On Tue, Jun 22, 2021 at 09:43:49AM +0200, Arnaud POULIQUEN wrote:
> Hi Mathieu
>
> On 6/22/21 1:16 AM, Mathieu Poirier wrote:
> > On Mon, Jun 21, 2021 at 02:58:00PM +0200, Arnaud Pouliquen wrote:
> >> Allow the user space application to create and release an rpmsg device
> >> by adding RPMSG_CREATE_DEV_IOCTL and RPMSG_RELEASE_DEV_IOCTL ioctrls to
> >> the /dev/rpmsg_ctrl interface
> >>
> >> The RPMSG_CREATE_DEV_IOCTL Ioctl can be used to instantiate a local rpmsg
> >> device.
> >> Depending on the back-end implementation, the associated rpmsg driver is
> >> probed and a NS announcement can be sent to the remote processor.
> >>
> >> The RPMSG_RELEASE_DEV_IOCTL allows the user application to release a
> >> rpmsg device created either by the remote processor or with the
> >> RPMSG_CREATE_DEV_IOCTL call.
> >> Depending on the back-end implementation, the associated rpmsg driver is
> >> removed and a NS destroy rpmsg can be sent to the remote processor.
> >>
> >> Suggested-by: Mathieu Poirier <mathieu.poirier@xxxxxxxxxx>
> >> Signed-off-by: Arnaud Pouliquen <arnaud.pouliquen@xxxxxxxxxxx>
> >> ---
> >> drivers/rpmsg/rpmsg_ctrl.c | 37 +++++++++++++++++++++++++++++++++----
> >> include/uapi/linux/rpmsg.h | 10 ++++++++++
> >> 2 files changed, 43 insertions(+), 4 deletions(-)
> >>
> >> diff --git a/drivers/rpmsg/rpmsg_ctrl.c b/drivers/rpmsg/rpmsg_ctrl.c
> >> index eeb1708548c1..cb19e32d05e1 100644
> >> --- a/drivers/rpmsg/rpmsg_ctrl.c
> >> +++ b/drivers/rpmsg/rpmsg_ctrl.c
> >> @@ -23,6 +23,7 @@
> >> #include <uapi/linux/rpmsg.h>
> >>
> >> #include "rpmsg_char.h"
> >> +#include "rpmsg_internal.h"
> >>
> >> static dev_t rpmsg_major;
> >>
> >> @@ -37,11 +38,13 @@ static DEFINE_IDA(rpmsg_minor_ida);
> >> * @rpdev: underlaying rpmsg device
> >> * @cdev: cdev for the ctrl device
> >> * @dev: device for the ctrl device
> >> + * @ctrl_lock: serialize the ioctrls.
> >> */
> >> struct rpmsg_ctrldev {
> >> struct rpmsg_device *rpdev;
> >> struct cdev cdev;
> >> struct device dev;
> >> + struct mutex ctrl_lock;
> >> };
> >>
> >> static int rpmsg_ctrldev_open(struct inode *inode, struct file *filp)
> >> @@ -70,9 +73,8 @@ static long rpmsg_ctrldev_ioctl(struct file *fp, unsigned int cmd,
> >> void __user *argp = (void __user *)arg;
> >> struct rpmsg_endpoint_info eptinfo;
> >> struct rpmsg_channel_info chinfo;
> >> -
> >> - if (cmd != RPMSG_CREATE_EPT_IOCTL)
> >> - return -EINVAL;
> >> + struct rpmsg_device *rpdev;
> >> + int ret = 0;
> >>
> >> if (copy_from_user(&eptinfo, argp, sizeof(eptinfo)))
> >> return -EFAULT;
> >> @@ -82,7 +84,33 @@ static long rpmsg_ctrldev_ioctl(struct file *fp, unsigned int cmd,
> >> chinfo.src = eptinfo.src;
> >> chinfo.dst = eptinfo.dst;
> >>
> >> - return rpmsg_chrdev_eptdev_create(ctrldev->rpdev, &ctrldev->dev, chinfo);
> >> + mutex_lock(&ctrldev->ctrl_lock);
> >
> > Have you been able to verify the VFS layer doesn't take care of serializing
> > ioctl() and file accesses in general? I did a quick search in the drivers/
> > directory and the vast majority of implementations don't use a lock.
>
> Please find my answer here https://lkml.org/lkml/2021/6/21/235
Ah ok, I hadn't noticed that one.
> I tested by suppressing the lock and adding msleep to check a potential atomic.
> in the rpmsg_ctrldev_ioctl function. Nothing seems prevent the re-entrance.
Thanks for testing this out. I did some research on my side and turns out
unlocked_ioctl() was introduced to get rid of the Big Kernel Lock rather than
indicate that serialisation is not supported by this interface. Someone also
asked the exact same question here [1].
So yes, locking is required.
Reviewed-by: Mathieu Poirier <mathieu.poirier@xxxxxxxxxx>
[1]. https://stackoverflow.com/questions/18874962/linux-device-driver-file-operations-it-is-possible-to-have-race-conditions
>
> Regards,
> Arnaud
>
> >
> > Thanks,
> > Mathieu
> >
> >> + switch (cmd) {
> >> + case RPMSG_CREATE_EPT_IOCTL:
> >> + ret = rpmsg_chrdev_eptdev_create(ctrldev->rpdev, &ctrldev->dev, chinfo);
> >> + break;
> >> +
> >> + case RPMSG_CREATE_DEV_IOCTL:
> >> + rpdev = rpmsg_create_channel(ctrldev->rpdev, &chinfo);
> >> + if (!rpdev) {
> >> + dev_err(&ctrldev->dev, "failed to create %s channel\n", chinfo.name);
> >> + ret = -ENXIO;
> >> + }
> >> + break;
> >> +
> >> + case RPMSG_RELEASE_DEV_IOCTL:
> >> + ret = rpmsg_release_channel(ctrldev->rpdev, &chinfo);
> >> + if (ret)
> >> + dev_err(&ctrldev->dev, "failed to release %s channel (%d)\n",
> >> + chinfo.name, ret);
> >> + break;
> >> +
> >> + default:
> >> + ret = -EINVAL;
> >> + }
> >> + mutex_unlock(&ctrldev->ctrl_lock);
> >> +
> >> + return ret;
> >> };
> >>
> >> static const struct file_operations rpmsg_ctrldev_fops = {
> >> @@ -119,6 +147,7 @@ static int rpmsg_ctrldev_probe(struct rpmsg_device *rpdev)
> >> device_initialize(dev);
> >> dev->parent = &rpdev->dev;
> >>
> >> + mutex_init(&ctrldev->ctrl_lock);
> >> cdev_init(&ctrldev->cdev, &rpmsg_ctrldev_fops);
> >> ctrldev->cdev.owner = THIS_MODULE;
> >>
> >> diff --git a/include/uapi/linux/rpmsg.h b/include/uapi/linux/rpmsg.h
> >> index f5ca8740f3fb..1637e68177d9 100644
> >> --- a/include/uapi/linux/rpmsg.h
> >> +++ b/include/uapi/linux/rpmsg.h
> >> @@ -33,4 +33,14 @@ struct rpmsg_endpoint_info {
> >> */
> >> #define RPMSG_DESTROY_EPT_IOCTL _IO(0xb5, 0x2)
> >>
> >> +/**
> >> + * Instantiate a new local rpmsg service device.
> >> + */
> >> +#define RPMSG_CREATE_DEV_IOCTL _IOW(0xb5, 0x3, struct rpmsg_endpoint_info)
> >> +
> >> +/**
> >> + * Release a local rpmsg device.
> >> + */
> >> +#define RPMSG_RELEASE_DEV_IOCTL _IOW(0xb5, 0x4, struct rpmsg_endpoint_info)
> >> +
> >> #endif
> >> --
> >> 2.17.1
> >>
