Re: [PATCH RFC v11 5/19] ipe: introduce 'boot_verified' as a trust provider

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, Nov 3, 2023 at 6:15 PM Paul Moore <paul@xxxxxxxxxxxxxx> wrote:
> On Thu, Nov 2, 2023 at 6:46 PM Fan Wu <wufan@xxxxxxxxxxxxxxxxxxx> wrote:
> > On 10/26/2023 3:12 PM, Paul Moore wrote:
> > > On Thu, Oct 26, 2023 at 5:33 PM Fan Wu <wufan@xxxxxxxxxxxxxxxxxxx> wrote:
> > >> On 10/23/2023 8:52 PM, Paul Moore wrote:
> > >>> On Oct  4, 2023 Fan Wu <wufan@xxxxxxxxxxxxxxxxxxx> wrote:
> > >>>>
> > >>>> IPE is designed to provide system level trust guarantees, this usually
> > >>>> implies that trust starts from bootup with a hardware root of trust,
> > >>>> which validates the bootloader. After this, the bootloader verifies the
> > >>>> kernel and the initramfs.
> > >>>>
> > >>>> As there's no currently supported integrity method for initramfs, and
> > >>>> it's typically already verified by the bootloader, introduce a property
> > >>>> that causes the first superblock to have an execution to be "pinned",
> > >>>> which is typically initramfs.
> > >>>>
> > >>>> When the "pinned" device is unmounted, it will be "unpinned" and
> > >>>> `boot_verified` property will always evaluate to false afterward.
> > >>>>
> > >>>> We use a pointer with a spin_lock to "pin" the device instead of rcu
> > >>>> because rcu synchronization may sleep, which is not allowed when
> > >>>> unmounting a device.
> > >>>>
> > >>>> Signed-off-by: Deven Bowers <deven.desai@xxxxxxxxxxxxxxxxxxx>
> > >>>> Signed-off-by: Fan Wu <wufan@xxxxxxxxxxxxxxxxxxx>
> > >> ...
> > >>>> ---
> > >>>>    security/ipe/eval.c          | 72 +++++++++++++++++++++++++++++++++++-
> > >>>>    security/ipe/eval.h          |  2 +
> > >>>>    security/ipe/hooks.c         | 12 ++++++
> > >>>>    security/ipe/hooks.h         |  2 +
> > >>>>    security/ipe/ipe.c           |  1 +
> > >>>>    security/ipe/policy.h        |  2 +
> > >>>>    security/ipe/policy_parser.c | 35 +++++++++++++++++-
> > >>>>    7 files changed, 124 insertions(+), 2 deletions(-)
>
> ...
>
> > >>>> +/**
> > >>>> + * from_pinned - Determine whether @sb is the pinned super_block.
> > >>>> + * @sb: Supplies a super_block to check against the pinned super_block.
> > >>>> + *
> > >>>> + * Return:
> > >>>> + * * true   - @sb is the pinned super_block
> > >>>> + * * false  - @sb is not the pinned super_block
> > >>>> + */
> > >>>> +static bool from_pinned(const struct super_block *sb)
> > >>>> +{
> > >>>> +    bool rv;
> > >>>> +
> > >>>> +    if (!sb)
> > >>>> +            return false;
> > >>>> +    spin_lock(&pin_lock);
> > >>>> +    rv = !IS_ERR_OR_NULL(pinned_sb) && pinned_sb == sb;
> > >>>> +    spin_unlock(&pin_lock);
> > >>>
> > >>> It's okay for an initial version, but I still think you need to get
> > >>> away from this spinlock in from_pinned() as quickly as possible.
> > >>> Maybe I'm wrong, but this looks like a major source of lock contention.
> > >>>
> > >>> I understand the issue around RCU and the potential for matching on
> > >>> a reused buffer/address, but if you modified IPE to have its own LSM
> > >>> security blob in super_block::security you could mark the superblock
> > >>> when it was mounted and do a lockless lookup here in from_pinned().
> > >>
> > >> Thank you for the suggestion. After some testing, I discovered that
> > >> switching to RCU to pin the super block and using a security blob to
> > >> mark a pinned super block works. This approach do avoid many spinlock
> > >> operations. I'll incorporate these changes in the next version of the patch.
> > >
> > > I probably wasn't as clear as I should have been, I was thinking of
> > > doing away with the @pinned_sb global variable entirely, as well as
> > > its associated lock problems and simply marking the initramfs/initrd
> > > superblock when it was mounted.  I will admit that I haven't fully
> > > thought about all the implementation details, but I think you could
> > > leverage the security_sb_mount() hook to set a flag in IPE's
> > > superblock metadata when the initramfs was mounted.
> >
> > I wasn't able to find a way to let LSM pin initramfs/initrd during mount
> > time ...
>
> I haven't had to look at the kernel init code in a while, and I don't
> recall ever looking at the initramfs code, but I spent some time
> digging through the code and I wonder if it would be possible to mark
> the initramfs superblock in wait_for_initramfs() via a new LSM hook
> using @current->fs->root.mnt->mnt_sb?  Although I'm not completely
> sure that it's populated.  Have you already looked at an approach like
> this?

Thinking about this more, the current IPE approach of treating the
first file access as being present in the initramfs is not correct
(one could build a system without an initramfs).  I think we need to
do something like the above where the initramfs is explicitly marked
in the initramfs code.

> > But I think we could replace the global variable with a flag
> > variable ipe_sb_state so we could use atomic operation to only mark one
> > drive as pinned without any lock. The code will be like:
> >
> > static void pin_sb(const struct super_block *sb)
> > {
> >         if (!sb)
> >                 return;
> >
> >         if (!test_and_set_bit_lock(IPE_SB_PINNED, &ipe_sb_state)) {
> >                 ipe_sb(sb)->pinned = true;
> >         }
> > }
> >
> > Would this sound better?

-- 
paul-moore.com




[Index of Archives]     [Linux RAID]     [Linux SCSI]     [Linux ATA RAID]     [IDE]     [Linux Wireless]     [Linux Kernel]     [ATH6KL]     [Linux Bluetooth]     [Linux Netdev]     [Kernel Newbies]     [Security]     [Git]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Device Mapper]

  Powered by Linux