Re: [PATCH RFC v15 14/21] dm verity: consume root hash digest and signature data via LSM hook

From: Mike Snitzer
Date: Wed Mar 20 2024 - 14:50:32 EST


On Wed, Mar 20 2024 at 1:23P -0400,
Paul Moore <paul@xxxxxxxxxxxxxx> wrote:

> On Tue, Mar 19, 2024 at 10:19 PM Mike Snitzer <snitzer@xxxxxxxxxx> wrote:
> > On Tue, Mar 19 2024 at 7:00P -0400,
> > Paul Moore <paul@xxxxxxxxxxxxxx> wrote:
> > > On Mar 15, 2024 Fan Wu <wufan@xxxxxxxxxxxxxxxxxxx> wrote:
> > > >
> > > > dm-verity provides a strong guarantee of a block device's integrity. As
> > > > a generic way to check the integrity of a block device, it provides
> > > > those integrity guarantees to its higher layers, including the filesystem
> > > > level.
> > > >
> > > > An LSM that control access to a resource on the system based on the
> > > > available integrity claims can use this transitive property of
> > > > dm-verity, by querying the underlying block_device of a particular
> > > > file.
> > > >
> > > > The digest and signature information need to be stored in the block
> > > > device to fulfill the next requirement of authorization via LSM policy.
> > > > This will enable the LSM to perform revocation of devices that are still
> > > > mounted, prohibiting execution of files that are no longer authorized
> > > > by the LSM in question.
> > > >
> > > > This patch adds two security hook calls in dm-verity to save the
> > > > dm-verity roothash and the roothash signature to the block device's
> > > > LSM blobs. The hook calls are depended on CONFIG_IPE_PROP_DM_VERITY,
> > > > which will be introduced in the next commit.
> > > >
> > > > Signed-off-by: Deven Bowers <deven.desai@xxxxxxxxxxxxxxxxxxx>
> > > > Signed-off-by: Fan Wu <wufan@xxxxxxxxxxxxxxxxxxx>
> > > > ---
> > > > v2:
> > > > + No Changes
> > > >
> > > > v3:
> > > > + No changes
> > > >
> > > > v4:
> > > > + No changes
> > > >
> > > > v5:
> > > > + No changes
> > > >
> > > > v6:
> > > > + Fix an improper cleanup that can result in
> > > > a leak
> > > >
> > > > v7:
> > > > + Squash patch 08/12, 10/12 to [11/16]
> > > > + Use part0 for block_device, to retrieve the block_device, when
> > > > calling security_bdev_setsecurity
> > > >
> > > > v8:
> > > > + Undo squash of 08/12, 10/12 - separating drivers/md/ from
> > > > security/ & block/
> > > > + Use common-audit function for dmverity_signature.
> > > > + Change implementation for storing the dm-verity digest to use the
> > > > newly introduced dm_verity_digest structure introduced in patch
> > > > 14/20.
> > > > + Create new structure, dm_verity_digest, containing digest algorithm,
> > > > size, and digest itself to pass to the LSM layer. V7 was missing the
> > > > algorithm.
> > > > + Create an associated public header containing this new structure and
> > > > the key values for the LSM hook, specific to dm-verity.
> > > > + Additional information added to commit, discussing the layering of
> > > > the changes and how the information passed will be used.
> > > >
> > > > v9:
> > > > + No changes
> > > >
> > > > v10:
> > > > + No changes
> > > >
> > > > v11:
> > > > + Add an optional field to save signature
> > > > + Move the security hook call to the new finalize hook
> > > >
> > > > v12:
> > > > + No changes
> > > >
> > > > v13:
> > > > + No changes
> > > >
> > > > v14:
> > > > + Correct code format
> > > > + Remove unnecessary header and switch to dm_disk()
> > > >
> > > > v15:
> > > > + Refactor security_bdev_setsecurity() to security_bdev_setintegrity()
> > > > + Remove unnecessary headers
> > > > ---
> > > > drivers/md/dm-verity-target.c | 73 +++++++++++++++++++++++++++++++++++
> > > > drivers/md/dm-verity.h | 6 +++
> > > > include/linux/dm-verity.h | 12 ++++++
> > > > include/linux/security.h | 2 +
> > > > 4 files changed, 93 insertions(+)
> > > > create mode 100644 include/linux/dm-verity.h
> > > >
> > > > diff --git a/drivers/md/dm-verity-target.c b/drivers/md/dm-verity-target.c
> > > > index bb5da66da4c1..e94cc6a755d5 100644
> > > > --- a/drivers/md/dm-verity-target.c
> > > > +++ b/drivers/md/dm-verity-target.c
> > > > @@ -22,6 +22,8 @@
> > > > #include <linux/scatterlist.h>
> > > > #include <linux/string.h>
> > > > #include <linux/jump_label.h>
> > > > +#include <linux/security.h>
> > > > +#include <linux/dm-verity.h>
> > > >
> > > > #define DM_MSG_PREFIX "verity"
> > > >
> > > > @@ -1017,6 +1019,38 @@ static void verity_io_hints(struct dm_target *ti, struct queue_limits *limits)
> > > > blk_limits_io_min(limits, limits->logical_block_size);
> > > > }
> > > >
> > > > +#ifdef CONFIG_IPE_PROP_DM_VERITY
> > > > +
> > > > +static int verity_init_sig(struct dm_verity *v, const void *sig,
> > > > + size_t sig_size)
> > > > +{
> > > > + v->sig_size = sig_size;
> > > > + v->root_digest_sig = kmalloc(v->sig_size, GFP_KERNEL);
> > > > + if (!v->root_digest)
> > > > + return -ENOMEM;
> > >
> > > Either you meant to copy @sig into @v->root_digest_sig and forgot to
> > > add the code for that, or we don't need to include @sig as a parameter
> > > to this function. I'm guessing it is the former as it wouldn't make
> > > sense to even have dm_verity::root_digest_sig if we weren't stashing
> > > it here.
> > >
> > > I'd also suggest looking at kmemdup() instead of a kmalloc()/memcpy()
> > > combo.
> > >
> > > > + return 0;
> > > > +}
> > > > +
> > > > +static void verity_free_sig(struct dm_verity *v)
> > > > +{
> > > > + kfree(v->root_digest_sig);
> > > > +}
> > > > +#else
> > > > +
> > > > +static inline int verity_init_sig(struct dm_verity *v, const void *sig,
> > > > + size_t sig_size)
> > > > +{
> > > > + return 0;
> > > > +}
> > > > +
> > > > +static inline void verity_free_sig(struct dm_verity *v)
> > > > +{
> > > > +}
> > > > +
> > > > +#endif /* CONFIG_IPE_PROP_DM_VERITY */
> > >
> > > It's been a while since I looked at this patch in the patchset, so
> > > maybe I'm missing something, but in general we don't want CONFIG_XXX
> > > checks in the kernel, outside of security/, that are specific to a
> > > particular LSM (what happens when multiple LSMs need this?). Please
> > > use CONFIG_SECURITY instead.
> > >
> > > > static void verity_dtr(struct dm_target *ti)
> > > > {
> > > > struct dm_verity *v = ti->private;
> > > > @@ -1035,6 +1069,7 @@ static void verity_dtr(struct dm_target *ti)
> > > > kfree(v->salt);
> > > > kfree(v->root_digest);
> > > > kfree(v->zero_digest);
> > > > + verity_free_sig(v);
> > > >
> > > > if (v->tfm)
> > > > crypto_free_ahash(v->tfm);
> > > > @@ -1434,6 +1469,13 @@ static int verity_ctr(struct dm_target *ti, unsigned int argc, char **argv)
> > > > ti->error = "Root hash verification failed";
> > > > goto bad;
> > > > }
> > > > +
> > > > + r = verity_init_sig(v, verify_args.sig, verify_args.sig_size);
> > > > + if (r < 0) {
> > > > + ti->error = "Cannot allocate root digest signature";
> > > > + goto bad;
> > > > + }
> > > > +
> > > > v->hash_per_block_bits =
> > > > __fls((1 << v->hash_dev_block_bits) / v->digest_size);
> > > >
> > > > @@ -1584,6 +1626,34 @@ int dm_verity_get_root_digest(struct dm_target *ti, u8 **root_digest, unsigned i
> > > > return 0;
> > > > }
> > > >
> > > > +#ifdef CONFIG_IPE_PROP_DM_VERITY
> > > > +
> > > > +static int verity_finalize(struct dm_target *ti)
> > > > +{
> > > > + struct block_device *bdev;
> > > > + struct dm_verity_digest root_digest;
> > > > + struct dm_verity *v;
> > > > + int r;
> > > > +
> > > > + v = ti->private;
> > > > + bdev = dm_disk(dm_table_get_md(ti->table))->part0;
> > > > + root_digest.digest = v->root_digest;
> > > > + root_digest.digest_len = v->digest_size;
> > > > + root_digest.alg = v->alg_name;
> > > > +
> > > > + r = security_bdev_setintegrity(bdev, LSM_INTGR_DMV_ROOTHASH, &root_digest,
> > > > + sizeof(root_digest));
> > > > + if (r)
> > > > + return r;
> > > > +
> > > > + return security_bdev_setintegrity(bdev,
> > > > + LSM_INTGR_DMV_SIG,
> > > > + v->root_digest_sig,
> > > > + v->sig_size);
> > >
> > > What happens if the second call fails, should we clear the
> > > LSM_INTGR_DMV_ROOTHASH state in the LSM?
> > >
> > > > +}
> > > > +
> > > > +#endif /* CONFIG_IPE_PROP_DM_VERITY */
> > >
> > > See my comments about CONFIG_SECURITY above. In fact, I would suggest
> > > moving this up into that part of the file so you only need one #ifdef
> > > block relating to CONFIG_SECURITY.
> > >
> > > I would also recommend making a dummy function so we can get rid of
> > > the conditional compilation in @verity_target below. For example:
> > >
> > > #ifdef CONFIG_SECURITY
> > > static int verity_finalize(struct dm_target *ti)
> > > {
> > > /* real implementation */
> > > }
> > > #else
> > > static int verity_finalize(struct dm_target *ti)
> > > {
> > > return 0;
> > > }
> > > #endif /* CONFIG_SECURITY */
> > >
> > > > static struct target_type verity_target = {
> > > > .name = "verity",
> > > > .features = DM_TARGET_SINGLETON | DM_TARGET_IMMUTABLE,
> > > > @@ -1596,6 +1666,9 @@ static struct target_type verity_target = {
> > > > .prepare_ioctl = verity_prepare_ioctl,
> > > > .iterate_devices = verity_iterate_devices,
> > > > .io_hints = verity_io_hints,
> > > > +#ifdef CONFIG_IPE_PROP_DM_VERITY
> > > > + .finalize = verity_finalize,
> > > > +#endif /* CONFIG_IPE_PROP_DM_VERITY */
> > > > };
> > > > module_dm(verity);
> > >
> > > If you create a dummy verity_finalize() function like above you can
> > > get rid of the #ifdef checks.
> >
> > Think it is better to leave it as-is, to avoid calling the .finalize
> > hook if it isn't actually needed.
>
> Fair enough, my personal preference is to minimize Kconfig conditional
> code flow changes such as this, but I understand your point of view
> and device-mapper is your code after all.
>
> I believe the other issues still need to be addressed,

Yes.

> but other than that are you generally okay with the new "finalize"
> hook approach?

I am, not seeing how we could avoid it.

Thanks,
Mike