linux-kernel - Re: [PATCH v6 2/4] drm: Add API for capturing frame CRCs

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 8 Sep 2016 14:24:25 +0100
From:   Emil Velikov <emil.l.velikov@...il.com>
To:     Tomeu Vizoso <tomeu.vizoso@...labora.com>
Cc:     "Linux-Kernel@...r. Kernel. Org" <linux-kernel@...r.kernel.org>,
        linux-doc@...r.kernel.org, Jonathan Corbet <corbet@....net>,
        ML dri-devel <dri-devel@...ts.freedesktop.org>,
        Daniel Vetter <daniel.vetter@...el.com>,
        Emil Velikov <emil.velikov@...labora.com>
Subject: Re: [PATCH v6 2/4] drm: Add API for capturing frame CRCs

Hi Tomeu,

A couple of small nitpicks and a rather nasty looking bug, related to
your earlier question.

On 7 September 2016 at 11:27, Tomeu Vizoso <tomeu.vizoso@...labora.com> wrote:

> +static ssize_t crc_control_write(struct file *file, const char __user *ubuf,
> +                                size_t len, loff_t *offp)
> +{

> +       if (source[len] == '\n')
> +               source[len] = '\0';
> +
Considering the bug below, I'm considering if there's a case were we
don't want to explicitly set the terminating byte ?


> +/*
> + * 1 frame field of 10 chars plus a number of CRC fields of 10 chars each, space
> + * separated, with a newline at the end and null-terminated.
> + */
NULL-terminated what the things that was missing, explaining the
maths. Yet note the code sort of contradicts it.

TL;DR: above we conditionally NULL terminate the data, yet (below) we
feed garbage (always) instead of \0 byte.

> +#define LINE_LEN(values_cnt)   (10 + 11 * values_cnt + 1 + 1)
> +#define MAX_LINE_LEN           (LINE_LEN(DRM_MAX_CRC_NR))
> +
> +static ssize_t crtc_crc_read(struct file *filep, char __user *user_buf,
> +                            size_t count, loff_t *pos)
> +{
> +       struct drm_crtc *crtc = filep->f_inode->i_private;
> +       struct drm_crtc_crc *crc = &crtc->crc;
> +       struct drm_crtc_crc_entry *entry;
> +       char buf[MAX_LINE_LEN];
Here buf is filled with garbage...


> +       if (entry->has_frame_counter)
> +               sprintf(buf, "0x%08x", entry->frame);
> +       else
> +               sprintf(buf, "XXXXXXXXXX");
> +
> +       for (i = 0; i < crc->values_cnt; i++)
> +               sprintf(buf + 10 + i * 11, " 0x%08x", entry->crcs[i]);
> +       sprintf(buf + 10 + crc->values_cnt * 11, "\n");
> +
... and now all the data is in, incl. the \n byte.

> +       if (copy_to_user(user_buf, buf, LINE_LEN(crc->values_cnt)))
And here we copy the whole thing incl. the 'should be \0 but is
actually garbage' byte.

This doesn't look good :-\

> --- a/drivers/gpu/drm/drm_drv.c
> +++ b/drivers/gpu/drm/drm_drv.c

> @@ -221,6 +222,14 @@ static int drm_minor_register(struct drm_device *dev, unsigned int type)
>                 return ret;
>         }
>
> +       if (type == DRM_MINOR_PRIMARY) {
> +               drm_for_each_crtc(crtc, dev) {
> +                       ret = drm_debugfs_crtc_add(crtc);
> +                       if (ret)
> +                               DRM_ERROR("DRM: Failed to initialize CRC debugfs.\n");
> +               }
> +       }
> +
Minor: We're missing teardown in the error path.


> --- /dev/null
> +++ b/include/drm/drm_debugfs_crc.h

> +static inline int drm_debugfs_crtc_crc_add(struct drm_crtc *crtc)
Minor: The function is internal only and used only when
defined(CONFIG_DEBUG_FS). Thus it should stay in
drivers/gpu/drm/foo.h.

Rule of thumb: include/drm defines the API used by the drivers, while
drivers/gpu/drm/foo_internal.h the internal API between the core DRM
modules.


Regards,
Emil