[linux-cifs-client] [patch] skip posix open if nameidata is null

[Shirish Pargaonkar]

On Fri, Apr 2, 2010 at 5:44 PM, Jeff Layton <jlayton at samba.org> wrote:
> On Fri,  2 Apr 2010 12:32:58 -0500
> shirishpargaonkar at gmail.com wrote:
>
>> While creating a file on a server which supports unix extensions
>> such as Samba, if a file is being created which does not supply
>> nameidata (i.e. nd is null), cifs client can oops when calling
>> cifs_posix_open.
>>
>> The simplest solution is, do not open a file using posix semantics
>> if nameidata parameter is NULL even if server supports posix semantics.
>>
>> I do not see a way to reach vfsmount (structure) field if nameidata
>> is not supplied to call posix open (cifs_posix_open).
>>
>>
>> Signed-off-by: Shirish Pargaonkar <shirishpargaonkar at gmail.com>
>> Reported-by: Eugene Teo <eugeneteo at kernel.sg>
>> ---
>>
>> diff --git a/fs/cifs/dir.c b/fs/cifs/dir.c
>> index e9f7ecc..eef8d83 100644
>> --- a/fs/cifs/dir.c
>> +++ b/fs/cifs/dir.c
>> @@ -317,7 +317,7 @@ cifs_create(struct inode *inode, struct dentry *direntry, int mode,
>>       else
>>               oflags = FMODE_READ;
>>
>> -     if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
>> +     if (nd && tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
>>           (CIFS_UNIX_POSIX_PATH_OPS_CAP &
>>                       le64_to_cpu(tcon->fsUnixInfo.Capability))) {
>>               rc = cifs_posix_open(full_path, &newinode, nd->path.mnt,
>
> There's nothing inherently special about posix opens that prevents them
> from working when nd is NULL. Instead of doing this, it would probably
> be preferable to have posix opens do what "regular" opens do when nd is
> NULL -- close the file before returning from cifs_create.
>
> --
> Jeff Layton <jlayton at samba.org>
>

Agree that just because nd is null, caller should not miss out on
posix semantics, but the way code is written
now, there is no way to access super_block within cifs_posix_open if
nd is NULL, will have to modify
cifs_posix_open to expect both vfsmount and super_block when invoked
and handle null nd/vfsmount case
within.

Regards,

Shirish