Using memcpy instead of copy_to_user. xprtrdma code.

Using memcpy instead of copy_to_user. xprtrdma code.

[Aleksey Senin]

This specific patch -  - implements using memcpy instead of
copy_to_user. Why is it considered to be safe in this specific case?
All readings about how to copy data are mentioning to use
copy_to_user/copy_from_user. Why use direct copy here? What prevents
the kernel from failure if the page is not present or doesn't have
required access rights?

@@ -103,8 +102,8 @@ static int read_reset_stat(struct ctl_table
*table, int write,
                len -= *ppos;
                if (len > *lenp)
                        len = *lenp;
-               if (len && copy_to_user(buffer, str_buf, len))
-                       return -EFAULT;
+               if (len)
+                       memcpy(buffer, str_buf, len);

Reference to the specific patch in the services of commits:
https://lkml.org/lkml/2020/4/17/60

Commit itself:
32927393dc1ccd60fb2bdc05b9e8e88753761469

RE: Using memcpy instead of copy_to_user. xprtrdma code.

[David Laight]

From: Aleksey Senin <aleksey-linux-kernel@senin.name>
> Sent: 25 January 2022 18:04
> 
> This specific patch -  - implements using memcpy instead of
> copy_to_user. Why is it considered to be safe in this specific case?
> All readings about how to copy data are mentioning to use
> copy_to_user/copy_from_user. Why use direct copy here? What prevents
> the kernel from failure if the page is not present or doesn't have
> required access rights?
> 
> @@ -103,8 +102,8 @@ static int read_reset_stat(struct ctl_table
> *table, int write,
>                 len -= *ppos;
>                 if (len > *lenp)
>                         len = *lenp;
> -               if (len && copy_to_user(buffer, str_buf, len))
> -                       return -EFAULT;
> +               if (len)
> +                       memcpy(buffer, str_buf, len);
> 
> Reference to the specific patch in the services of commits:
> https://lkml.org/lkml/2020/4/17/60

Read the commit message.

> 
> Commit itself:
> 32927393dc1ccd60fb2bdc05b9e8e88753761469

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

Re: Using memcpy instead of copy_to_user. xprtrdma code.

[Aleksey Senin]

Hi,
It seems clue is in new proc_sys_call_handler code that allocates the
kernel buffer first, calls to particular function and than copying
obtained buffer to user space using copy_to_user. Right?

On Tue, Jan 25, 2022 at 4:21 PM David Laight <David.Laight@aculab.com> wrote:
>
> From: Aleksey Senin <aleksey-linux-kernel@senin.name>
> > Sent: 25 January 2022 18:04
> >
> > This specific patch -  - implements using memcpy instead of
> > copy_to_user. Why is it considered to be safe in this specific case?
> > All readings about how to copy data are mentioning to use
> > copy_to_user/copy_from_user. Why use direct copy here? What prevents
> > the kernel from failure if the page is not present or doesn't have
> > required access rights?
> >
> > @@ -103,8 +102,8 @@ static int read_reset_stat(struct ctl_table
> > *table, int write,
> >                 len -= *ppos;
> >                 if (len > *lenp)
> >                         len = *lenp;
> > -               if (len && copy_to_user(buffer, str_buf, len))
> > -                       return -EFAULT;
> > +               if (len)
> > +                       memcpy(buffer, str_buf, len);
> >
> > Reference to the specific patch in the services of commits:
> > https://lkml.org/lkml/2020/4/17/60
>
> Read the commit message.
>
> >
> > Commit itself:
> > 32927393dc1ccd60fb2bdc05b9e8e88753761469
>
> -
> Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
> Registration No: 1397386 (Wales)