linux-arm-msm.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH v1 0/2] soc: qcom: rmtfs-mem: Support non-root rmtfs daemons
@ 2018-12-18  0:08 Evan Green
  2018-12-18  0:08 ` [PATCH v1 1/2] soc: qcom: rmtfs-mem: Add class to enable uevents Evan Green
  2018-12-18  0:08 ` [PATCH v1 2/2] soc: qcom: rmtfs-mem: Make sysfs attributes world-readable Evan Green
  0 siblings, 2 replies; 6+ messages in thread
From: Evan Green @ 2018-12-18  0:08 UTC (permalink / raw)
  To: Andy Gross, Bjorn Andersson
  Cc: Brian Norris, Ben Chan, Evan Green, linux-arm-msm, David Brown,
	linux-kernel

This series contains minor fixes needed to better support running an
rmtfs daemon from an unprivileged process:
 - Enable uevents on the child character device by adding a struct
class to the parent. I needed these so that I could change the
ownership of /dev/qcom_rmtfs_mem1, and drove myself crazy trying to
understand why my udev rules never fired.
 - Enable access to phys_addr and size sysfs attributes. The daemon
needs to read these, and they don't really contain anything sensitive,
so expose them.

I still need CAP_NET_ADMIN to be able to bind to the right qrtr port,
but at least with these changes I can run as a different user, and drop
all other privileges.


Evan Green (2):
  soc: qcom: rmtfs-mem: Add class to enable uevents
  soc: qcom: rmtfs-mem: Make sysfs attributes world-readable

 drivers/soc/qcom/rmtfs_mem.c | 29 +++++++++++++++++++++--------
 1 file changed, 21 insertions(+), 8 deletions(-)

-- 
2.18.1

^ permalink raw reply	[flat|nested] 6+ messages in thread

* [PATCH v1 1/2] soc: qcom: rmtfs-mem: Add class to enable uevents
  2018-12-18  0:08 [PATCH v1 0/2] soc: qcom: rmtfs-mem: Support non-root rmtfs daemons Evan Green
@ 2018-12-18  0:08 ` Evan Green
  2018-12-21  1:18   ` Brian Norris
  2018-12-18  0:08 ` [PATCH v1 2/2] soc: qcom: rmtfs-mem: Make sysfs attributes world-readable Evan Green
  1 sibling, 1 reply; 6+ messages in thread
From: Evan Green @ 2018-12-18  0:08 UTC (permalink / raw)
  To: Andy Gross, Bjorn Andersson
  Cc: Brian Norris, Ben Chan, Evan Green, linux-arm-msm, David Brown,
	linux-kernel

Currently the qcom_rmtfs_memN devices are entirely invisible to the udev world.
Add a class to the rmtfs device so that uevents fire when the device is added.

Signed-off-by: Evan Green <evgreen@chromium.org>
---

 drivers/soc/qcom/rmtfs_mem.c | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)

diff --git a/drivers/soc/qcom/rmtfs_mem.c b/drivers/soc/qcom/rmtfs_mem.c
index 97bb5989aa211..0bf800ee2a978 100644
--- a/drivers/soc/qcom/rmtfs_mem.c
+++ b/drivers/soc/qcom/rmtfs_mem.c
@@ -132,6 +132,11 @@ static int qcom_rmtfs_mem_release(struct inode *inode, struct file *filp)
 	return 0;
 }
 
+static struct class rmtfs_class = {
+	.owner          = THIS_MODULE,
+	.name           = "rmtfs",
+};
+
 static const struct file_operations qcom_rmtfs_mem_fops = {
 	.owner = THIS_MODULE,
 	.open = qcom_rmtfs_mem_open,
@@ -173,9 +178,15 @@ static int qcom_rmtfs_mem_probe(struct platform_device *pdev)
 
 	}
 
+	ret  = class_register(&rmtfs_class);
+	if (ret)
+		return ret;
+
 	rmtfs_mem = kzalloc(sizeof(*rmtfs_mem), GFP_KERNEL);
-	if (!rmtfs_mem)
-		return -ENOMEM;
+	if (!rmtfs_mem) {
+		ret = -ENOMEM;
+		goto unregister_class;
+	}
 
 	rmtfs_mem->addr = rmem->base;
 	rmtfs_mem->client_id = client_id;
@@ -199,8 +210,8 @@ static int qcom_rmtfs_mem_probe(struct platform_device *pdev)
 
 	dev_set_name(&rmtfs_mem->dev, "qcom_rmtfs_mem%d", client_id);
 	rmtfs_mem->dev.id = client_id;
+	rmtfs_mem->dev.class = &rmtfs_class;
 	rmtfs_mem->dev.devt = MKDEV(MAJOR(qcom_rmtfs_mem_major), client_id);
-
 	ret = cdev_device_add(&rmtfs_mem->cdev, &rmtfs_mem->dev);
 	if (ret) {
 		dev_err(&pdev->dev, "failed to add cdev: %d\n", ret);
@@ -235,11 +246,13 @@ static int qcom_rmtfs_mem_probe(struct platform_device *pdev)
 
 	return 0;
 
+
 remove_cdev:
 	cdev_device_del(&rmtfs_mem->cdev, &rmtfs_mem->dev);
 put_device:
 	put_device(&rmtfs_mem->dev);
-
+unregister_class:
+	class_unregister(&rmtfs_class);
 	return ret;
 }
 
@@ -258,7 +271,7 @@ static int qcom_rmtfs_mem_remove(struct platform_device *pdev)
 
 	cdev_device_del(&rmtfs_mem->cdev, &rmtfs_mem->dev);
 	put_device(&rmtfs_mem->dev);
-
+	class_unregister(&rmtfs_class);
 	return 0;
 }
 
-- 
2.18.1

^ permalink raw reply related	[flat|nested] 6+ messages in thread

* [PATCH v1 2/2] soc: qcom: rmtfs-mem: Make sysfs attributes world-readable
  2018-12-18  0:08 [PATCH v1 0/2] soc: qcom: rmtfs-mem: Support non-root rmtfs daemons Evan Green
  2018-12-18  0:08 ` [PATCH v1 1/2] soc: qcom: rmtfs-mem: Add class to enable uevents Evan Green
@ 2018-12-18  0:08 ` Evan Green
  2018-12-21  1:21   ` Brian Norris
  1 sibling, 1 reply; 6+ messages in thread
From: Evan Green @ 2018-12-18  0:08 UTC (permalink / raw)
  To: Andy Gross, Bjorn Andersson
  Cc: Brian Norris, Ben Chan, Evan Green, linux-arm-msm, David Brown,
	linux-kernel

In order to run an rmtfs daemon as an unprivileged user, that user would
need access to the phys_addr and size sysfs attributes. Sharing these
attributes with unprivileged users doesn't really leak anything
sensitive, since if you have access to physical memory, the jig is
up anyway.

Make those attributes readable by all.

Signed-off-by: Evan Green <evgreen@chromium.org>
---

 drivers/soc/qcom/rmtfs_mem.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/soc/qcom/rmtfs_mem.c b/drivers/soc/qcom/rmtfs_mem.c
index 0bf800ee2a978..4eeb9f02e7889 100644
--- a/drivers/soc/qcom/rmtfs_mem.c
+++ b/drivers/soc/qcom/rmtfs_mem.c
@@ -45,9 +45,9 @@ static ssize_t qcom_rmtfs_mem_show(struct device *dev,
 			      struct device_attribute *attr,
 			      char *buf);
 
-static DEVICE_ATTR(phys_addr, 0400, qcom_rmtfs_mem_show, NULL);
-static DEVICE_ATTR(size, 0400, qcom_rmtfs_mem_show, NULL);
-static DEVICE_ATTR(client_id, 0400, qcom_rmtfs_mem_show, NULL);
+static DEVICE_ATTR(phys_addr, 0444, qcom_rmtfs_mem_show, NULL);
+static DEVICE_ATTR(size, 0444, qcom_rmtfs_mem_show, NULL);
+static DEVICE_ATTR(client_id, 0444, qcom_rmtfs_mem_show, NULL);
 
 static ssize_t qcom_rmtfs_mem_show(struct device *dev,
 			      struct device_attribute *attr,
-- 
2.18.1

^ permalink raw reply related	[flat|nested] 6+ messages in thread

* Re: [PATCH v1 1/2] soc: qcom: rmtfs-mem: Add class to enable uevents
  2018-12-18  0:08 ` [PATCH v1 1/2] soc: qcom: rmtfs-mem: Add class to enable uevents Evan Green
@ 2018-12-21  1:18   ` Brian Norris
  2018-12-21 17:56     ` Evan Green
  0 siblings, 1 reply; 6+ messages in thread
From: Brian Norris @ 2018-12-21  1:18 UTC (permalink / raw)
  To: Evan Green
  Cc: Andy Gross, Bjorn Andersson, Ben Chan, linux-arm-msm,
	David Brown, linux-kernel

Hi Evan,

On Mon, Dec 17, 2018 at 04:08:33PM -0800, Evan Green wrote:
> Currently the qcom_rmtfs_memN devices are entirely invisible to the udev world.
> Add a class to the rmtfs device so that uevents fire when the device is added.
> 
> Signed-off-by: Evan Green <evgreen@chromium.org>
> ---
> 
>  drivers/soc/qcom/rmtfs_mem.c | 23 ++++++++++++++++++-----
>  1 file changed, 18 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/soc/qcom/rmtfs_mem.c b/drivers/soc/qcom/rmtfs_mem.c
> index 97bb5989aa211..0bf800ee2a978 100644
> --- a/drivers/soc/qcom/rmtfs_mem.c
> +++ b/drivers/soc/qcom/rmtfs_mem.c
> @@ -132,6 +132,11 @@ static int qcom_rmtfs_mem_release(struct inode *inode, struct file *filp)
>  	return 0;
>  }
>  
> +static struct class rmtfs_class = {
> +	.owner          = THIS_MODULE,
> +	.name           = "rmtfs",
> +};
> +
>  static const struct file_operations qcom_rmtfs_mem_fops = {
>  	.owner = THIS_MODULE,
>  	.open = qcom_rmtfs_mem_open,
> @@ -173,9 +178,15 @@ static int qcom_rmtfs_mem_probe(struct platform_device *pdev)
>  
>  	}
>  
> +	ret  = class_register(&rmtfs_class);
> +	if (ret)
> +		return ret;

Hmm, is this how classes are supposed to work? Usually, you have the
possibility of more than 1 device per class, and therefore you don't
register the class in the driver probe -- you register it in the init()
routine, or something similar. As it is, I expect this will break if
there were ever a second rmtfs device.

Brian

> +
>  	rmtfs_mem = kzalloc(sizeof(*rmtfs_mem), GFP_KERNEL);
> -	if (!rmtfs_mem)
> -		return -ENOMEM;
> +	if (!rmtfs_mem) {
> +		ret = -ENOMEM;
> +		goto unregister_class;
> +	}
>  
>  	rmtfs_mem->addr = rmem->base;
>  	rmtfs_mem->client_id = client_id;
> @@ -199,8 +210,8 @@ static int qcom_rmtfs_mem_probe(struct platform_device *pdev)
>  
>  	dev_set_name(&rmtfs_mem->dev, "qcom_rmtfs_mem%d", client_id);
>  	rmtfs_mem->dev.id = client_id;
> +	rmtfs_mem->dev.class = &rmtfs_class;
>  	rmtfs_mem->dev.devt = MKDEV(MAJOR(qcom_rmtfs_mem_major), client_id);
> -
>  	ret = cdev_device_add(&rmtfs_mem->cdev, &rmtfs_mem->dev);
>  	if (ret) {
>  		dev_err(&pdev->dev, "failed to add cdev: %d\n", ret);
> @@ -235,11 +246,13 @@ static int qcom_rmtfs_mem_probe(struct platform_device *pdev)
>  
>  	return 0;
>  
> +
>  remove_cdev:
>  	cdev_device_del(&rmtfs_mem->cdev, &rmtfs_mem->dev);
>  put_device:
>  	put_device(&rmtfs_mem->dev);
> -
> +unregister_class:
> +	class_unregister(&rmtfs_class);
>  	return ret;
>  }
>  
> @@ -258,7 +271,7 @@ static int qcom_rmtfs_mem_remove(struct platform_device *pdev)
>  
>  	cdev_device_del(&rmtfs_mem->cdev, &rmtfs_mem->dev);
>  	put_device(&rmtfs_mem->dev);
> -
> +	class_unregister(&rmtfs_class);
>  	return 0;
>  }
>  
> -- 
> 2.18.1
> 

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH v1 2/2] soc: qcom: rmtfs-mem: Make sysfs attributes world-readable
  2018-12-18  0:08 ` [PATCH v1 2/2] soc: qcom: rmtfs-mem: Make sysfs attributes world-readable Evan Green
@ 2018-12-21  1:21   ` Brian Norris
  0 siblings, 0 replies; 6+ messages in thread
From: Brian Norris @ 2018-12-21  1:21 UTC (permalink / raw)
  To: Evan Green
  Cc: Andy Gross, Bjorn Andersson, Ben Chan, linux-arm-msm,
	David Brown, linux-kernel

On Mon, Dec 17, 2018 at 04:08:34PM -0800, Evan Green wrote:
> In order to run an rmtfs daemon as an unprivileged user, that user would
> need access to the phys_addr and size sysfs attributes. Sharing these
> attributes with unprivileged users doesn't really leak anything
> sensitive, since if you have access to physical memory, the jig is
> up anyway.
> 
> Make those attributes readable by all.
> 
> Signed-off-by: Evan Green <evgreen@chromium.org>
> ---

Seems fine to me.

Reviewed-by: Brian Norris <briannorris@chromium.org>

> 
>  drivers/soc/qcom/rmtfs_mem.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/soc/qcom/rmtfs_mem.c b/drivers/soc/qcom/rmtfs_mem.c
> index 0bf800ee2a978..4eeb9f02e7889 100644
> --- a/drivers/soc/qcom/rmtfs_mem.c
> +++ b/drivers/soc/qcom/rmtfs_mem.c
> @@ -45,9 +45,9 @@ static ssize_t qcom_rmtfs_mem_show(struct device *dev,
>  			      struct device_attribute *attr,
>  			      char *buf);
>  
> -static DEVICE_ATTR(phys_addr, 0400, qcom_rmtfs_mem_show, NULL);
> -static DEVICE_ATTR(size, 0400, qcom_rmtfs_mem_show, NULL);
> -static DEVICE_ATTR(client_id, 0400, qcom_rmtfs_mem_show, NULL);
> +static DEVICE_ATTR(phys_addr, 0444, qcom_rmtfs_mem_show, NULL);
> +static DEVICE_ATTR(size, 0444, qcom_rmtfs_mem_show, NULL);
> +static DEVICE_ATTR(client_id, 0444, qcom_rmtfs_mem_show, NULL);
>  
>  static ssize_t qcom_rmtfs_mem_show(struct device *dev,
>  			      struct device_attribute *attr,
> -- 
> 2.18.1
> 

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH v1 1/2] soc: qcom: rmtfs-mem: Add class to enable uevents
  2018-12-21  1:18   ` Brian Norris
@ 2018-12-21 17:56     ` Evan Green
  0 siblings, 0 replies; 6+ messages in thread
From: Evan Green @ 2018-12-21 17:56 UTC (permalink / raw)
  To: Brian Norris
  Cc: Andy Gross, Bjorn Andersson, Ben Chan, linux-arm-msm,
	David Brown, linux-kernel

On Thu, Dec 20, 2018 at 5:19 PM Brian Norris <briannorris@chromium.org> wrote:
>
> Hi Evan,
>
> On Mon, Dec 17, 2018 at 04:08:33PM -0800, Evan Green wrote:
> > Currently the qcom_rmtfs_memN devices are entirely invisible to the udev world.
> > Add a class to the rmtfs device so that uevents fire when the device is added.
> >
> > Signed-off-by: Evan Green <evgreen@chromium.org>
> > ---
> >
> >  drivers/soc/qcom/rmtfs_mem.c | 23 ++++++++++++++++++-----
> >  1 file changed, 18 insertions(+), 5 deletions(-)
> >
> > diff --git a/drivers/soc/qcom/rmtfs_mem.c b/drivers/soc/qcom/rmtfs_mem.c
> > index 97bb5989aa211..0bf800ee2a978 100644
> > --- a/drivers/soc/qcom/rmtfs_mem.c
> > +++ b/drivers/soc/qcom/rmtfs_mem.c
> > @@ -132,6 +132,11 @@ static int qcom_rmtfs_mem_release(struct inode *inode, struct file *filp)
> >       return 0;
> >  }
> >
> > +static struct class rmtfs_class = {
> > +     .owner          = THIS_MODULE,
> > +     .name           = "rmtfs",
> > +};
> > +
> >  static const struct file_operations qcom_rmtfs_mem_fops = {
> >       .owner = THIS_MODULE,
> >       .open = qcom_rmtfs_mem_open,
> > @@ -173,9 +178,15 @@ static int qcom_rmtfs_mem_probe(struct platform_device *pdev)
> >
> >       }
> >
> > +     ret  = class_register(&rmtfs_class);
> > +     if (ret)
> > +             return ret;
>
> Hmm, is this how classes are supposed to work? Usually, you have the
> possibility of more than 1 device per class, and therefore you don't
> register the class in the driver probe -- you register it in the init()
> routine, or something similar. As it is, I expect this will break if
> there were ever a second rmtfs device.
>

Fair enough. I'll move this to an init routine.

^ permalink raw reply	[flat|nested] 6+ messages in thread

end of thread, other threads:[~2018-12-21 17:56 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-12-18  0:08 [PATCH v1 0/2] soc: qcom: rmtfs-mem: Support non-root rmtfs daemons Evan Green
2018-12-18  0:08 ` [PATCH v1 1/2] soc: qcom: rmtfs-mem: Add class to enable uevents Evan Green
2018-12-21  1:18   ` Brian Norris
2018-12-21 17:56     ` Evan Green
2018-12-18  0:08 ` [PATCH v1 2/2] soc: qcom: rmtfs-mem: Make sysfs attributes world-readable Evan Green
2018-12-21  1:21   ` Brian Norris

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).