On Thu, 2017-05-11 at 10:16 +0200, Christoph Hellwig wrote: > On Wed, May 10, 2017 at 05:00:47PM -0400, Mimi Zohar wrote: > > Without i_version support the file is measured/appraised once. With > > i_version support it will be re-measured/appraised. As a file system > > is mounted/remounted, some sort of message should be emitted > > indicating whether i_version is supported. > > You can check for (sb->s_flags & MS_I_VERSION) to see if it's supported. Yes, I defined a new LSM hook to catch the new mounts, but there are lots of mounts, even after to limiting it to non-kernel mounts (MS_KERNMOUNT) and only checking if the MS_I_VERSION is set on filesystems mounted read-write. It would be nice if there was a way of saying not pseudo filesystems (eg. CGROUP_SUPER_MAGIC, DEBUGFS_MAGIC, DEVPTS_SUPER_MAGIC, PROC_SUPER_MAGIC, SECURITYFS_MAGIC, SYSFS_MAGIC, etc). > > > That does not imply that > > there is no value in measuring/appraising the file only once. > > > > With this patch, the "opt-in" behavior, is only for measurement, not > > appraisal. For appraisal, it still enforces file hash/signature > > verification, as it should, based on policy. > > > > Christoph, could we call ->read_iter() in the NULL case as Boaz > > suggested? > > No - that way you get deadlocks for every fs that uses i_rwsem in > ->read_iter, which is perfectly valid behavior. > > We can set ->integrity_read for every file system that's been tested > with IMA, though. Do you have a list of known-good file systems? In addition to the ones you've already defined, we need definitions in ramfs/file-mmu.c and file-nommu.c, and the corresponding tmpfs, to get the initial measurements from the initramfs. We know that stacked filesystems have similar locking problems. I'm loop back mounting each filesystem and testing to see if files are being measured/re-measured properly. I haven't finished yet, but there haven't been any problems so far. Mimi
