On Thu, 2017-09-28 at 16:39 -0700, Linus Torvalds wrote: > On Thu, Sep 28, 2017 at 3:02 PM, Dave Chinner <david@xxxxxxxxxxxxx> wrote: > > On Thu, Sep 28, 2017 at 08:39:33AM -0400, Mimi Zohar wrote: > >> Don't attempt to take the i_rwsem, if it has already been taken > >> exclusively. > >> > >> Signed-off-by: Mimi Zohar <zohar@xxxxxxxxxxxxxxxxxx> > > > > That's bloody awful. > > > > The locking in filesystem IO paths is already complex enough without > > adding a new IO path semantic that says "caller has already locked > > the i_rwsem in some order and some dependencies that we have no idea > > about". > > I do have to admit that I never got a satisfactory answer on why IMA > doesn't just use its own private per-inode lock for this all. > > It isn't using the i_rwsem for file consistency reasons anyway, so it > seems to be purely about serializing the actual signature generation > with the xattr writing, but since IMA does those both, why isn't IMA > just using its own lock (not the filesystem lock) to do that? Originally IMA did define it's own lock, prior to IMA-appraisal. IMA- appraisal introduced writing the file hash as an xattr, which required taking the i_mutex. process_measurement() and ima_file_free() took the iint->mutex first and then the i_mutex, while setxattr, chmod and chown took the locks in reverse order. To resolve the potential deadlock, the iint->mutex was eliminated. Mimi
