[Ocfs2-devel] Invalid file rights

[Ocfs2-devel] Invalid file rights

[Chen, Yukun]

Hi All

I find a file rights bug on kernel 2.4 with svn 1313 code. Steps to
duplicate

Being root, create a file in a ocfs2 partition with the sentence below

 open(filename, O_CREAT | O_RDWR )

Then the file will be created with a right 000 --- nobody can write/read
it at 

all. 

But, if I use "touch filename" , the file will be created with a right
of 644.

Any ideas on it? (a bug # 113 has been reported to oss.oracle.com
bugzilla)

Thanx.


Aaron

Intel China Software Lab

Tel:  8621-52574545 Ext.1587

E_mail:yukun.chen@intel.com

[Ocfs2-devel] Invalid file rights

[Mark Fasheh]

On Tue, Aug 03, 2004 at 03:25:42PM +0800, Chen, Yukun wrote:
> Hi All
> 
> I find a file rights bug on kernel 2.4 with svn 1313 code. Steps to
> duplicate
> 
> Being root, create a file in a ocfs2 partition with the sentence below
> 
>  open(filename, O_CREAT | O_RDWR )
> 
> Then the file will be created with a right 000 --- nobody can write/read
> it at all. 
Right, and we get the mode from the kernel for this, right? Is there some
policy which we're missing? In ocfs_mknod_locked we do "fe->i_mode = mode;"
so it's definitely being set on disk. later on we also do "inode->i_mode =
mode;".

> But, if I use "touch filename" , the file will be created with a right
> of 644.
Well, that'd be because:

[mfasheh@expo ocfs2]$ strace touch file
<snip>
open("file", O_WRONLY|O_NONBLOCK|O_CREAT|O_NOCTTY|O_LARGEFILE, 0666) = 3
<snip>

So that's "touch" setting the mode there.
 
> Any ideas on it? (a bug # 113 has been reported to oss.oracle.com
> bugzilla)
So the "touch" case is obvious, but I'm not sure that the 1st one is a bug
either... If it is, please explain why.
	--Mark

--
Mark Fasheh
Software Developer, Oracle Corp
mark.fasheh@oracle.com

[Ocfs2-devel] Invalid file rights

[Chen, Yukun]

Yeah. You are right.

I take it for granted that a file will be granted some default rights =
for its owner, such as read/write, when it is created with =
open(...,...|O_CREAT) .
I traced the source code of ocfs2 and found the kernel return a mode =
8000 really that means nobody can access the file. Also, I checked it =
with ext3 filesystem and got the same results.

So, I think it is nature and it is not a bug yet.
Thanx.

Aaron

-----Original Message-----
From: Mark Fasheh [mailto:mark.fasheh@oracle.com]=20
Sent: 2004=C4=EA8=D4=C24=C8=D5 2:49
To: Chen, Yukun
Cc: ocfs2-devel@oss.oracle.com
Subject: Re: [Ocfs2-devel] Invalid file rights

On Tue, Aug 03, 2004 at 03:25:42PM +0800, Chen, Yukun wrote:
> Hi All
>=20
> I find a file rights bug on kernel 2.4 with svn 1313 code. Steps to
> duplicate
>=20
> Being root, create a file in a ocfs2 partition with the sentence below
>=20
>  open(filename, O_CREAT | O_RDWR )
>=20
> Then the file will be created with a right 000 --- nobody can =
write/read
> it at all.=20
Right, and we get the mode from the kernel for this, right? Is there =
some
policy which we're missing? In ocfs_mknod_locked we do "fe->i_mode =3D =
mode;"
so it's definitely being set on disk. later on we also do "inode->i_mode =
=3D
mode;".

> But, if I use "touch filename" , the file will be created with a right
> of 644.
Well, that'd be because:

[mfasheh@expo ocfs2]$ strace touch file
<snip>
open("file", O_WRONLY|O_NONBLOCK|O_CREAT|O_NOCTTY|O_LARGEFILE, 0666) =3D =
3
<snip>

So that's "touch" setting the mode there.
=20
> Any ideas on it? (a bug # 113 has been reported to oss.oracle.com
> bugzilla)
So the "touch" case is obvious, but I'm not sure that the 1st one is a =
bug
either... If it is, please explain why.
	--Mark

--
Mark Fasheh
Software Developer, Oracle Corp
mark.fasheh@oracle.com

[Ocfs2-devel] Invalid file rights

[Kurt Hackel]

Yeah I think this is why libc provides...
       int open(const char *pathname, int flags);
       int open(const char *pathname, int flags, mode_t mode);

That "overloaded" open with the mode is only used for the O_CREAT case. 

Thanks!
-kurt

On Wed, Aug 04, 2004 at 09:51:20AM +0800, Chen, Yukun wrote:
> Yeah. You are right.
> 
> I take it for granted that a file will be granted some default rights for its owner, such as read/write, when it is created with open(...,...|O_CREAT) .
> I traced the source code of ocfs2 and found the kernel return a mode 8000 really that means nobody can access the file. Also, I checked it with ext3 filesystem and got the same results.
> 
> So, I think it is nature and it is not a bug yet.
> Thanx.
> 
> Aaron
> 
> -----Original Message-----
> From: Mark Fasheh [mailto:mark.fasheh@oracle.com] 
> Sent: 2004??8??4?? 2:49
> To: Chen, Yukun
> Cc: ocfs2-devel@oss.oracle.com
> Subject: Re: [Ocfs2-devel] Invalid file rights
> 
> On Tue, Aug 03, 2004 at 03:25:42PM +0800, Chen, Yukun wrote:
> > Hi All
> > 
> > I find a file rights bug on kernel 2.4 with svn 1313 code. Steps to
> > duplicate
> > 
> > Being root, create a file in a ocfs2 partition with the sentence below
> > 
> >  open(filename, O_CREAT | O_RDWR )
> > 
> > Then the file will be created with a right 000 --- nobody can write/read
> > it at all. 
> Right, and we get the mode from the kernel for this, right? Is there some
> policy which we're missing? In ocfs_mknod_locked we do "fe->i_mode = mode;"
> so it's definitely being set on disk. later on we also do "inode->i_mode =
> mode;".
> 
> > But, if I use "touch filename" , the file will be created with a right
> > of 644.
> Well, that'd be because:
> 
> [mfasheh@expo ocfs2]$ strace touch file
> <snip>
> open("file", O_WRONLY|O_NONBLOCK|O_CREAT|O_NOCTTY|O_LARGEFILE, 0666) = 3
> <snip>
> 
> So that's "touch" setting the mode there.
>  
> > Any ideas on it? (a bug # 113 has been reported to oss.oracle.com
> > bugzilla)
> So the "touch" case is obvious, but I'm not sure that the 1st one is a bug
> either... If it is, please explain why.
> 	--Mark
> 
> --
> Mark Fasheh
> Software Developer, Oracle Corp
> mark.fasheh@oracle.com
> _______________________________________________
> Ocfs2-devel mailing list
> Ocfs2-devel@oss.oracle.com
> http://oss.oracle.com/mailman/listinfo/ocfs2-devel