Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <186c5de2-a07c-7401-bf8e-863defaff5ba@schaufler-ca.com>
Date:   Mon, 3 Apr 2023 11:33:51 -0700
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:102.0) Gecko/20100101
 Thunderbird/102.9.1
Subject: Re: [PATCH v7 07/11] LSM: Helpers for attribute names and filling an
 lsm_ctx
Content-Language: en-US
To:     =?UTF-8?Q?Micka=c3=abl_Sala=c3=bcn?= <mic@digikod.net>,
        paul@paul-moore.com, linux-security-module@vger.kernel.org
Cc:     jmorris@namei.org, keescook@chromium.org,
        john.johansen@canonical.com, penguin-kernel@i-love.sakura.ne.jp,
        stephen.smalley.work@gmail.com, linux-kernel@vger.kernel.org,
        linux-api@vger.kernel.org, Casey Schaufler <casey@schaufler-ca.com>
References: <20230315224704.2672-1-casey@schaufler-ca.com>
 <20230315224704.2672-8-casey@schaufler-ca.com>
 <544a4809-1a79-9dd7-61a5-5fce1f4a5f10@digikod.net>
 <bbd6abb7-7443-0f34-788b-0e441dc5afee@schaufler-ca.com>
 <6af85c04-0388-026a-e4ee-08d9aad6b86a@digikod.net>
From:   Casey Schaufler <casey@schaufler-ca.com>
In-Reply-To: <6af85c04-0388-026a-e4ee-08d9aad6b86a@digikod.net>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 4/3/2023 11:06 AM, Mickaël Salaün wrote:
>
> On 03/04/2023 20:03, Casey Schaufler wrote:
>> On 4/3/2023 2:47 AM, Mickaël Salaün wrote:
>>>
>>> On 15/03/2023 23:47, Casey Schaufler wrote:
>>>> Add lsm_name_to_attr(), which translates a text string to a
>>>> LSM_ATTR value if one is available.
>>>>
>>>> Add lsm_fill_user_ctx(), which fills a struct lsm_ctx, including
>>>> the trailing attribute value.
>>>>
>>>> All are used in module specific components of LSM system calls.
>>>>
>>>> Signed-off-by: Casey Schaufler <casey@schaufler-ca.com>
>>>> ---
>>>>    include/linux/security.h | 13 ++++++++++
>>>>    security/lsm_syscalls.c  | 51
>>>> ++++++++++++++++++++++++++++++++++++++++
>>>>    security/security.c      | 31 ++++++++++++++++++++++++
>>>>    3 files changed, 95 insertions(+)
>>>
>>> [...]
>>>
>>>> diff --git a/security/security.c b/security/security.c
>>>> index 2c57fe28c4f7..f7b814a3940c 100644
>>>> --- a/security/security.c
>>>> +++ b/security/security.c
>>>> @@ -753,6 +753,37 @@ static int lsm_superblock_alloc(struct
>>>> super_block *sb)
>>>>        return 0;
>>>>    }
>>>>    +/**
>>>> + * lsm_fill_user_ctx - Fill a user space lsm_ctx structure
>>>> + * @ctx: an LSM context to be filled
>>>> + * @context: the new context value
>>>> + * @context_size: the size of the new context value
>>>> + * @id: LSM id
>>>> + * @flags: LSM defined flags
>>>> + *
>>>> + * Fill all of the fields in a user space lsm_ctx structure.
>>>> + * Caller is assumed to have verified that @ctx has enough space
>>>> + * for @context.
>>>> + * Returns 0 on success, -EFAULT on a copyout error.
>>>> + */
>>>> +int lsm_fill_user_ctx(struct lsm_ctx __user *ctx, void *context,
>>>> +              size_t context_size, u64 id, u64 flags)
>>>> +{
>>>> +    struct lsm_ctx local;
>>>> +    void __user *vc = ctx;
>>>> +
>>>> +    local.id = id;
>>>> +    local.flags = flags;
>>>> +    local.ctx_len = context_size;
>>>> +    local.len = context_size + sizeof(local);
>>>> +    vc += sizeof(local);
>>>> +    if (copy_to_user(ctx, &local, sizeof(local)))
>>>> +        return -EFAULT;
>>>> +    if (context_size > 0 && copy_to_user(vc, context, context_size))
>>>> +        return -EFAULT;
>>>
>>> Can we do a single copy_to_user() call?
>>
>> It would be possible, but would require allocating memory and copying
>> the context. I don't see that as an improvement.
>>
>>> That would avoid inconsistent user space data, could speed up a bit
>>> the operation, and make the code easier to understand. To use the
>>> stack, we need to know the maximum size of context_size for all use
>>> cases, which seems reasonable and can be checked at build time (on
>>> each LSM side, and potentially with specific context type passed as
>>> enum instead of context_size) and run time (for this generic helper).
>>
>> Knowning the maximum size of attributes for all LSMs and hard coding
>> that here would make maintaining this code really painful.
>
> Hmm, I forgot about variable-length strings, but maybe a reasonable
> common maximum size (that could fit on the stack) could be found?

Putting a maximum size limit on LSM attributes just to reduce the
number of copy_to_user() calls in this helper function does not make
a whole lot of sense to me.

>
>>
>>>
>>>
>>>> +    return 0;
>>>> +}
>>>> +
>>>>    /*
>>>>     * The default value of the LSM hook is defined in
>>>> linux/lsm_hook_defs.h and
>>>>     * can be accessed with: