2023-08-07 21:47:01

by Bart Van Assche

[permalink] [raw]
Subject: Re: [RFC, PATCH 1/1] rpmb: add Replay Protected Memory Block (RPMB) driver

On 7/21/23 18:40, Shyam Saini wrote:
> +config RPMB
> + tristate "RPMB partition interface"
> + help
> + Unified RPMB partition interface for RPMB capable devices such as
> + eMMC and UFS. Provides interface for in kernel security controllers to
> + access RPMB partition.
> +
> + If unsure, select N.

Please also mention NVMe.

Please change the word "partition" into "unit" to avoid confusion with
the concept "LBA range partition".

> +static DEFINE_IDA(rpmb_ida);

How are accesses to this IDA serialized?

> +/**
> + * rpmb_get_capacity() - returns the capacity of the rpmb device
> + * @rdev: rpmb device
> + *
> + * Return:
> + * * capacity of the device in units of 128K, on success
> + * * -EINVAL on wrong parameters
> + * * -EOPNOTSUPP if device doesn't support the requested operation
> + * * < 0 if the operation fails
> + */

Why in units of 128 KiB?

> +/**
> + * rpmb_dev_find_by_device() - retrieve rpmb device from the parent device
> + * @parent: parent device of the rpmb device
> + * @target: RPMB target/region within the physical device
> + *
> + * Return: NULL if there is no rpmb device associated with the parent device
> + */

Can an NVMe controller have multiple RPMB units? From the NVMe
specification: "The controller may support multiple RPMB targets."

Can rpmb_dev_find_by_device() be used if multiple RPMB units are
associated with a single controller?

> +/**
> + * rpmb_dev_register - register RPMB partition with the RPMB subsystem
> + * @dev: storage device of the rpmb device
> + * @target: RPMB target/region within the physical device
> + * @ops: device specific operations
> + *
> + * Return: a pointer to rpmb device
> + */
> +struct rpmb_dev *rpmb_dev_register(struct device *dev, u8 target,
> + const struct rpmb_ops *ops)
> +{
> + struct rpmb_dev *rdev;
> + int id;
> + int ret;
> +
> + if (!dev || !ops)
> + return ERR_PTR(-EINVAL);
> +
> + if (!ops->program_key)
> + return ERR_PTR(-EINVAL);
> +
> + if (!ops->get_capacity)
> + return ERR_PTR(-EINVAL);
> +
> + if (!ops->get_write_counter)
> + return ERR_PTR(-EINVAL);
> +
> + if (!ops->write_blocks)
> + return ERR_PTR(-EINVAL);
> +
> + if (!ops->read_blocks)
> + return ERR_PTR(-EINVAL);
> +
> + rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
> + if (!rdev)
> + return ERR_PTR(-ENOMEM);
> +
> + id = ida_simple_get(&rpmb_ida, 0, 0, GFP_KERNEL);
> + if (id < 0) {
> + ret = id;
> + goto exit;
> + }
> +
> + mutex_init(&rdev->lock);
> + rdev->ops = ops;
> + rdev->id = id;
> + rdev->target = target;
> +
> + dev_set_name(&rdev->dev, "rpmb%d", id);
> + rdev->dev.class = &rpmb_class;
> + rdev->dev.parent = dev;
> +
> + rpmb_cdev_prepare(rdev);
> +
> + ret = device_register(&rdev->dev);
> + if (ret)
> + goto exit;
> +
> + rpmb_cdev_add(rdev);
> +
> + dev_dbg(&rdev->dev, "registered device\n");
> +
> + return rdev;
> +
> +exit:
> + if (id >= 0)
> + ida_simple_remove(&rpmb_ida, id);
> + kfree(rdev);
> + return ERR_PTR(ret);
> +}

How is user space software supposed to map an NVMe RPMB target ID to an
RPMB device name?

> +MODULE_AUTHOR("Intel Corporation");

Shouldn't this be the name of a person instead of the name of a company?

Thanks,

Bart.


2023-08-22 21:31:56

by Shyam Saini

[permalink] [raw]
Subject: Re: [RFC, PATCH 1/1] rpmb: add Replay Protected Memory Block (RPMB) driver


Hi Bart,

sorry for not replying earlier, as I am very new to NVMe/UFS spec and was
figuring out few details about them.

> On 7/21/23 18:40, Shyam Saini wrote:
>> +config RPMB
>> + tristate "RPMB partition interface"
>> + help
>> + Unified RPMB partition interface for RPMB capable devices such as
>> + eMMC and UFS. Provides interface for in kernel security
>> controllers to
>> + access RPMB partition.
>> +
>> + If unsure, select N.
>
> Please also mention NVMe.

Sure,
> Please change the word "partition" into "unit" to avoid confusion with the
> concept "LBA range partition".

sure, in next iteration

>> +static DEFINE_IDA(rpmb_ida);
>
> How are accesses to this IDA serialized?

I will look into that.

>> +/**
>> + * rpmb_get_capacity() - returns the capacity of the rpmb device
>> + * @rdev: rpmb device
>> + *
>> + * Return:
>> + * * capacity of the device in units of 128K, on success
>> + * * -EINVAL on wrong parameters
>> + * * -EOPNOTSUPP if device doesn't support the requested operation
>> + * * < 0 if the operation fails
>> + */
>
> Why in units of 128 KiB?

I think UFS/eMMC RPMB spec suggests size of RPMB multiple of 128K
and NVMe spec suggests RPMB Data Area to be multiple of 128K as well.

>> +/**
>> + * rpmb_dev_find_by_device() - retrieve rpmb device from the parent
>> device
>> + * @parent: parent device of the rpmb device
>> + * @target: RPMB target/region within the physical device
>> + *
>> + * Return: NULL if there is no rpmb device associated with the parent
>> device
>> + */
>
> Can an NVMe controller have multiple RPMB units? From the NVMe specification:
> "The controller may support multiple RPMB targets."

That we have to figure, I see NVMe device can have upto 7 RPMB
targets/units

> Can rpmb_dev_find_by_device() be used if multiple RPMB units are associated
> with a single controller?

That's not finalised yet, but we some ideas from Optee folks on the other
replies.

>> +/**
>> + * rpmb_dev_register - register RPMB partition with the RPMB subsystem
>> + * @dev: storage device of the rpmb device
>> + * @target: RPMB target/region within the physical device
>> + * @ops: device specific operations
>> + *
>> + * Return: a pointer to rpmb device
>> + */
>> +struct rpmb_dev *rpmb_dev_register(struct device *dev, u8 target,
>> + const struct rpmb_ops *ops)
>> +{
>> + struct rpmb_dev *rdev;
>> + int id;
>> + int ret;
>> +
>> + if (!dev || !ops)
>> + return ERR_PTR(-EINVAL);
>> +
>> + if (!ops->program_key)
>> + return ERR_PTR(-EINVAL);
>> +
>> + if (!ops->get_capacity)
>> + return ERR_PTR(-EINVAL);
>> +
>> + if (!ops->get_write_counter)
>> + return ERR_PTR(-EINVAL);
>> +
>> + if (!ops->write_blocks)
>> + return ERR_PTR(-EINVAL);
>> +
>> + if (!ops->read_blocks)
>> + return ERR_PTR(-EINVAL);
>> +
>> + rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
>> + if (!rdev)
>> + return ERR_PTR(-ENOMEM);
>> +
>> + id = ida_simple_get(&rpmb_ida, 0, 0, GFP_KERNEL);
>> + if (id < 0) {
>> + ret = id;
>> + goto exit;
>> + }
>> +
>> + mutex_init(&rdev->lock);
>> + rdev->ops = ops;
>> + rdev->id = id;
>> + rdev->target = target;
>> +
>> + dev_set_name(&rdev->dev, "rpmb%d", id);
>> + rdev->dev.class = &rpmb_class;
>> + rdev->dev.parent = dev;
>> +
>> + rpmb_cdev_prepare(rdev);
>> +
>> + ret = device_register(&rdev->dev);
>> + if (ret)
>> + goto exit;
>> +
>> + rpmb_cdev_add(rdev);
>> +
>> + dev_dbg(&rdev->dev, "registered device\n");
>> +
>> + return rdev;
>> +
>> +exit:
>> + if (id >= 0)
>> + ida_simple_remove(&rpmb_ida, id);
>> + kfree(rdev);
>> + return ERR_PTR(ret);
>> +}
>
> How is user space software supposed to map an NVMe RPMB target ID to an RPMB
> device name?

I am not sure, this driver aims to provide in kernel RPMB access APIs,
user space support may be added later on, but i will look if the current
RFC version has any implication on future user-space support.

>> +MODULE_AUTHOR("Intel Corporation");
>
> Shouldn't this be the name of a person instead of the name of a company?
>

Thanks, I will address that in next iteration.

Please keep posted your reviews and feedback.

Best Regards,
Shyam