LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: Christian Brauner <christian@brauner.io>
To: Amir Goldstein <amir73il@gmail.com>
Cc: Jan Kara <jack@suse.cz>,
	linux-fsdevel <linux-fsdevel@vger.kernel.org>,
	linux-kernel <linux-kernel@vger.kernel.org>,
	Matthew Bobrowski <mbobrowski@mbobrowski.org>
Subject: Re: [PATCH] fanotify: remove redundant capable(CAP_SYS_ADMIN)s
Date: Thu, 23 May 2019 13:58:46 +0200	[thread overview]
Message-ID: <20190523115845.w7neydaka5xivwyi@brauner.io> (raw)
In-Reply-To: <CAOQ4uxji4jRvJnLvXe0yR4Ls7VxM_tjAypX1TqBe5FYr_7GnXw@mail.gmail.com>

On Thu, May 23, 2019 at 02:40:39PM +0300, Amir Goldstein wrote:
> On Thu, May 23, 2019 at 1:42 PM Christian Brauner <christian@brauner.io> wrote:
> >
> > On Thu, May 23, 2019 at 01:25:08PM +0300, Amir Goldstein wrote:
> > > On Thu, May 23, 2019 at 12:55 PM Christian Brauner <christian@brauner.io> wrote:
> > > >
> > > > On Wed, May 22, 2019 at 11:00:22PM +0300, Amir Goldstein wrote:
> > > > > On Wed, May 22, 2019 at 9:57 PM Christian Brauner <christian@brauner.io> wrote:
> > > > > >
> > > > > > On May 22, 2019 8:29:37 PM GMT+02:00, Amir Goldstein <amir73il@gmail.com> wrote:
> > > > > > >On Wed, May 22, 2019 at 7:32 PM Christian Brauner
> > > > > > ><christian@brauner.io> wrote:
> > > > > > >>
> > > > > > >> This removes two redundant capable(CAP_SYS_ADMIN) checks from
> > > > > > >> fanotify_init().
> > > > > > >> fanotify_init() guards the whole syscall with capable(CAP_SYS_ADMIN)
> > > > > > >at the
> > > > > > >> beginning. So the other two capable(CAP_SYS_ADMIN) checks are not
> > > > > > >needed.
> > > > > > >
> > > > > > >It's intentional:
> > > > > > >
> > > > > > >commit e7099d8a5a34d2876908a9fab4952dabdcfc5909
> > > > > > >Author: Eric Paris <eparis@redhat.com>
> > > > > > >Date:   Thu Oct 28 17:21:57 2010 -0400
> > > > > > >
> > > > > > >    fanotify: limit the number of marks in a single fanotify group
> > > > > > >
> > > > > > >There is currently no limit on the number of marks a given fanotify
> > > > > > >group
> > > > > > >can have.  Since fanotify is gated on CAP_SYS_ADMIN this was not seen
> > > > > > >as
> > > > > > >a serious DoS threat.  This patch implements a default of 8192, the
> > > > > > >same as
> > > > > > >inotify to work towards removing the CAP_SYS_ADMIN gating and
> > > > > > >eliminating
> > > > > > >    the default DoS'able status.
> > > > > > >
> > > > > > >    Signed-off-by: Eric Paris <eparis@redhat.com>
> > > > > > >
> > > > > > >There idea is to eventually remove the gated CAP_SYS_ADMIN.
> > > > > > >There is no reason that fanotify could not be used by unprivileged
> > > > > > >users
> > > > > > >to setup inotify style watch on an inode or directories children, see:
> > > > > > >https://patchwork.kernel.org/patch/10668299/
> > > > > > >
> > > > > > >>
> > > > > > >> Fixes: 5dd03f55fd2 ("fanotify: allow userspace to override max queue
> > > > > > >depth")
> > > > > > >> Fixes: ac7e22dcfaf ("fanotify: allow userspace to override max
> > > > > > >marks")
> > > > > > >
> > > > > > >Fixes is used to tag bug fixes for stable.
> > > > > > >There is no bug.
> > > > > > >
> > > > > > >Thanks,
> > > > > > >Amir.
> > > > > >
> > > > > > Interesting. When do you think the gate can be removed?
> > > > >
> > > > > Nobody is working on this AFAIK.
> > > > > What I posted was a simple POC, but I have no use case for this.
> > > > > In the patchwork link above, Jan has listed the prerequisites for
> > > > > removing the gate.
> > > > >
> > > > > One of the prerequisites is FAN_REPORT_FID, which is now merged.
> > > > > When events gets reported with fid instead of fd, unprivileged user
> > > > > (hopefully) cannot use fid for privilege escalation.
> > > > >
> > > > > > I was looking into switching from inotify to fanotify but since it's not usable from
> > > > > > non-initial userns it's a no-no
> > > > > > since we support nested workloads.
> > > > >
> > > > > One of Jan's questions was what is the benefit of using inotify-compatible
> > > > > fanotify vs. using inotify.
> > > > > So what was the reason you were looking into switching from inotify to fanotify?
> > > > > Is it because of mount/filesystem watch? Because making those available for
> > > >
> > > > Yeah. Well, I would need to look but you could probably do it safely for
> > > > filesystems mountable in user namespaces (which are few).
> > > > Can you do a bind-mount and then place a watch on the bind-mount or is
> > > > this superblock based?
> > > >
> > >
> > > Either.
> > > FAN_MARK_MOUNT was there from day 1 of fanotify.
> > > FAN_MARK_FILESYSTEM was merged to Linux Linux 4.20.
> > >
> > > But directory modification events that are supported since v5.1 are
> > > not available
> > > with FAN_MARK_MOUNT, see:
> >
> > Because you're worried about unprivileged users spying on events? Or
> > something else?
> 
> Something else. The current fsnotify_move/create/delete() VFS hooks
> have no path/mount information, so it is not possible to filter them by
> mount only by inode/sb.
> Fixing that would not be trivial, but first a strong use case would need
> to be presented.
> 
> > Because if you can do a bind-mount there's nothing preventing an
> > unprivileged user to do a hand-rolled recursive inotify that would
> > amount to the same thing anyway.
> 
> There is. unprivileged user cannot traverse into directories it is not
> allowed to read/search.

Right, I should've mentioned: when you're userns root and you have
access to all files. The part that is interesting to me is getting rid
of capable(CAP_SYS_ADMIN).

> 
> > (And btw, v5.1 really is a major step forward and I would really like to
> >  use this api tbh.)
> >
> 
> You haven't answered my question. What is the reason you are interested
> in the new API? What does it provide that the old API does not?
> I know the 2 APIs differ. I just want to know which difference interests *you*,
> because without a strong use case, it will be hard for me to make progress
> upstream.
> 
> Is what you want really a "bind-mount" watch or a "subtree watch"?
> The distinction is important. I am thinking about solutions for the latter,
> although there is no immediate solution in the horizon - only ideas.

Both cases would be interesting. But subtree watch is what would
probably help a lot already. So let me explain.
For LXD - not sure if you know what that is - we allow user to "hotplug"
mounts or certain whitelisted devices into a user namespace container.
One of the nifty features is that we let users specify a "required"
property. When "required" is "false" the user can give us a path, e.g.
/bla/bla/bla/target and then we place a watch on the closest existing
ancestor of my-device. When the target shows up we hotplug it for the
user. Now, as you imagine maintaining that cache until "target" shows up
is a royal pain.

I think that we can get rid of at least some of the complexity if
subtree watch and bind-mount watches would work.

Thanks!
Christian

  reply	other threads:[~2019-05-23 11:58 UTC|newest]

Thread overview: 15+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-05-22 16:31 Christian Brauner
2019-05-22 18:29 ` Amir Goldstein
2019-05-22 18:57   ` Christian Brauner
2019-05-22 20:00     ` Amir Goldstein
2019-05-23  9:55       ` Christian Brauner
2019-05-23 10:25         ` Amir Goldstein
2019-05-23 10:42           ` Christian Brauner
2019-05-23 11:40             ` Amir Goldstein
2019-05-23 11:58               ` Christian Brauner [this message]
2019-05-23 13:16                 ` Amir Goldstein
2019-05-23 13:35                   ` Christian Brauner
2019-05-23 14:40                     ` Jan Kara
2019-05-23 14:43                       ` Christian Brauner
2019-05-23 15:15                       ` Amir Goldstein
2019-06-05 10:26           ` Matthew Bobrowski

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20190523115845.w7neydaka5xivwyi@brauner.io \
    --to=christian@brauner.io \
    --cc=amir73il@gmail.com \
    --cc=jack@suse.cz \
    --cc=linux-fsdevel@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mbobrowski@mbobrowski.org \
    --subject='Re: [PATCH] fanotify: remove redundant capable(CAP_SYS_ADMIN)s' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).