[OFFTOPIC] f_pos ?

[OFFTOPIC] f_pos ?

[FabF]

Hi,
	
	I try to understand how readdir process works and I can't understand
f_pos management :

        Having in mind things work that way :

        user : ls
        glibc : 
                open (->sys_open)
                getdentries64 (->sys_getdentries64)
                
        kernel:
                sys_getdentries64
                ->vfs_readdir
                        ->ext2_readdir

At that point, I don't understand why ext2_readdir is playing with
filp->f_pos .... It should be 0 ...Why does it care about offset ?

Thanks in advance for your precious help,
FabF

Re: [OFFTOPIC] f_pos ?

[Maneesh Soni]

On Sun, Jul 04, 2004 at 07:31:29PM +0000, FabF wrote:
> Hi,
> 	
> 	I try to understand how readdir process works and I can't understand
> f_pos management :
> 
>         Having in mind things work that way :
> 
>         user : ls
>         glibc : 
>                 open (->sys_open)
>                 getdentries64 (->sys_getdentries64)
>                 
>         kernel:
>                 sys_getdentries64
>                 ->vfs_readdir
>                         ->ext2_readdir
> 
> At that point, I don't understand why ext2_readdir is playing with
> filp->f_pos .... It should be 0 ...Why does it care about offset ?
> 

I think it may not be 0 all the time. A seekdir() could change could change
the offset to non-zero.

Thanks
Maneesh

-- 
Maneesh Soni
Linux Technology Center, 
IBM Austin
email: maneesh@in.ibm.com
Phone: 1-512-838-1896 Fax: 
T/L : 6781896

Re: [OFFTOPIC] f_pos ? [PATCH 2.6.7-mm6] ext2_readdir commenting

[FabF]

[-- Attachment #1: Type: text/plain, Size: 1274 bytes --]

On Tue, 2004-07-06 at 21:58, Maneesh Soni wrote:
> On Sun, Jul 04, 2004 at 07:31:29PM +0000, FabF wrote:
> > Hi,
> > 	
> > 	I try to understand how readdir process works and I can't understand
> > f_pos management :
> > 
> >         Having in mind things work that way :
> > 
> >         user : ls
> >         glibc : 
> >                 open (->sys_open)
> >                 getdentries64 (->sys_getdentries64)
> >                 
> >         kernel:
> >                 sys_getdentries64
> >                 ->vfs_readdir
> >                         ->ext2_readdir
> > 
> > At that point, I don't understand why ext2_readdir is playing with
> > filp->f_pos .... It should be 0 ...Why does it care about offset ?
> > 
> 
> I think it may not be 0 all the time. A seekdir() could change could change
> the offset to non-zero.
Thanks Maneesh :) Anton brightly explained me readdir : it can be called
more than once (buffering ...) so the reason for that offset.Action
taking place in filldir return.'over' in ext2 readdir could bring
confusion though....It's 'over' for kernel duty but the whole path is
not necessarily explored yet.

I think the trivial comment patch attached (against 2.6.7-mm6) could be
interesting for that issue.

Regards,
FabF

> 
> Thanks
> Maneesh

[-- Attachment #2: ext2readdir_comment.patch --]
[-- Type: text/x-patch, Size: 619 bytes --]

ext2_readdir commenting

Signed off by FabF
---


diff -puN fs/ext2/dir.c~ext2readdir_comment fs/ext2/dir.c
--- linux-2.6.7/fs/ext2/dir.c~ext2readdir_comment	2004-07-06 22:20:14.514980596 +0200
+++ linux-2.6.7-heatwave/fs/ext2/dir.c	2004-07-06 22:31:52.822458517 +0200
@@ -245,6 +245,12 @@ static inline void ext2_set_de_type(ext2
 		de->file_type = 0;
 }
 
+/*
+ *	ext2_readdir()
+ *
+ * copy directory entries to userland from its current offset (filp->f_pos)
+ * up to the end or when buffer is full (over=-EFAULT | -EINVAL)
+ */
 static int
 ext2_readdir (struct file * filp, void * dirent, filldir_t filldir)
 {
_

Re: [OFFTOPIC] f_pos ?

[FabF]

On Tue, 2004-07-06 at 21:58, Maneesh Soni wrote:
> On Sun, Jul 04, 2004 at 07:31:29PM +0000, FabF wrote:
> > Hi,
> > 	
> > 	I try to understand how readdir process works and I can't understand
> > f_pos management :
> > 
> >         Having in mind things work that way :
> > 
> >         user : ls
> >         glibc : 
> >                 open (->sys_open)
> >                 getdentries64 (->sys_getdentries64)
> >                 
> >         kernel:
> >                 sys_getdentries64
> >                 ->vfs_readdir
> >                         ->ext2_readdir
> > 
> > At that point, I don't understand why ext2_readdir is playing with
> > filp->f_pos .... It should be 0 ...Why does it care about offset ?
> > 
> 
> I think it may not be 0 all the time. A seekdir() could change could change
> the offset to non-zero.
> 
btw, someone could tell why old ext2 readdir had "do the readahead"
feature ; current ext3 as well but new ext2 implementation doesn't ???

Regards,
FabF

> Thanks
> Maneesh