On Thu, 2012-03-22 at 14:22 +0000, Al Viro wrote: > On Wed, Mar 21, 2012 at 02:54:15PM -0400, Mimi Zohar wrote: > > If a file is closed before it is munmapped, __fput() is called with > > the mmap_sem taken. With IMA-appraisal enabled, this results in an > > mmap_sem/i_mutex lockdep. ima_defer_fput() increments the f_count > > to defer __fput() being called until after the mmap_sem is released. > > NAK. It's far too heavy-weight for what's a very common path. Worse, > it causes different locking conditions on IMA and non-IMA kernels and makes > bugs harder to spot. Rejected with extreme prejudice; please, fix your > locking instead. Al, thanks for replying! I realize you are very prejudiced against IMA, but have you taken a look at IMA-appraisal? It probably would have been better if we used another name in order to avoid this prejudice, but unfortunately we didn't. :( IMA-appraisal is about enforcing file integrity. > 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. I'm open to any suggestions. 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
