Re: [PATCH] eventfs: Stop using dcache_readdir() for getdents()

[Steven Rostedt]

On Wed, 3 Jan 2024 10:38:09 -0800
Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:

> On Wed, 3 Jan 2024 at 10:12, Linus Torvalds
> <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> >
> > Much better. Now eventfs looks more like a real filesystem, and less
> > like an eldritch horror monster that is parts of dcache tackled onto a
> > pseudo-filesystem.  
> 
> Oh, except I think you still need to just remove the 'set_gid()' mess.
> 
> It's disgusting and it's wrong, and it's not even what the 'uid'
> option does (it only sets the root inode uid).
> 
> If you remount the filesystem with different gid values, you get to
> keep both broken pieces. And if it isn't a remount, then setting the
> root uid is sufficient.
> 
> I think the whole thing was triggered by commit 49d67e445742, and
> maybe the fix is to just revert that commit.
> 
> That commit makes no sense in general, since the default mounting
> position for tracefs that the kernel sets up is only accessible to
> root anyway.
> 
> Alternatively, just do the ->permissions() thing, and allow access to
> the group in the mount options.
> 
> Getting rid of set_gid() would be this attached lovely patch:
> 
>  fs/tracefs/inode.c | 83 ++----------------------------------------------------
>  1 file changed, 2 insertions(+), 81 deletions(-)
> 
> and would get rid of the final (?) piece of disgusting dcache hackery
> that tracefs most definitely should not have.
>

I'll look at that and play with it. I understand VFS much better now that I
spent so much time with eventfs. That commit had to do with allowing OTH
read access, which is a security issue as the trace files expose a lot of
the kernel internals.

I think these changes are a bit much for -rc8, don't you? Or do you want
all this in before v6.7 is released. I'd be more comfortable with adding
these changes in the upcoming merge window, where I can have more time
playing with them.

-- Steve