mbox series

[v5,0/6] evm: Prepare for moving to the LSM infrastructure

Message ID 20221123095202.599252-1-roberto.sassu@huaweicloud.com (mailing list archive)
Headers show
Series evm: Prepare for moving to the LSM infrastructure | expand

Message

Roberto Sassu Nov. 23, 2022, 9:51 a.m. UTC
From: Roberto Sassu <roberto.sassu@huawei.com>

One of the challenges that must be tackled to move IMA and EVM to the LSM
infrastructure is to ensure that EVM is capable to correctly handle
multiple stacked LSMs providing an xattr at file creation. At the moment,
there are few issues that would prevent a correct integration. This patch
set aims at solving them.

From the LSM infrastructure side, the LSM stacking feature added the
possibility of registering multiple implementations of the security hooks,
that are called sequentially whenever someone calls the corresponding
security hook. However, security_inode_init_security() is currently limited
to support one xattr provided by LSM and one by EVM. It is ready to support
multiple xattrs, as it can call the initxattrs() callback provided by
filesystems. security_old_inode_init_security() can only support one xattr
due to its API.

In addition, using the call_int_hook() macro causes some issues. According
to the documentation in include/linux/lsm_hooks.h, it is a legitimate case
that an LSM returns -EOPNOTSUPP when it does not want to provide an xattr.
However, the loop defined in the macro would stop calling subsequent LSMs
if that happens. In the case of security_old_inode_init_security(), using
the macro would also cause a memory leak due to replacing the *value
pointer, if multiple LSMs provide an xattr.

From EVM side, the first operation to be done is to change the definition
of evm_inode_init_security() to be compatible with the security hook
definition. Unfortunately, the current definition does not provide enough
information for EVM, as it must have visibility of all xattrs provided by
LSMs to correctly calculate the HMAC. This patch set changes the security
hook definition by replacing the name, value and len triple with the xattr
array allocated by security_inode_init_security().

Secondly, given that the place where EVM can fill an xattr is not provided
anymore with the changed definition, EVM must know how many elements are in
the xattr array. EVM can rely on the fact that the xattr array must be
terminated with an element with name field set to NULL. If EVM is moved to
the LSM infrastructure, the infrastructure will provide additional
information.

Casey suggested to use the reservation mechanism currently implemented for
other security blobs, for xattrs. In this way,
security_inode_init_security() can know after LSM initialization how many
slots for xattrs should be allocated, and LSMs know the offset in the
array from where they can start writing xattrs.

One of the problem was that LSMs can decide at run-time, although they
reserved a slot, to not use it (for example because they were not
initialized). Given that the initxattrs() method implemented by filesystems
expect that the array elements are contiguous, they would miss the slots
after the one not being initialized. security_check_compact_filled_xattrs()
has been introduced to overcome this problem and also to check the
correctness of the xattrs provided by the LSMs.

Filesystems calling security_inode_init_security() are already able to set
multiple xattrs. Those which were using security_old_inode_init_security(),
ocfs2 and reiserfs, were converted to using security_inode_init_security()
and security_old_inode_init_security() was removed. However, they are still
limited to setting only one xattr despite more are available, as their code
is not ready to set multiple xattrs at inode creation time. Support for it
can be added by those filesystems later.

This patch set has been tested by introducing several instances of a
TestLSM (some providing an xattr, some not, one with a wrong implementation
to see how the LSM infrastructure handles it, one providing multiple xattrs
and another providing an xattr but in a disabled state). The patch is not
included in this set but it is available here:

https://github.com/robertosassu/linux/commit/e13a03236df0c399dccb73df5fe4cfceb4bb1d89

The test, added to ima-evm-utils, is available here:

https://github.com/robertosassu/ima-evm-utils/blob/evm-multiple-lsms-v5-devel-v3/tests/evm_multiple_lsms.test

The test takes a UML kernel built by Github Actions and launches it several
times, each time with a different combination of LSMs and filesystems (ext4,
reiserfs, ocfs2). After boot, it first checks that there is an xattr for each
LSM providing it (for reiserfs and ocfs2 just the first LSM), and then (for
ext4) calculates the HMAC in user space and compares it with the HMAC
calculated by EVM in kernel space.

A test report can be obtained here:

https://github.com/robertosassu/ima-evm-utils/actions/runs/3525619568/jobs/5912560168

The patch set has been tested with both the SElinux and Smack test suites.
Below, there is the summary of the test results:

SELinux Test Suite result (without patches):
Files=73, Tests=1346, 225 wallclock secs ( 0.43 usr  0.23 sys +  6.11 cusr 58.70 csys = 65.47 CPU)
Result: FAIL
Failed 4/73 test programs. 13/1346 subtests failed.

SELinux Test Suite result (with patches):
Files=73, Tests=1346, 225 wallclock secs ( 0.44 usr  0.22 sys +  6.15 cusr 59.94 csys = 66.75 CPU)
Result: FAIL
Failed 4/73 test programs. 13/1346 subtests failed.

Smack Test Suite result (without patches):
95 Passed, 0 Failed, 100% Success rate

Smack Test Suite result (with patches):
95 Passed, 0 Failed, 100% Success rate

The patch set has also been successfully tested with a WIP branch where
IMA/EVM have been moved to the LSM infrastructure. It is available here:

https://github.com/robertosassu/linux/commits/ima-evm-lsms-v1-devel-v9-test

This is the patch that moves EVM to the LSM infrastructure:

https://github.com/robertosassu/linux/commit/a1e0682ef75cc8087327e6f446f586c02dd87c60

The only trivial changes, after this patch set, would be to allocate one
element less in the xattr array (because EVM will reserve its own xattr),
and to simply remove the call to evm_inode_init_security().

The test report when IMA and EVM are moved to the LSM infrastructure is
available here:

https://github.com/robertosassu/ima-evm-utils/actions/runs/3525750217/jobs/5912853636

Changelog

v4:
- Remove patch to call reiserfs_security_free(), already queued
- Switch ocfs2 and reiserfs to security_inode_init_security() (suggested by
  Mimi)
- Remove security_old_inode_init_security() (suggested by Paul)
- Rename security_check_compact_xattrs() to
  security_check_compact_filled_xattrs() and add function description
  (suggested by Mimi)
- Rename checked_xattrs parameter of security_check_compact_filled_xattrs()
  to num_filled_xattrs (suggested by Mimi)
- Rename cur_xattrs variable in security_inode_init_security() to
  num_filled_xattrs (suggested by Mimi)

v3:
- Don't free the xattr name in reiserfs_security_free()
- Don't include fs_data parameter in inode_init_security hook
- Don't change evm_inode_init_security(), as it will be removed if EVM is
  stacked
- Fix inode_init_security hook documentation
- Drop lsm_find_xattr_slot(), use simple xattr reservation mechanism and
  introduce security_check_compact_xattrs() to compact the xattr array
- Don't allocate xattr array if LSMs didn't reserve any xattr
- Return zero if initxattrs() is not provided to
  security_inode_init_security(), -EOPNOTSUPP if value is not provided to
  security_old_inode_init_security()
- Request LSMs to fill xattrs if only value (not the triple) is provided to
  security_old_inode_init_security(), to avoid unnecessary memory
  allocation

v2:
- rewrite selinux_old_inode_init_security() to use
  security_inode_init_security()
- add lbs_xattr field to lsm_blob_sizes structure, to give the ability to
  LSMs to reserve slots in the xattr array (suggested by Casey)
- add new parameter base_slot to inode_init_security hook definition

v1:
- add calls to reiserfs_security_free() and initialize sec->value to NULL
  (suggested by Tetsuo and Mimi)
- change definition of inode_init_security hook, replace the name, value
  and len triple with the xattr array (suggested by Casey)
- introduce lsm_find_xattr_slot() helper for LSMs to find an unused slot in
  the passed xattr array

Roberto Sassu (6):
  reiserfs: Switch to security_inode_init_security()
  ocfs2: Switch to security_inode_init_security()
  security: Remove security_old_inode_init_security()
  security: Allow all LSMs to provide xattrs for inode_init_security
    hook
  evm: Align evm_inode_init_security() definition with LSM
    infrastructure
  evm: Support multiple LSMs providing an xattr

 fs/ocfs2/namei.c                    |  18 ++---
 fs/ocfs2/xattr.c                    |  30 +++++++-
 fs/reiserfs/xattr_security.c        |  23 ++++--
 include/linux/evm.h                 |  12 +--
 include/linux/lsm_hook_defs.h       |   3 +-
 include/linux/lsm_hooks.h           |  17 ++--
 include/linux/security.h            |  12 ---
 security/integrity/evm/evm.h        |   2 +
 security/integrity/evm/evm_crypto.c |   9 ++-
 security/integrity/evm/evm_main.c   |  28 +++++--
 security/security.c                 | 115 +++++++++++++++++++++-------
 security/selinux/hooks.c            |  19 +++--
 security/smack/smack_lsm.c          |  26 ++++---
 13 files changed, 213 insertions(+), 101 deletions(-)

Comments

Mimi Zohar Nov. 23, 2022, 12:28 p.m. UTC | #1
Hi Roberto,

On Wed, 2022-11-23 at 10:51 +0100, Roberto Sassu wrote:
> From: Roberto Sassu <roberto.sassu@huawei.com>
> 
> One of the challenges that must be tackled to move IMA and EVM to the LSM
> infrastructure is to ensure that EVM is capable to correctly handle
> multiple stacked LSMs providing an xattr at file creation. At the moment,
> there are few issues that would prevent a correct integration. This patch
> set aims at solving them.

Let's take a step back and understand the purpose of this patch set. 
Regardless of whether IMA and EVM are moved to the "LSM
infrastructure", EVM needs to support per LSM xattrs.  A side affect is
the removal of the security_old_inode_init_security hook.  This patch
set cover letter and patch descriptions should be limited to EVM
support for per LSM (multiple) xattrs.  The motivation, concerns, and
problems of making IMA and EVM LSMs will be documented in the patch set
that actual makes them LSMs.  Please remove all references to "move IMA
and EVM to the LSM infrastructure".

When EVM was upstreamed, there were filesystem limitations on the
number and size of the extended attributes.  In addition there were
performance concerns, which resulted in staging the LSM, IMA and EVM
xattrs, before calling initxattrs to write them at the same time.  With
this patch set, not only are per LSM xattrs supported, but multiple per
LSM xattrs are supported as well.  Have the size limitation concerns
been addressed by the different filesystems?   If not, then at minimum
this patch set needs to at least mention it and the possible
ramifications.
Roberto Sassu Nov. 23, 2022, 12:44 p.m. UTC | #2
On Wed, 2022-11-23 at 07:28 -0500, Mimi Zohar wrote:
> Hi Roberto,
> 
> On Wed, 2022-11-23 at 10:51 +0100, Roberto Sassu wrote:
> > From: Roberto Sassu <roberto.sassu@huawei.com>
> > 
> > One of the challenges that must be tackled to move IMA and EVM to the LSM
> > infrastructure is to ensure that EVM is capable to correctly handle
> > multiple stacked LSMs providing an xattr at file creation. At the moment,
> > there are few issues that would prevent a correct integration. This patch
> > set aims at solving them.
> 
> Let's take a step back and understand the purpose of this patch set. 
> Regardless of whether IMA and EVM are moved to the "LSM
> infrastructure", EVM needs to support per LSM xattrs.  A side affect is
> the removal of the security_old_inode_init_security hook.  This patch
> set cover letter and patch descriptions should be limited to EVM
> support for per LSM (multiple) xattrs.  The motivation, concerns, and
> problems of making IMA and EVM LSMs will be documented in the patch set
> that actual makes them LSMs.  Please remove all references to "move IMA
> and EVM to the LSM infrastructure".

Hi Mimi

ok, will do.

> When EVM was upstreamed, there were filesystem limitations on the
> number and size of the extended attributes.  In addition there were
> performance concerns, which resulted in staging the LSM, IMA and EVM
> xattrs, before calling initxattrs to write them at the same time.  With
> this patch set, not only are per LSM xattrs supported, but multiple per
> LSM xattrs are supported as well.  Have the size limitation concerns
> been addressed by the different filesystems?   If not, then at minimum
> this patch set needs to at least mention it and the possible
> ramifications.

With your patch, 9d8f13ba3f483 ("security: new
security_inode_init_security API adds function callback") you made it
possible to set multiple xattrs at inode creation time.

This patch set pushes further to the limits, as there could be more
xattrs to be added to the inode. I will mention that.

If there are too many xattrs, I guess the only solution would be to use
less LSMs, or a different filesystem. The per filesystem limit could be
increased separately case by case.

Thanks

Roberto
Mimi Zohar Nov. 23, 2022, 1:06 p.m. UTC | #3
On Wed, 2022-11-23 at 13:44 +0100, Roberto Sassu wrote:
> On Wed, 2022-11-23 at 07:28 -0500, Mimi Zohar wrote:
> > Hi Roberto,
> > 
> > On Wed, 2022-11-23 at 10:51 +0100, Roberto Sassu wrote:
> > > From: Roberto Sassu <roberto.sassu@huawei.com>
> > > 
> > > One of the challenges that must be tackled to move IMA and EVM to the LSM
> > > infrastructure is to ensure that EVM is capable to correctly handle
> > > multiple stacked LSMs providing an xattr at file creation. At the moment,
> > > there are few issues that would prevent a correct integration. This patch
> > > set aims at solving them.
> > 
> > Let's take a step back and understand the purpose of this patch set. 
> > Regardless of whether IMA and EVM are moved to the "LSM
> > infrastructure", EVM needs to support per LSM xattrs.  A side affect is
> > the removal of the security_old_inode_init_security hook.  This patch
> > set cover letter and patch descriptions should be limited to EVM
> > support for per LSM (multiple) xattrs.  The motivation, concerns, and
> > problems of making IMA and EVM LSMs will be documented in the patch set
> > that actual makes them LSMs.  Please remove all references to "move IMA
> > and EVM to the LSM infrastructure".
> 
> Hi Mimi
> 
> ok, will do.
> 
> > When EVM was upstreamed, there were filesystem limitations on the
> > number and size of the extended attributes.  In addition there were
> > performance concerns, which resulted in staging the LSM, IMA and EVM
> > xattrs, before calling initxattrs to write them at the same time.  With
> > this patch set, not only are per LSM xattrs supported, but multiple per
> > LSM xattrs are supported as well.  Have the size limitation concerns
> > been addressed by the different filesystems?   If not, then at minimum
> > this patch set needs to at least mention it and the possible
> > ramifications.
> 
> With your patch, 9d8f13ba3f483 ("security: new
> security_inode_init_security API adds function callback") you made it
> possible to set multiple xattrs at inode creation time.

True, and even then there were concerns.

> This patch set pushes further to the limits, as there could be more
> xattrs to be added to the inode. I will mention that.

Thanks

> If there are too many xattrs, I guess the only solution would be to use
> less LSMs, or a different filesystem. The per filesystem limit could be
> increased separately case by case.

Agreed, but unless it is documented somewhere, nobody but us will know
there is a potential problem.  At least document it here in the cover
letter, which we'll include in the merge message.

FYI, the xattr.7 man page contains a section "Filesystem differences".