Assuming NULL

Assuming NULL

[Jan Engelhardt]

Hi developers,



some places in fs/*.c have conditions like

(namei.c, 238, in "int permission()"):
        if(inode->i_op && inode->i_op->permission)

Others just have
(namei.c, 813, in "int fastcall link_path_walk()"):
        if(!inode->i_op->lookup)

My question is: Which one is right wrt the case "i_op ==/!= NULL"?
There are two ways:

- the kernel assumes i_op (and similar) is always non-NULL
  => then we can remove a lot of checks, like the first example above

- the kernel does not assume...
  => then we need some extra checks, like in the second example above



Jan Engelhardt                                                               
--                                                                            
| Gesellschaft fuer Wissenschaftliche Datenverarbeitung Goettingen,
| Am Fassberg, 37077 Goettingen, www.gwdg.de

Re: Assuming NULL

[Måns Rullgård]

Jan Engelhardt <jengelh@linux01.gwdg.de> writes:

> Hi developers,
>
> some places in fs/*.c have conditions like
>
> (namei.c, 238, in "int permission()"):
>         if(inode->i_op && inode->i_op->permission)
>
> Others just have
> (namei.c, 813, in "int fastcall link_path_walk()"):
>         if(!inode->i_op->lookup)
>
> My question is: Which one is right wrt the case "i_op ==/!= NULL"?
> There are two ways:
>
> - the kernel assumes i_op (and similar) is always non-NULL
>   => then we can remove a lot of checks, like the first example above
>
> - the kernel does not assume...
>   => then we need some extra checks, like in the second example above

And a third:

- in some places it's safe to assume non-NULL, but not always
  => then we need to check only the unsafe places

-- 
Måns Rullgård
mru@inprovide.com

Re: Assuming NULL

[Jan Engelhardt]

>And a third:
>
>- in some places it's safe to assume non-NULL, but not always
>  => then we need to check only the unsafe places

Fine, it would be nice to know where in my module (providing a filesystem) I 
can assume something and where not. (And not only my module.)



Jan Engelhardt                                                               
--                                                                            
| Gesellschaft fuer Wissenschaftliche Datenverarbeitung Goettingen,
| Am Fassberg, 37077 Goettingen, www.gwdg.de

Re: Assuming NULL

[Willy Tarreau]

Hi Jan,

On Sat, Jun 11, 2005 at 06:24:22PM +0200, Jan Engelhardt wrote:
(...)
> My question is: Which one is right wrt the case "i_op ==/!= NULL"?
> There are two ways:
> 
> - the kernel assumes i_op (and similar) is always non-NULL
>   => then we can remove a lot of checks, like the first example above
> 
> - the kernel does not assume...
>   => then we need some extra checks, like in the second example above

and in any case, adding a comment telling why it CAN or why it CANNOT
be NULL would prevent other people from having to redo the same work
in 6 months.

Willy

Re: Assuming NULL

[Christoph Hellwig]

On Sat, Jun 11, 2005 at 06:24:22PM +0200, Jan Engelhardt wrote:
> Hi developers,
> 
> 
> 
> some places in fs/*.c have conditions like
> 
> (namei.c, 238, in "int permission()"):
>         if(inode->i_op && inode->i_op->permission)
> 
> Others just have
> (namei.c, 813, in "int fastcall link_path_walk()"):
>         if(!inode->i_op->lookup)
> 
> My question is: Which one is right wrt the case "i_op ==/!= NULL"?
> There are two ways:
> 
> - the kernel assumes i_op (and similar) is always non-NULL
>   => then we can remove a lot of checks, like the first example above

i_op must not be NULL .alloc_inode() intitializes it to &empty_iops,
and setting it to NULL would be a bug.