Re: [PATCH 04/16] vfs: do_last(): use inode variable

[Nick Piggin]

On 8 May 2012 00:28, Miklos Szeredi <miklos@xxxxxxxxxx> wrote:
> Nick Piggin <npiggin@xxxxxxxxx> writes:
>
>> On 25 April 2012 22:44, Miklos Szeredi <miklos@xxxxxxxxxx> wrote:
>>> From: Miklos Szeredi <mszeredi@xxxxxxx>
>>>
>>> Use helper variable instead of path->dentry->d_inode before complete_walk().
>>> This will allow this code to be used in RCU mode.
>>
>> What do you mean, allow it to be used?
>
> I mean allow the code to be shared between RCU and non-RCU mode. ÂSee
> 10/16.
>
>>
>>>
>>> Signed-off-by: Miklos Szeredi <mszeredi@xxxxxxx>
>>> ---
>>> Âfs/namei.c | Â Â7 ++++---
>>> Â1 files changed, 4 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/fs/namei.c b/fs/namei.c
>>> index 46d4bf6..f21ddb3 100644
>>> --- a/fs/namei.c
>>> +++ b/fs/namei.c
>>> @@ -2360,15 +2360,16 @@ static struct file *do_last(struct nameidata *nd, struct path *path,
>>> Â Â Â Âif (error)
>>> Â Â Â Â Â Â Â Ând->flags |= LOOKUP_JUMPED;
>>>
>>> + Â Â Â inode = path->dentry->d_inode ;
>>> Â Â Â Âerror = -ENOENT;
>>> - Â Â Â if (!path->dentry->d_inode)
>>> + Â Â Â if (!inode)
>>> Â Â Â Â Â Â Â Âgoto exit_dput;
>>>
>>> - Â Â Â if (path->dentry->d_inode->i_op->follow_link)
>>> + Â Â Â if (inode->i_op->follow_link)
>>> Â Â Â Â Â Â Â Âreturn NULL;
>>>
>>> Â Â Â Âpath_to_nameidata(path, nd);
>>> - Â Â Â nd->inode = path->dentry->d_inode;
>>> + Â Â Â nd->inode = inode;
>>> Â Â Â Â/* Why this, you ask? Â_Now_ we might have grown LOOKUP_JUMPED... */
>>> Â Â Â Âerror = complete_walk(nd);
>>> Â Â Â Âif (error)
>>
>> In rcu-walk mode, dentry->d_inode should not be accessed at all,
>> outside of the core lookup code that (should) have the correct
>> barriers and sequence locks.
>>
>> That logic should not escape into here, so I'm just not sure what
>> you're doing here.
>
> Right, dentry->d_inode is *not* going to be dereferenced in RCU mode.
> In RCU mode it will jump to the place just after the "inode =
> path->dentry->d_inode;" line. ÂSee patch 10/16.

Hmm, OK, I admittedly didn't apply all patches and look at the result.

I wonder if that's getting too hairy... I'm sure consolidation is worthwhile
though. Perhaps comment or even a BUG_ON to ensure it is not in
rcu-walk mode, at points where you load the inode?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/