On Thu, 2012-03-22 at 15:09 +0000, Al Viro wrote: > On Thu, Mar 22, 2012 at 10:53:04AM -0400, Mimi Zohar wrote: > > > BTW, you've missed several other places in mm/* doing fput(), so it wouldn't > > > be enough to paper over your problem anyway. > > > > > > Final fput() *can* happen under mmap_sem. Period. > > > > I think I got that loud and clear; otherwise we wouldn't have come up > > with deferring the __fput(). We have a very real problem here - writing > > extended attributes requires taking the i_mutex. > > Don't do it, then? If you _must_ write to xattr on final fput, I'd suggest > starting to figure out if xattr needs its protection to be ->i_mutex - it > might make sense to introduce a separate mutex for xattr crap. Or not - I'm > not familiar enough with the guts of xattr handling in individual filesystems > to tell if that would work (e.g. if it would need unpleasant changes to > ->setattr() instances)... After looking into this, the individual filesystems do their own xattr locking. The i_mutex, however, is currently required to access inode->i_op->setxattr() (and the isec). In addition, IMA-appraisal requires the i_mutex in order to calculate the file hash. Calling ima_file_free() after the mmap_sem is released, as opposed to queueing the __fput(), won't work, as the file needs to be open in order to calculate the file hash. Calling ima_file_free() before taking the mmap_sem, could work, but at that point we don't have access to the file handle. Do you see any other options? thanks, Mimi -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
