Re: [fuse-devel] invalidating children dentries on changing directory permissions

[Jean-Pierre A. <jea...@wa...>]

Hi,

Satya Prakash GS wrote:
>>In this situation your reply by
>>fuse_reply_attr() defines the new permissions to enter into
>>the cached entries and the kernel is supposed to check the
>>full path for any subsequent access to the file. You do not
>>have to do anything special.
>
> I double-checked this and we are sending correct attributes as
> part of reply. After the 3 second timeout things are fine but within the
> 3 second window ls -l seems to be displaying the attributes of the children.

That is a bad bug.

The more bad as I already reported it more than six years
ago on this list (with a patch proposal which was wrong)
http://article.gmane.org/gmane.comp.file-systems.fuse.devel/8099
http://article.gmane.org/gmane.comp.file-systems.fuse.devel/8119

This was one of the reasons why ntfs-3g gave up on
using the fuse cache.

Jean-Pierre

>
>
> On Jan 28 2016, Satya Prakash GS <g.satyaprakash-Re5JQEeQqe8AvxtiuMwx3w@...> wrote:
>>> Is there an api in low-level which will invalidate all the dentries of
>>> the children in a directory.
>
>>No, but you can invalidate them one-by-one using fuse_notify_inval*.
>
> But this is an expensive operation if I have to walk over all the files (or the ones which have
> lookup count > 0) in a directory.
>
>
>>> When setattr is called on a directory with execute
>>> permissions revoked for a particular user, further lookups of children in
>>> the directory should not be served from kernel. How can this be
>>> achieved.
>
>>The kernel ought to reset its cache automatically in this situation.
>
> Looks like this isn't happening within the lookup timeout window. I am not a kernel
> expert can somebody please confirm this behaviour.
>
>>Best,
>>-Nikolaus
>
> Thanks,
> Satya.