxx>
> >
> > From commit 068a0df76023926af958a336a78bef60468d2033
> > "[media] media: vb2: add length check for mmap"
> > patch verifies that the mmap() size requested by userspace
> > doesn't exceed the buffer size.
> >
> > As the mmap() size is rounded up to the next page boundary
> > the check will fail for buffer sizes that are not multiple
> > of the page size.
> >
> > This patch fixes the check by aligning the buffer size to page
> > size during the check. Alongside fixes the vmalloc allocator
> > to round up the size.
> >
> > Signed-off-by: Lad, Prabhakar <prabhakar.csengg@xxxxxxxxx>
> > Cc: Laurent Pinchart <laurent.pinchart@xxxxxxxxxxxxxxxx>
> > Cc: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>
> > Cc: Seung-Woo Kim <sw0312.kim@xxxxxxxxxxx>
> > Cc: Hans Verkuil <hans.verkuil@xxxxxxxxx>
> > Cc: Mauro Carvalho Chehab <mchehab@xxxxxxxxxx>
> > ---
> >
> > drivers/media/v4l2-core/videobuf2-core.c | 2 +-
> > drivers/media/v4l2-core/videobuf2-vmalloc.c | 2 +-
> > 2 files changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/media/v4l2-core/videobuf2-core.c
> > b/drivers/media/v4l2-core/videobuf2-core.c index 58c1744..223fcd4 100644
> > --- a/drivers/media/v4l2-core/videobuf2-core.c
> > +++ b/drivers/media/v4l2-core/videobuf2-core.c
> > @@ -1886,7 +1886,7 @@ int vb2_mmap(struct vb2_queue *q, struct
> > vm_area_struct *vma)>
> > vb = q->bufs[buffer];
> >
> > - if (vb->v4l2_planes[plane].length < (vma->vm_end - vma->vm_start)) {
> > + if (PAGE_ALIGN(vb->v4l2_planes[plane].length) < (vma->vm_end -
> > vma->vm_start)) {>
> > dprintk(1, "Invalid length\n");
> > return -EINVAL;
> >
> > }
>
> That is tricky, as it assumes that vb->v4l2_planes[plane].length was round
> up to PAGE_SIZE at each memops driver, but the vb2 core doesn't enforce it.
>
> IMO, it would be cleaner to round vb->v4l2_planes[plane].length up
> at VB2 core, before calling the memops alloc functions at the drivers.
I don't think we should round vb->v4l2_planes[plane].length up. That variable
stores the buffer length required by the driver, and will be used to perform
size checks when importing a dmabuf buffer. We don't want to prevent a buffer
large enough for the driver but not page size aligned to be imported.
What we could do is round in the core the size passed to the alloc function,
without storing the rounded value in vb->v4l2_planes[plane].length.
And, reading down, I realize that this is exactly what you meant :-) The
proposed patch looks good to me.
> Also, VB2 is already complex enough to put it there without proper
> comments (and there's a minor codingstyle issue there: line is bigger
> than 80 cols).
A comment is definitely a good idea.
> > diff --git a/drivers/media/v4l2-core/videobuf2-vmalloc.c
> > b/drivers/media/v4l2-core/videobuf2-vmalloc.c index 313d977..bf3b95c
> > 100644
> > --- a/drivers/media/v4l2-core/videobuf2-vmalloc.c
> > +++ b/drivers/media/v4l2-core/videobuf2-vmalloc.c
> > @@ -44,7 +44,7 @@ static void *vb2_vmalloc_alloc(void *alloc_ctx, unsigned
> > long size, gfp_t gfp_fl>
> > return NULL;
> >
> > buf->size = size;
> >
> > - buf->vaddr = vmalloc_user(buf->size);
> > + buf->vaddr = vmalloc_user(PAGE_ALIGN(buf->size));
>
> See? You needed to put an alignment here as well, not because vmalloc
> needs it, but because this is needed by VB2 core.
>
> Also, on the other drivers, buf->size is stored page aligned, while
> here, you're doing different, without any documented reason for doing
> that, instead of doing the same as on the other memops drivers.
>
> That mistake reflects, for example, when the driver prints the failure:
>
> if (!buf->vaddr) {
> pr_debug("vmalloc of size %ld failed\n", buf->size);
>
> as it will show a different size than what you actually required.
> As those memory starving errors can also produce a dump at the mm
> core, the size there won't match the size on the above printed message.
>
> Also, it is a very bad idea to delegate the core's requirement of
> do page alignment from the core to the memops drivers, as other
> patches may change the logic there, or a new memops could be added,
> and the same problem will hit again (and unnoticed, as the check
> routine do page alignments).
Agreed. The memory allocator shouldn't need to guess the core requirements.
> > buf->handler.refcount = &buf->refcount;
> > buf->handler.put = vb2_vmalloc_put;
> > buf->handler.arg = buf;
>
> IMO, a cleaner version would be the following (untested) code.
>
> -
>
> [media] videobuf2: fix the length check for mmap
>
> Memory maps typically require that the buffer size to be page
> aligned. Currently, two memops drivers do such alignment
> internally, but videobuf-vmalloc doesn't.
>
> Also, the buffer overflow check doesn't take it into account.
>
> So, instead of doing it at each memops driver, enforce it at
> VB2 core.
>
> Reported-by: Prabhakar lad <prabhakar.csengg@xxxxxxxxx>
> Signed-off-by: Mauro Carvalho Chehab <mchehab@xxxxxxxxxx>
>
> diff --git a/drivers/media/v4l2-core/videobuf2-core.c
> b/drivers/media/v4l2-core/videobuf2-core.c index 58c1744..7d833ee 100644
> --- a/drivers/media/v4l2-core/videobuf2-core.c
> +++ b/drivers/media/v4l2-core/videobuf2-core.c
> @@ -54,10 +54,15 @@ static int __vb2_buf_mem_alloc(struct vb2_buffer *vb)
> void *mem_priv;
> int plane;
>
> - /* Allocate memory for all planes in this buffer */
> + /*
> + * Allocate memory for all planes in this buffer
> + * NOTE: mmapped areas should be page aligned
> + */
> for (plane = 0; plane < vb->num_planes; ++plane) {
> + unsigned long size = PAGE_ALIGN(q->plane_sizes[plane]);
> +
> mem_priv = call_memop(q, alloc, q->alloc_ctx[plane],
> - q->plane_sizes[plane], q->gfp_flags);
> + size, q->gfp_flags);
> if (IS_ERR_OR_NULL(mem_priv))
> goto free;
>
> @@ -1852,6 +1857,7 @@ int vb2_mmap(struct vb2_queue *q, struct
> vm_area_struct *vma) struct vb2_buffer *vb;
> unsigned int buffer, plane;
> int ret;
> + unsigned long length;
>
> if (q->memory != V4L2_MEMORY_MMAP) {
> dprintk(1, "Queue is not currently set up for mmap\n");
> @@ -1886,8 +1892,15 @@ int vb2_mmap(struct vb2_queue *q, struct
> vm_area_struct *vma)
>
> vb = q->bufs[buffer];
>
> - if (vb->v4l2_planes[plane].length < (vma->vm_end - vma->vm_start)) {
> - dprintk(1, "Invalid length\n");
> + /*
> + * MMAP requires page_aligned buffers.
> + * The buffer length was page_aligned at __vb2_buf_mem_alloc(),
> + * so, we need to do the same here.
> + */
> + length = PAGE_ALIGN(vb->v4l2_planes[plane].length);
> + if (length < (vma->vm_end - vma->vm_start)) {
> + dprintk(1,
> + "MMAP invalid, as it would overflow buffer length\n");
> return -EINVAL;
> }
>
> diff --git a/drivers/media/v4l2-core/videobuf2-dma-contig.c
> b/drivers/media/v4l2-core/videobuf2-dma-contig.c index ae35d25..fd56f25
> 100644
> --- a/drivers/media/v4l2-core/videobuf2-dma-contig.c
> +++ b/drivers/media/v4l2-core/videobuf2-dma-contig.c
> @@ -162,9 +162,6 @@ static void *vb2_dc_alloc(void *alloc_ctx, unsigned long
> size, gfp_t gfp_flags) if (!buf)
> return ERR_PTR(-ENOMEM);
>
> - /* align image size to PAGE_SIZE */
> - size = PAGE_ALIGN(size);
> -
> buf->vaddr = dma_alloc_coherent(dev, size, &buf->dma_addr,
> GFP_KERNEL | gfp_flags);
> if (!buf->vaddr) {
> diff --git a/drivers/media/v4l2-core/videobuf2-dma-sg.c
> b/drivers/media/v4l2-core/videobuf2-dma-sg.c index 59522b2..16ae3dc 100644
> --- a/drivers/media/v4l2-core/videobuf2-dma-sg.c
> +++ b/drivers/media/v4l2-core/videobuf2-dma-sg.c
> @@ -55,7 +55,8 @@ static void *vb2_dma_sg_alloc(void *alloc_ctx, unsigned
> long size, gfp_t gfp_fla buf->write = 0;
> buf->offset = 0;
> buf->sg_desc.size = size;
> - buf->sg_desc.num_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
> + /* size is already page aligned */
> + buf->sg_desc.num_pages = size >> PAGE_SHIFT;
>
> buf->sg_desc.sglist = vzalloc(buf->sg_desc.num_pages *
> sizeof(*buf->sg_desc.sglist));
--
Regards,
Laurent Pinchart
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
