Message-ID: <CAK8P3a2e3zNqMJSN-LAAjYmy8Gr=wjn5MMDMinxawOWcMgo7Ww@mail.gmail.com>
Date:   Thu, 4 Mar 2021 22:08:11 +0100
From:   Arnd Bergmann <arnd@...aro.org>
To:     Alex Bennée <alex.bennee@...aro.org>
Cc:     "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        Maxim Uvarov <maxim.uvarov@...aro.org>,
        Joakim Bech <joakim.bech@...aro.org>,
        Ilias Apalodimas <ilias.apalodimas@...aro.org>,
        ruchika.gupta@...aro.org,
        "Winkler, Tomas" <tomas.winkler@...el.com>, yang.huang@...el.com,
        bing.zhu@...el.com, Matti.Moell@...nsynergy.com,
        hmo@...nsynergy.com, linux-mmc <linux-mmc@...r.kernel.org>,
        linux-scsi <linux-scsi@...r.kernel.org>,
        linux-nvme@...r.kernel.org, Ulf Hansson <ulf.hansson@...aro.org>,
        Linus Walleij <linus.walleij@...aro.org>,
        Arnd Bergmann <arnd.bergmann@...aro.org>,
        Alexander Usyskin <alexander.usyskin@...el.com>,
        Avri Altman <avri.altman@...disk.com>
Subject: Re: [RFC PATCH 2/5] char: rpmb: provide a user space interface

On Wed, Mar 3, 2021 at 2:54 PM Alex Bennée <alex.bennee@...aro.org> wrote:
>
> +       /* the rpmb is single open! */
> +       if (test_and_set_bit(RPMB_DEV_OPEN, &rdev->status))
> +               return -EBUSY;

open counters on device nodes are fundamentally broken, because
they do not stop you from using dup() or sharing the file descriptor
across a fork. Just remove this.

> +static long rpmb_ioctl_ver_cmd(struct rpmb_dev *rdev,
> +                              struct rpmb_ioc_ver_cmd __user *ptr)
> +{
> +       struct rpmb_ioc_ver_cmd ver = {
> +               .api_version = RPMB_API_VERSION,
> +       };
> +
> +       return copy_to_user(ptr, &ver, sizeof(ver)) ? -EFAULT : 0;
> +}

Similarly, API versions are fundamentally flawed, as the kernel requires
us to keep compatibility with existing user space. Remove this as well.

> +static long rpmb_ioctl_cap_cmd(struct rpmb_dev *rdev,
> +                              struct rpmb_ioc_cap_cmd __user *ptr)
> +{
> +       struct rpmb_ioc_cap_cmd cap;

Better do a memset() here to ensure this does not leak kernel
stack data to user space.


> +static const struct file_operations rpmb_fops = {
> +       .open           = rpmb_open,
> +       .release        = rpmb_release,
> +       .unlocked_ioctl = rpmb_ioctl,
> +       .owner          = THIS_MODULE,
> +       .llseek         = noop_llseek,
> +};

Add

       .compat_ioctl = compat_ptr_ioctl

to make it work for 32-bit user space on 64-bit kernels.

> @@ -0,0 +1,17 @@
> +/* SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0 */
> +/*
> + * Copyright (C) 2015-2018 Intel Corp. All rights reserved
> + */
> +#ifdef CONFIG_RPMB_INTF_DEV
> +int __init rpmb_cdev_init(void);
> +void __exit rpmb_cdev_exit(void);
> +void rpmb_cdev_prepare(struct rpmb_dev *rdev);
> +void rpmb_cdev_add(struct rpmb_dev *rdev);
> +void rpmb_cdev_del(struct rpmb_dev *rdev);
> +#else
> +static inline int __init rpmb_cdev_init(void) { return 0; }

I don't think it's necessary to make the user interface optional,
I'd just always provide these.

>
> +#define RPMB_API_VERSION 0x80000001

Remove this

> + */
> +struct rpmb_ioc_ver_cmd {
> +       __u32 api_version;
> +} __packed;

And this

> +
> +enum rpmb_auth_method {
> +       RPMB_HMAC_ALGO_SHA_256 = 0,
> +};
> +
> +/**
> + * struct rpmb_ioc_cap_cmd - rpmb capabilities
> + *
> + * @target: rpmb target/region within RPMB partition.
> + * @capacity: storage capacity (in units of 128K)
> + * @block_size: storage data block size (in units of 256B)
> + * @wr_cnt_max: maximal number of block that can be written in a single request.
> + * @rd_cnt_max: maximal number of block that can be read in a single request.
> + * @auth_method: authentication method: currently always HMAC_SHA_256
> + * @reserved: reserved to align to 4 bytes.
> + */
> +struct rpmb_ioc_cap_cmd {
> +       __u16 target;
> +       __u16 capacity;
> +       __u16 block_size;
> +       __u16 wr_cnt_max;
> +       __u16 rd_cnt_max;
> +       __u16 auth_method;
> +       __u16 reserved;
> +} __attribute__((packed));
> +

Remove the packed attribute, it does not change the structure layout but
just makes it less efficient to access on architectures that turn unaligned
loads and stores into byte accesses.

> +/**
> + * struct rpmb_ioc_blocks_cmd - read/write blocks to/from RPMB
> + *
> + * @keyid: key_serial_t of key to use
> + * @addr: index into device (units of 256B blocks)
> + * @count: number of 256B blocks
> + * @data: pointer to data to write/read
> + */
> +struct rpmb_ioc_blocks_cmd {
> +       __s32 key; /* key_serial_t */
> +       __u32 addr;
> +       __u32 count;
> +       __u8 __user *data;
> +} __attribute__((packed));

ioctl structures should generally not have pointers in them. If this can be done
one block at a time, you can have the 256 bytes as part of the structure.

This probably needs a redesign anyway based on Tomas' feedback though.

If you end up needing a pointer, use a __u64 member  with
u64_to_user_ptr() as described in Documentation/driver-api/ioctl.rst

> +#define RPMB_IOC_VER_CMD     _IOR(0xB8, 80, struct rpmb_ioc_ver_cmd)
> +#define RPMB_IOC_CAP_CMD     _IOR(0xB8, 81, struct rpmb_ioc_cap_cmd)
> +#define RPMB_IOC_PKEY_CMD    _IOW(0xB8, 82, key_serial_t)
> +#define RPMB_IOC_COUNTER_CMD _IOR(0xB8, 84, int)
> +#define RPMB_IOC_WBLOCKS_CMD _IOW(0xB8, 85, struct rpmb_ioc_blocks_cmd)
> +#define RPMB_IOC_RBLOCKS_CMD _IOR(0xB8, 86, struct rpmb_ioc_blocks_cmd)

The last one should be _IOWR(), not _IOR(), since you write the metadata and
read the data.

      Arnd

Powered by blists - more mailing lists