[PATCH] kmsg: limit message size

[PATCH] kmsg: limit message size

[Sasha Levin]

There are no size checks in kmsg_write(), and we try allocating enough
memory to store everything userspace gave us, which may be too much for
kmalloc to allocate.

Furthermore, we can have an integer overflow if len==INT_MAX, in that case
we'll corrupt kernel memory.

This was tested with several userspace programs that write to kmsg, and haven't
found a case where the program attempts to write more than PAGE_SIZE.

Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
---
 drivers/char/mem.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/drivers/char/mem.c b/drivers/char/mem.c
index d6e9d08..c90964b 100644
--- a/drivers/char/mem.c
+++ b/drivers/char/mem.c
@@ -815,6 +815,9 @@ static ssize_t kmsg_writev(struct kiocb *iocb, const struct iovec *iv,
 	ssize_t ret = -EFAULT;
 	size_t len = iov_length(iv, count);
 
+	if (len > PAGE_SIZE)
+		return -E2BIG;
+
 	line = kmalloc(len + 1, GFP_KERNEL);
 	if (line == NULL)
 		return -ENOMEM;
-- 
1.7.8.5

Re: [PATCH] kmsg: limit message size

[Andrew Morton]

On Sat,  5 May 2012 16:56:12 +0200
Sasha Levin <levinsasha928@gmail.com> wrote:

> There are no size checks in kmsg_write(), and we try allocating enough
> memory to store everything userspace gave us, which may be too much for
> kmalloc to allocate.
> 
> Furthermore, we can have an integer overflow if len==INT_MAX, in that case
> we'll corrupt kernel memory.
> 
> This was tested with several userspace programs that write to kmsg, and haven't
> found a case where the program attempts to write more than PAGE_SIZE.
> 
> Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
> ---
>  drivers/char/mem.c |    3 +++
>  1 files changed, 3 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/char/mem.c b/drivers/char/mem.c
> index d6e9d08..c90964b 100644
> --- a/drivers/char/mem.c
> +++ b/drivers/char/mem.c
> @@ -815,6 +815,9 @@ static ssize_t kmsg_writev(struct kiocb *iocb, const struct iovec *iv,
>  	ssize_t ret = -EFAULT;
>  	size_t len = iov_length(iv, count);
>  
> +	if (len > PAGE_SIZE)
> +		return -E2BIG;
> +
>  	line = kmalloc(len + 1, GFP_KERNEL);
>  	if (line == NULL)
>  		return -ENOMEM;

Well, this is a write(), and write() is permitted to return
less-than-asked-for.  So what we could do here is to write the first N
bytes and then return N to userspace.  Well-behaved userspace will
notice this and then do some more writing from offset N.

Could I ask that you try this and test it a bit, see whether there is
any well-behaved userspace out there?  Things like cat and echo
_should_ dtrt.

Re: [PATCH] kmsg: limit message size

[Sasha Levin]

On Wed, May 9, 2012 at 12:57 AM, Andrew Morton
<akpm@linux-foundation.org> wrote:
> On Sat,  5 May 2012 16:56:12 +0200
> Sasha Levin <levinsasha928@gmail.com> wrote:
>
>> There are no size checks in kmsg_write(), and we try allocating enough
>> memory to store everything userspace gave us, which may be too much for
>> kmalloc to allocate.
>>
>> Furthermore, we can have an integer overflow if len==INT_MAX, in that case
>> we'll corrupt kernel memory.
>>
>> This was tested with several userspace programs that write to kmsg, and haven't
>> found a case where the program attempts to write more than PAGE_SIZE.
>>
>> Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
>> ---
>>  drivers/char/mem.c |    3 +++
>>  1 files changed, 3 insertions(+), 0 deletions(-)
>>
>> diff --git a/drivers/char/mem.c b/drivers/char/mem.c
>> index d6e9d08..c90964b 100644
>> --- a/drivers/char/mem.c
>> +++ b/drivers/char/mem.c
>> @@ -815,6 +815,9 @@ static ssize_t kmsg_writev(struct kiocb *iocb, const struct iovec *iv,
>>       ssize_t ret = -EFAULT;
>>       size_t len = iov_length(iv, count);
>>
>> +     if (len > PAGE_SIZE)
>> +             return -E2BIG;
>> +
>>       line = kmalloc(len + 1, GFP_KERNEL);
>>       if (line == NULL)
>>               return -ENOMEM;
>
> Well, this is a write(), and write() is permitted to return
> less-than-asked-for.  So what we could do here is to write the first N
> bytes and then return N to userspace.  Well-behaved userspace will
> notice this and then do some more writing from offset N.
>
> Could I ask that you try this and test it a bit, see whether there is
> any well-behaved userspace out there?  Things like cat and echo
> _should_ dtrt.

Yup, cat and echo do dtrt. I'll resend a patch which limits the write
instead of canceling it altogether.