[PATCH] drivers/base: Fix length checks in create_syslog_header()/dev_vprintk_emit()

[PATCH] drivers/base: Fix length checks in create_syslog_header()/dev_vprintk_emit()

[Ben Hutchings]

[-- Attachment #1: Type: text/plain, Size: 1552 bytes --]

snprintf() returns the number of bytes that could have been written
(excluding the null), not the actual number of bytes written.  Given a
long enough subsystem or device name, these functions will advance
beyond the end of the on-stack buffer in dev_vprintk_exit(), resulting
in an information leak or stack corruption.  I don't know whether such
a long name is currently possible.

In case snprintf() returns a value >= the buffer size, do not add
structured logging information.  Also WARN the first time this
happens, so we can fix the driver or increase the buffer size.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
---
 drivers/base/core.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/drivers/base/core.c b/drivers/base/core.c
index 67b180d..989a93c 100644
--- a/drivers/base/core.c
+++ b/drivers/base/core.c
@@ -2022,6 +2022,8 @@ create_syslog_header(const struct device *dev, char *hdr, size_t hdrlen)
 		return 0;
 
 	pos += snprintf(hdr + pos, hdrlen - pos, "SUBSYSTEM=%s", subsys);
+	if (pos >= hdrlen)
+		goto overflow;
 
 	/*
 	 * Add device identifier DEVICE=:
@@ -2053,7 +2055,14 @@ create_syslog_header(const struct device *dev, char *hdr, size_t hdrlen)
 				"DEVICE=+%s:%s", subsys, dev_name(dev));
 	}
 
+	if (pos >= hdrlen)
+		goto overflow;
+
 	return pos;
+
+overflow:
+	dev_WARN_ONCE(dev, 1, "device/subsystem name too long");
+	return 0;
 }
 EXPORT_SYMBOL(create_syslog_header);
 

-- 
Ben Hutchings
Computers are not intelligent.	They only think they are.

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 828 bytes --]

Re: [PATCH] drivers/base: Fix length checks in create_syslog_header()/dev_vprintk_emit()

[Greg Kroah-Hartman]

On Fri, Dec 27, 2013 at 06:18:18PM +0100, Ben Hutchings wrote:
> snprintf() returns the number of bytes that could have been written
> (excluding the null), not the actual number of bytes written.  Given a
> long enough subsystem or device name, these functions will advance
> beyond the end of the on-stack buffer in dev_vprintk_exit(), resulting
> in an information leak or stack corruption.  I don't know whether such
> a long name is currently possible.
> 
> In case snprintf() returns a value >= the buffer size, do not add
> structured logging information.  Also WARN the first time this
> happens, so we can fix the driver or increase the buffer size.
> 
> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
> ---
>  drivers/base/core.c | 9 +++++++++
>  1 file changed, 9 insertions(+)
> 
> diff --git a/drivers/base/core.c b/drivers/base/core.c
> index 67b180d..989a93c 100644
> --- a/drivers/base/core.c
> +++ b/drivers/base/core.c
> @@ -2022,6 +2022,8 @@ create_syslog_header(const struct device *dev, char *hdr, size_t hdrlen)
>  		return 0;
>  
>  	pos += snprintf(hdr + pos, hdrlen - pos, "SUBSYSTEM=%s", subsys);
> +	if (pos >= hdrlen)
> +		goto overflow;
>  
>  	/*
>  	 * Add device identifier DEVICE=:
> @@ -2053,7 +2055,14 @@ create_syslog_header(const struct device *dev, char *hdr, size_t hdrlen)
>  				"DEVICE=+%s:%s", subsys, dev_name(dev));
>  	}
>  
> +	if (pos >= hdrlen)
> +		goto overflow;
> +
>  	return pos;
> +
> +overflow:
> +	dev_WARN_ONCE(dev, 1, "device/subsystem name too long");

Why only warn once?  Any device/subsystem mix should be complained
about, if for only that we should be really annoying about it to get it
resolved.

thanks,

greg k-h

Re: [PATCH] drivers/base: Fix length checks in create_syslog_header()/dev_vprintk_emit()

[Ben Hutchings]

[-- Attachment #1: Type: text/plain, Size: 2082 bytes --]

On Fri, 2013-12-27 at 10:47 -0800, Greg Kroah-Hartman wrote:
> On Fri, Dec 27, 2013 at 06:18:18PM +0100, Ben Hutchings wrote:
> > snprintf() returns the number of bytes that could have been written
> > (excluding the null), not the actual number of bytes written.  Given a
> > long enough subsystem or device name, these functions will advance
> > beyond the end of the on-stack buffer in dev_vprintk_exit(), resulting
> > in an information leak or stack corruption.  I don't know whether such
> > a long name is currently possible.
> > 
> > In case snprintf() returns a value >= the buffer size, do not add
> > structured logging information.  Also WARN the first time this
> > happens, so we can fix the driver or increase the buffer size.
> > 
> > Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
> > ---
> >  drivers/base/core.c | 9 +++++++++
> >  1 file changed, 9 insertions(+)
> > 
> > diff --git a/drivers/base/core.c b/drivers/base/core.c
> > index 67b180d..989a93c 100644
> > --- a/drivers/base/core.c
> > +++ b/drivers/base/core.c
> > @@ -2022,6 +2022,8 @@ create_syslog_header(const struct device *dev, char *hdr, size_t hdrlen)
> >  		return 0;
> >  
> >  	pos += snprintf(hdr + pos, hdrlen - pos, "SUBSYSTEM=%s", subsys);
> > +	if (pos >= hdrlen)
> > +		goto overflow;
> >  
> >  	/*
> >  	 * Add device identifier DEVICE=:
> > @@ -2053,7 +2055,14 @@ create_syslog_header(const struct device *dev, char *hdr, size_t hdrlen)
> >  				"DEVICE=+%s:%s", subsys, dev_name(dev));
> >  	}
> >  
> > +	if (pos >= hdrlen)
> > +		goto overflow;
> > +
> >  	return pos;
> > +
> > +overflow:
> > +	dev_WARN_ONCE(dev, 1, "device/subsystem name too long");
> 
> Why only warn once?  Any device/subsystem mix should be complained
> about, if for only that we should be really annoying about it to get it
> resolved.

This would expand the volume of logging for the problem device by a
factor of ~50 so it doesn't seem like a good failure mode.

Ben.

-- 
Ben Hutchings
Computers are not intelligent.	They only think they are.

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 828 bytes --]

Re: [PATCH] drivers/base: Fix length checks in create_syslog_header()/dev_vprintk_emit()

[Joe Perches]

On Fri, 2013-12-27 at 20:10 +0100, Ben Hutchings wrote:
> On Fri, 2013-12-27 at 10:47 -0800, Greg Kroah-Hartman wrote:
> > On Fri, Dec 27, 2013 at 06:18:18PM +0100, Ben Hutchings wrote:
> > > snprintf() returns the number of bytes that could have been written
> > > (excluding the null), not the actual number of bytes written.  Given a
> > > long enough subsystem or device name, these functions will advance
> > > beyond the end of the on-stack buffer in dev_vprintk_exit(), resulting
> > > in an information leak or stack corruption.  I don't know whether such
> > > a long name is currently possible.
> > > 
> > > In case snprintf() returns a value >= the buffer size, do not add
> > > structured logging information.  Also WARN the first time this
> > > happens, so we can fix the driver or increase the buffer size.
[]
> > > diff --git a/drivers/base/core.c b/drivers/base/core.c
[]
> > > +overflow:
> > > +	dev_WARN_ONCE(dev, 1, "device/subsystem name too long");
> > 
> > Why only warn once?  Any device/subsystem mix should be complained
> > about, if for only that we should be really annoying about it to get it
> > resolved.
> 
> This would expand the volume of logging for the problem device by a
> factor of ~50 so it doesn't seem like a good failure mode.

There are at least 6 bits on struct device unused.
Maybe add another bool bit field to struct device?

	bool			offline_disabled:1;
	bool			offline:1;

Re: [PATCH] drivers/base: Fix length checks in create_syslog_header()/dev_vprintk_emit()

[Greg Kroah-Hartman]

On Fri, Dec 27, 2013 at 08:10:27PM +0100, Ben Hutchings wrote:
> On Fri, 2013-12-27 at 10:47 -0800, Greg Kroah-Hartman wrote:
> > On Fri, Dec 27, 2013 at 06:18:18PM +0100, Ben Hutchings wrote:
> > > snprintf() returns the number of bytes that could have been written
> > > (excluding the null), not the actual number of bytes written.  Given a
> > > long enough subsystem or device name, these functions will advance
> > > beyond the end of the on-stack buffer in dev_vprintk_exit(), resulting
> > > in an information leak or stack corruption.  I don't know whether such
> > > a long name is currently possible.
> > > 
> > > In case snprintf() returns a value >= the buffer size, do not add
> > > structured logging information.  Also WARN the first time this
> > > happens, so we can fix the driver or increase the buffer size.
> > > 
> > > Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
> > > ---
> > >  drivers/base/core.c | 9 +++++++++
> > >  1 file changed, 9 insertions(+)
> > > 
> > > diff --git a/drivers/base/core.c b/drivers/base/core.c
> > > index 67b180d..989a93c 100644
> > > --- a/drivers/base/core.c
> > > +++ b/drivers/base/core.c
> > > @@ -2022,6 +2022,8 @@ create_syslog_header(const struct device *dev, char *hdr, size_t hdrlen)
> > >  		return 0;
> > >  
> > >  	pos += snprintf(hdr + pos, hdrlen - pos, "SUBSYSTEM=%s", subsys);
> > > +	if (pos >= hdrlen)
> > > +		goto overflow;
> > >  
> > >  	/*
> > >  	 * Add device identifier DEVICE=:
> > > @@ -2053,7 +2055,14 @@ create_syslog_header(const struct device *dev, char *hdr, size_t hdrlen)
> > >  				"DEVICE=+%s:%s", subsys, dev_name(dev));
> > >  	}
> > >  
> > > +	if (pos >= hdrlen)
> > > +		goto overflow;
> > > +
> > >  	return pos;
> > > +
> > > +overflow:
> > > +	dev_WARN_ONCE(dev, 1, "device/subsystem name too long");
> > 
> > Why only warn once?  Any device/subsystem mix should be complained
> > about, if for only that we should be really annoying about it to get it
> > resolved.
> 
> This would expand the volume of logging for the problem device by a
> factor of ~50 so it doesn't seem like a good failure mode.

I want that failure mode to be really obvious so that the user tells the
driver author to fix their code.  A single error message usually just
gets ignored.

So please fail really loudly.

thanks,

greg k-h