[PATCH] xen/blkback: prevent leak of mode during multiple backend_changed calls

[PATCH] xen/blkback: prevent leak of mode during multiple backend_changed calls

[Olaf Hering]

backend_changed might be called multiple times, which will leak
be->mode. free the previous value before storing the current mode value.

Signed-off-by: Olaf Hering <olaf@aepfle.de>
---

!! Not compile tested !!

 drivers/block/xen-blkback/xenbus.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/drivers/block/xen-blkback/xenbus.c b/drivers/block/xen-blkback/xenbus.c
index a6585a4..8650b19 100644
--- a/drivers/block/xen-blkback/xenbus.c
+++ b/drivers/block/xen-blkback/xenbus.c
@@ -502,7 +502,7 @@ static void backend_changed(struct xenbus_watch *watch,
 		= container_of(watch, struct backend_info, backend_watch);
 	struct xenbus_device *dev = be->dev;
 	int cdrom = 0;
-	char *device_type;
+	char *mode, *device_type;
 
 	DPRINTK("");
 
@@ -528,13 +528,15 @@ static void backend_changed(struct xenbus_watch *watch,
 		return;
 	}
 
-	be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
-	if (IS_ERR(be->mode)) {
-		err = PTR_ERR(be->mode);
+	mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
+	kfree(be->mode);
+	if (IS_ERR(mode)) {
+		err = PTR_ERR(mode);
 		be->mode = NULL;
 		xenbus_dev_fatal(dev, err, "reading mode");
 		return;
 	}
+	be->mode = mode;
 
 	device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
 	if (!IS_ERR(device_type)) {
@@ -555,7 +557,7 @@ static void backend_changed(struct xenbus_watch *watch,
 		be->minor = minor;
 
 		err = xen_vbd_create(be->blkif, handle, major, minor,
-				 (NULL == strchr(be->mode, 'w')), cdrom);
+				 (NULL == strchr(mode, 'w')), cdrom);
 		if (err) {
 			be->major = 0;
 			be->minor = 0;
-- 
1.8.0.1

Re: [PATCH] xen/blkback: prevent leak of mode during multiple backend_changed calls

[Jan Beulich]

>>> On 05.12.12 at 11:01, Olaf Hering <olaf@aepfle.de> wrote:
> backend_changed might be called multiple times, which will leak
> be->mode. free the previous value before storing the current mode value.

As said before - this is one possible route to take. But did you
consider at all the alternative of preventing the function from
getting called more than once for a given device? As also said
before, I think that would have other bad effects, and hence
should be preferred (and would likely also result in a smaller
patch).

And _if_ this here is the route to go, I'd clearly see this and your
earlier patch to be folded into just one (dealing with both leaks
in one go).

Jan

> Signed-off-by: Olaf Hering <olaf@aepfle.de>
> ---
> 
> !! Not compile tested !!
> 
>  drivers/block/xen-blkback/xenbus.c | 12 +++++++-----
>  1 file changed, 7 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/block/xen-blkback/xenbus.c 
> b/drivers/block/xen-blkback/xenbus.c
> index a6585a4..8650b19 100644
> --- a/drivers/block/xen-blkback/xenbus.c
> +++ b/drivers/block/xen-blkback/xenbus.c
> @@ -502,7 +502,7 @@ static void backend_changed(struct xenbus_watch *watch,
>  		= container_of(watch, struct backend_info, backend_watch);
>  	struct xenbus_device *dev = be->dev;
>  	int cdrom = 0;
> -	char *device_type;
> +	char *mode, *device_type;
>  
>  	DPRINTK("");
>  
> @@ -528,13 +528,15 @@ static void backend_changed(struct xenbus_watch *watch,
>  		return;
>  	}
>  
> -	be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
> -	if (IS_ERR(be->mode)) {
> -		err = PTR_ERR(be->mode);
> +	mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
> +	kfree(be->mode);
> +	if (IS_ERR(mode)) {
> +		err = PTR_ERR(mode);
>  		be->mode = NULL;
>  		xenbus_dev_fatal(dev, err, "reading mode");
>  		return;
>  	}
> +	be->mode = mode;
>  
>  	device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
>  	if (!IS_ERR(device_type)) {
> @@ -555,7 +557,7 @@ static void backend_changed(struct xenbus_watch *watch,
>  		be->minor = minor;
>  
>  		err = xen_vbd_create(be->blkif, handle, major, minor,
> -				 (NULL == strchr(be->mode, 'w')), cdrom);
> +				 (NULL == strchr(mode, 'w')), cdrom);
>  		if (err) {
>  			be->major = 0;
>  			be->minor = 0;
> -- 
> 1.8.0.1

Re: [PATCH] xen/blkback: prevent leak of mode during multiple backend_changed calls

[Olaf Hering]

On Wed, Dec 05, Jan Beulich wrote:

> >>> On 05.12.12 at 11:01, Olaf Hering <olaf@aepfle.de> wrote:
> > backend_changed might be called multiple times, which will leak
> > be->mode. free the previous value before storing the current mode value.
> 
> As said before - this is one possible route to take. But did you
> consider at all the alternative of preventing the function from
> getting called more than once for a given device? As also said
> before, I think that would have other bad effects, and hence
> should be preferred (and would likely also result in a smaller
> patch).

Maybe it could be done like this, adding a flag to the backend device
and exit early if its called twice.

Olaf


diff --git a/drivers/block/xen-blkback/xenbus.c b/drivers/block/xen-blkback/xenbus.c
index a6585a4..2822e73 100644
--- a/drivers/block/xen-blkback/xenbus.c
+++ b/drivers/block/xen-blkback/xenbus.c
@@ -28,6 +28,7 @@ struct backend_info {
 	unsigned		major;
 	unsigned		minor;
 	char			*mode;
+	unsigned		alive;
 };
 
 static struct kmem_cache *xen_blkif_cachep;
@@ -506,6 +507,9 @@ static void backend_changed(struct xenbus_watch *watch,
 
 	DPRINTK("");
 
+	if (be->alive)
+		return;
+
 	err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
 			   &major, &minor);
 	if (XENBUS_EXIST_ERR(err)) {
@@ -548,8 +552,11 @@ static void backend_changed(struct xenbus_watch *watch,
 		char *p = strrchr(dev->otherend, '/') + 1;
 		long handle;
 		err = strict_strtoul(p, 0, &handle);
-		if (err)
+		if (err) {
+			kfree(be->mode);
+			be->mode = NULL;
 			return;
+		}
 
 		be->major = major;
 		be->minor = minor;
@@ -560,6 +567,8 @@ static void backend_changed(struct xenbus_watch *watch,
 			be->major = 0;
 			be->minor = 0;
 			xenbus_dev_fatal(dev, err, "creating vbd structure");
+			kfree(be->mode);
+			be->mode = NULL;
 			return;
 		}
 
@@ -569,10 +578,13 @@ static void backend_changed(struct xenbus_watch *watch,
 			be->major = 0;
 			be->minor = 0;
 			xenbus_dev_fatal(dev, err, "creating sysfs entries");
+			kfree(be->mode);
+			be->mode = NULL;
 			return;
 		}
 
 		/* We're potentially connected now */
+		be->alive = 1;
 		xen_update_blkif_status(be->blkif);
 	}
 }
-- 
1.8.0.1

Re: [PATCH] xen/blkback: prevent leak of mode during multiple backend_changed calls

[Jan Beulich]

>>> On 06.12.12 at 17:23, Olaf Hering <olaf@aepfle.de> wrote:
> On Wed, Dec 05, Jan Beulich wrote:
> 
>> >>> On 05.12.12 at 11:01, Olaf Hering <olaf@aepfle.de> wrote:
>> > backend_changed might be called multiple times, which will leak
>> > be->mode. free the previous value before storing the current mode value.
>> 
>> As said before - this is one possible route to take. But did you
>> consider at all the alternative of preventing the function from
>> getting called more than once for a given device? As also said
>> before, I think that would have other bad effects, and hence
>> should be preferred (and would likely also result in a smaller
>> patch).
> 
> Maybe it could be done like this, adding a flag to the backend device
> and exit early if its called twice.

Maybe, but it looks odd to me. But then again I had hoped Konrad
would have an opinion here...

Also I don't see why you need to free be->mode now on all error
paths - afaict it would still get freed when "be" gets freed (with
your earlier patch).

Jan

> --- a/drivers/block/xen-blkback/xenbus.c
> +++ b/drivers/block/xen-blkback/xenbus.c
> @@ -28,6 +28,7 @@ struct backend_info {
>  	unsigned		major;
>  	unsigned		minor;
>  	char			*mode;
> +	unsigned		alive;
>  };
>  
>  static struct kmem_cache *xen_blkif_cachep;
> @@ -506,6 +507,9 @@ static void backend_changed(struct xenbus_watch *watch,
>  
>  	DPRINTK("");
>  
> +	if (be->alive)
> +		return;
> +
>  	err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
>  			   &major, &minor);
>  	if (XENBUS_EXIST_ERR(err)) {
> @@ -548,8 +552,11 @@ static void backend_changed(struct xenbus_watch *watch,
>  		char *p = strrchr(dev->otherend, '/') + 1;
>  		long handle;
>  		err = strict_strtoul(p, 0, &handle);
> -		if (err)
> +		if (err) {
> +			kfree(be->mode);
> +			be->mode = NULL;
>  			return;
> +		}
>  
>  		be->major = major;
>  		be->minor = minor;
> @@ -560,6 +567,8 @@ static void backend_changed(struct xenbus_watch *watch,
>  			be->major = 0;
>  			be->minor = 0;
>  			xenbus_dev_fatal(dev, err, "creating vbd structure");
> +			kfree(be->mode);
> +			be->mode = NULL;
>  			return;
>  		}
>  
> @@ -569,10 +578,13 @@ static void backend_changed(struct xenbus_watch 
> *watch,
>  			be->major = 0;
>  			be->minor = 0;
>  			xenbus_dev_fatal(dev, err, "creating sysfs entries");
> +			kfree(be->mode);
> +			be->mode = NULL;
>  			return;
>  		}
>  
>  		/* We're potentially connected now */
> +		be->alive = 1;
>  		xen_update_blkif_status(be->blkif);
>  	}
>  }
> -- 
> 1.8.0.1

Re: [PATCH] xen/blkback: prevent leak of mode during multiple backend_changed calls

[Olaf Hering]

On Thu, Dec 06, Jan Beulich wrote:

> >>> On 06.12.12 at 17:23, Olaf Hering <olaf@aepfle.de> wrote:
> > On Wed, Dec 05, Jan Beulich wrote:
> > 
> >> >>> On 05.12.12 at 11:01, Olaf Hering <olaf@aepfle.de> wrote:
> >> > backend_changed might be called multiple times, which will leak
> >> > be->mode. free the previous value before storing the current mode value.
> >> 
> >> As said before - this is one possible route to take. But did you
> >> consider at all the alternative of preventing the function from
> >> getting called more than once for a given device? As also said
> >> before, I think that would have other bad effects, and hence
> >> should be preferred (and would likely also result in a smaller
> >> patch).
> > 
> > Maybe it could be done like this, adding a flag to the backend device
> > and exit early if its called twice.
> 
> Maybe, but it looks odd to me. But then again I had hoped Konrad
> would have an opinion here...

Looking at this some more, if backend_changed is supposed to be called
exactly once then major/minor checks can be removed because they will be
always zero, like this:


 drivers/block/xen-blkback/xenbus.c | 68 ++++++++++++++++++--------------------
 1 file changed, 32 insertions(+), 36 deletions(-)

diff --git a/drivers/block/xen-blkback/xenbus.c b/drivers/block/xen-blkback/xenbus.c
index a6585a4..5ca77c3 100644
--- a/drivers/block/xen-blkback/xenbus.c
+++ b/drivers/block/xen-blkback/xenbus.c
@@ -28,6 +28,7 @@ struct backend_info {
 	unsigned		major;
 	unsigned		minor;
 	char			*mode;
+	unsigned		alive;
 };
 
 static struct kmem_cache *xen_blkif_cachep;
@@ -502,10 +503,14 @@ static void backend_changed(struct xenbus_watch *watch,
 		= container_of(watch, struct backend_info, backend_watch);
 	struct xenbus_device *dev = be->dev;
 	int cdrom = 0;
-	char *device_type;
+	char *device_type, *p;
+	long handle;
 
 	DPRINTK("");
 
+	if (be->alive)
+		return;
+
 	err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
 			   &major, &minor);
 	if (XENBUS_EXIST_ERR(err)) {
@@ -521,12 +526,7 @@ static void backend_changed(struct xenbus_watch *watch,
 		return;
 	}
 
-	if ((be->major || be->minor) &&
-	    ((be->major != major) || (be->minor != minor))) {
-		pr_warn(DRV_PFX "changing physical device (from %x:%x to %x:%x) not supported.\n",
-			be->major, be->minor, major, minor);
-		return;
-	}
+	be->alive = 1;
 
 	be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
 	if (IS_ERR(be->mode)) {
@@ -542,39 +542,35 @@ static void backend_changed(struct xenbus_watch *watch,
 		kfree(device_type);
 	}
 
-	if (be->major == 0 && be->minor == 0) {
-		/* Front end dir is a number, which is used as the handle. */
-
-		char *p = strrchr(dev->otherend, '/') + 1;
-		long handle;
-		err = strict_strtoul(p, 0, &handle);
-		if (err)
-			return;
-
-		be->major = major;
-		be->minor = minor;
+	/* Front end dir is a number, which is used as the handle. */
+	p = strrchr(dev->otherend, '/') + 1;
+	err = strict_strtoul(p, 0, &handle);
+	if (err)
+		return;
 
-		err = xen_vbd_create(be->blkif, handle, major, minor,
-				 (NULL == strchr(be->mode, 'w')), cdrom);
-		if (err) {
-			be->major = 0;
-			be->minor = 0;
-			xenbus_dev_fatal(dev, err, "creating vbd structure");
-			return;
-		}
+	be->major = major;
+	be->minor = minor;
 
-		err = xenvbd_sysfs_addif(dev);
-		if (err) {
-			xen_vbd_free(&be->blkif->vbd);
-			be->major = 0;
-			be->minor = 0;
-			xenbus_dev_fatal(dev, err, "creating sysfs entries");
-			return;
-		}
+	err = xen_vbd_create(be->blkif, handle, major, minor,
+			 (NULL == strchr(be->mode, 'w')), cdrom);
+	if (err) {
+		be->major = 0;
+		be->minor = 0;
+		xenbus_dev_fatal(dev, err, "creating vbd structure");
+		return;
+	}
 
-		/* We're potentially connected now */
-		xen_update_blkif_status(be->blkif);
+	err = xenvbd_sysfs_addif(dev);
+	if (err) {
+		xen_vbd_free(&be->blkif->vbd);
+		be->major = 0;
+		be->minor = 0;
+		xenbus_dev_fatal(dev, err, "creating sysfs entries");
+		return;
 	}
+
+	/* We're potentially connected now */
+	xen_update_blkif_status(be->blkif);
 }
 
 
-- 
1.8.0.1

Re: [PATCH] xen/blkback: prevent leak of mode during multiple backend_changed calls

[Konrad Rzeszutek Wilk]

On Thu, Dec 06, 2012 at 05:02:50PM +0000, Jan Beulich wrote:
> >>> On 06.12.12 at 17:23, Olaf Hering <olaf@aepfle.de> wrote:
> > On Wed, Dec 05, Jan Beulich wrote:
> > 
> >> >>> On 05.12.12 at 11:01, Olaf Hering <olaf@aepfle.de> wrote:
> >> > backend_changed might be called multiple times, which will leak
> >> > be->mode. free the previous value before storing the current mode value.
> >> 
> >> As said before - this is one possible route to take. But did you
> >> consider at all the alternative of preventing the function from
> >> getting called more than once for a given device? As also said
> >> before, I think that would have other bad effects, and hence
> >> should be preferred (and would likely also result in a smaller
> >> patch).
> > 
> > Maybe it could be done like this, adding a flag to the backend device
> > and exit early if its called twice.
> 
> Maybe, but it looks odd to me. But then again I had hoped Konrad
> would have an opinion here...

Sorry - was lurking around and hadn't paid any attention to this thread.
And it does not help that next week I am out :-)

> 
> Also I don't see why you need to free be->mode now on all error
> paths - afaict it would still get freed when "be" gets freed (with
> your earlier patch).
> 
> Jan
> 
> > --- a/drivers/block/xen-blkback/xenbus.c
> > +++ b/drivers/block/xen-blkback/xenbus.c
> > @@ -28,6 +28,7 @@ struct backend_info {
> >  	unsigned		major;
> >  	unsigned		minor;
> >  	char			*mode;
> > +	unsigned		alive;
> >  };
> >  
> >  static struct kmem_cache *xen_blkif_cachep;
> > @@ -506,6 +507,9 @@ static void backend_changed(struct xenbus_watch *watch,
> >  
> >  	DPRINTK("");
> >  
> > +	if (be->alive)
> > +		return;
> > +
> >  	err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
> >  			   &major, &minor);
> >  	if (XENBUS_EXIST_ERR(err)) {
> > @@ -548,8 +552,11 @@ static void backend_changed(struct xenbus_watch *watch,
> >  		char *p = strrchr(dev->otherend, '/') + 1;
> >  		long handle;
> >  		err = strict_strtoul(p, 0, &handle);
> > -		if (err)
> > +		if (err) {
> > +			kfree(be->mode);
> > +			be->mode = NULL;
> >  			return;
> > +		}
> >  
> >  		be->major = major;
> >  		be->minor = minor;
> > @@ -560,6 +567,8 @@ static void backend_changed(struct xenbus_watch *watch,
> >  			be->major = 0;
> >  			be->minor = 0;
> >  			xenbus_dev_fatal(dev, err, "creating vbd structure");
> > +			kfree(be->mode);
> > +			be->mode = NULL;
> >  			return;
> >  		}
> >  
> > @@ -569,10 +578,13 @@ static void backend_changed(struct xenbus_watch 
> > *watch,
> >  			be->major = 0;
> >  			be->minor = 0;
> >  			xenbus_dev_fatal(dev, err, "creating sysfs entries");
> > +			kfree(be->mode);
> > +			be->mode = NULL;
> >  			return;
> >  		}
> >  
> >  		/* We're potentially connected now */
> > +		be->alive = 1;
> >  		xen_update_blkif_status(be->blkif);
> >  	}
> >  }
> > -- 
> > 1.8.0.1
> 
> 
>