From: Deven Bowers Subject: Re: [RFC PATCH v5 06/11] dm-verity: move signature check after tree validation Date: Tue, 28 Jul 2020 16:55:04 -0700 Message-ID: References: <20200728213614.586312-1-deven.desai@linux.microsoft.com> <20200728213614.586312-7-deven.desai@linux.microsoft.com> <20200728215041.GF4053562@gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20200728215041.GF4053562@gmail.com> Content-Language: en-US Sender: linux-block-owner@vger.kernel.org To: Eric Biggers Cc: agk@redhat.com, axboe@kernel.dk, snitzer@redhat.com, jmorris@namei.org, serge@hallyn.com, zohar@linux.ibm.com, viro@zeniv.linux.org.uk, paul@paul-moore.com, eparis@redhat.com, jannh@google.com, dm-devel@redhat.com, linux-integrity@vger.kernel.org, linux-security-module@vger.kernel.org, linux-fsdevel@vger.kernel.org, linux-block@vger.kernel.org, linux-audit@redhat.com, tyhicks@linux.microsoft.com, linux-kernel@vger.kernel.org, corbet@lwn.net, sashal@kernel.org, jaskarankhurana@linux.microsoft.com, mdsakib@microsoft.com, nramas@linux.microsoft.com, pasha.tatashin@soleen.com List-Id: dm-devel.ids On 7/28/2020 2:50 PM, Eric Biggers wrote: > On Tue, Jul 28, 2020 at 02:36:06PM -0700, Deven Bowers wrote: >> The CONFIG_DM_VERITY_VERIFY_ROOTHASH_SIG introduced by Jaskaran was >> intended to be used to allow an LSM to enforce verifications for all >> dm-verity volumes. >> >> However, with it's current implementation, this signature verification >> occurs after the merkel-tree is validated, as a result the signature can >> pass initial verification by passing a matching root-hash and signature. >> This results in an unreadable block_device, but that has passed signature >> validation (and subsequently, would be marked as verified). >> >> This change moves the signature verification to after the merkel-tree has >> finished validation. >> >> Signed-off-by: Deven Bowers >> --- >> drivers/md/dm-verity-target.c | 44 ++++------ >> drivers/md/dm-verity-verify-sig.c | 140 ++++++++++++++++++++++-------- >> drivers/md/dm-verity-verify-sig.h | 24 +++-- >> drivers/md/dm-verity.h | 2 +- >> 4 files changed, 134 insertions(+), 76 deletions(-) >> >> diff --git a/drivers/md/dm-verity-target.c b/drivers/md/dm-verity-target.c >> index eec9f252e935..fabc173aa7b3 100644 >> --- a/drivers/md/dm-verity-target.c >> +++ b/drivers/md/dm-verity-target.c >> @@ -471,9 +471,9 @@ static int verity_verify_io(struct dm_verity_io *io) >> struct bvec_iter start; >> unsigned b; >> struct crypto_wait wait; >> + int r; >> >> for (b = 0; b < io->n_blocks; b++) { >> - int r; >> sector_t cur_block = io->block + b; >> struct ahash_request *req = verity_io_hash_req(v, io); >> >> @@ -530,6 +530,16 @@ static int verity_verify_io(struct dm_verity_io *io) >> return -EIO; >> } >> >> + /* >> + * At this point, the merkel tree has finished validating. >> + * if signature was specified, validate the signature here. >> + */ >> + r = verity_verify_root_hash(v); >> + if (r < 0) { >> + DMERR_LIMIT("signature mismatch"); >> + return r; >> + } >> + >> return 0; >> } > > This doesn't make any sense. > > This just moves the signature verification to some random I/O. > > The whole point of dm-verity is that data is verified on demand. You can't know > whether any particular data or hash block is consistent with the root hash or > not until it is read and verified. > > When the first I/O completes it might have just checked one block of a billion. > > Not to mention that you didn't consider locking at all. > > - Eric > I appear to have dangerously misunderstood how dm-verity works under the covers. What I thought was happening here was that *this* would be where the first I/O that completes validation and has been verified - the root hash signature could then be checked against the root hash, and then no-op for remaining blocks, provided the signature validates. The reason why I was proposing moving the signature check, is that I was afraid of the block_device being created in dm-verity with a root-hash that belongs to a different device + a signature that verifies that root-hash, would get past verity_ctr, as despite the root hash not matching the hash tree, the signature and the root hash will be verified. At this point, a block_device structure would be resident in the kernel with the security attributes I propose in the next patch in the series. This device would never be read successfully, but the structure with the attribute would exist. This felt odd because there would be a structure in the kernel with an attribute that says it passed a security check, but the block_device is effectively invalid. I realize now that that's a pretty ridiculous situation because the theoretical attack with access to manipulate the kernel memory in such a way to make it viable could just override whatever is needed to make the exploit work, and isn't unique to dm-verity. I'm going to drop this patch in the next iteration of this series.