[PATCH 1/3] ima: Remove inode lock

[Roberto Sassu]

On Wed, 2024-10-09 at 18:25 +0200, Roberto Sassu wrote:
> On Wed, 2024-10-09 at 11:37 -0400, Paul Moore wrote:
> > On Wed, Oct 9, 2024 at 11:36 AM Paul Moore <paul at paul-moore.com> wrote:
> > > On Tue, Oct 8, 2024 at 12:57 PM Roberto Sassu
> > > <roberto.sassu at huaweicloud.com> wrote:
> > > > 
> > > > From: Roberto Sassu <roberto.sassu at huawei.com>
> > > > 
> > > > Move out the mutex in the ima_iint_cache structure to a new structure
> > > > called ima_iint_cache_lock, so that a lock can be taken regardless of
> > > > whether or not inode integrity metadata are stored in the inode.
> > > > 
> > > > Introduce ima_inode_security() to simplify accessing the new structure in
> > > > the inode security blob.
> > > > 
> > > > Move the mutex initialization and annotation in the new function
> > > > ima_inode_alloc_security() and introduce ima_iint_lock() and
> > > > ima_iint_unlock() to respectively lock and unlock the mutex.
> > > > 
> > > > Finally, expand the critical region in process_measurement() guarded by
> > > > iint->mutex up to where the inode was locked, use only one iint lock in
> > > > __ima_inode_hash(), since the mutex is now in the inode security blob, and
> > > > replace the inode_lock()/inode_unlock() calls in ima_check_last_writer().
> > > > 
> > > > Signed-off-by: Roberto Sassu <roberto.sassu at huawei.com>
> > > > ---
> > > >  security/integrity/ima/ima.h      | 26 ++++++++---
> > > >  security/integrity/ima/ima_api.c  |  4 +-
> > > >  security/integrity/ima/ima_iint.c | 77 ++++++++++++++++++++++++++-----
> > > >  security/integrity/ima/ima_main.c | 39 +++++++---------
> > > >  4 files changed, 104 insertions(+), 42 deletions(-)
> > > 
> > > I'm not an IMA expert, but it looks reasonable to me, although
> > > shouldn't this carry a stable CC in the patch metadata?
> > > 
> > > Reviewed-by: Paul Moore <paul at paul-moore.com>
> > 
> > Sorry, one more thing ... did you verify this patchset resolves the
> > syzbot problem?  I saw at least one reproducer.
> 
> Uhm, could not reproduce the deadlock with the reproducer. However,
> without the patch I have a lockdep warning, and with I don't.
> 
> I asked syzbot to try the patches. Let's see.

@bpf: could you please manually trigger the tests in a PR? Next time
will add the bpf-next tag (or I can send a PR directly from Github).

This patch affects the BPF LSM, the bpf_ima_file_hash() and
bpf_ima_inode_hash() helpers.

Thanks

Roberto