Re: [PATCH] vfs: fix IMA lockdep circular locking dependency

[Al Viro <viro@ZenIV.linux.org.uk>]

On Wed, May 30, 2012 at 04:33:36PM -0400, Mimi Zohar wrote:
> On Wed, 2012-05-30 at 21:24 +0100, Al Viro wrote:
> > On Wed, May 30, 2012 at 03:42:47PM -0400, Eric Paris wrote:
> > > security_file_mmap() would pretty much only be used in do_mmap_pgoff()
> > > (or validate_mmap_request)
> > 
> > Callers, actually - the whole point is to lift it out of under ->mmap_sem.
> > The tricky part is reqprot vs. prot mess.
> 
> Linus already addressed it in the original patch, which I split into two
> as you suggested:
> 733559f - "security: move security_file_mmap() and rename to security_mmap_file()" 
> b3649e9 security: define and use the new security_mmap_addr() hook

Ehh...

* elf_map() should just be using vm_map/vm_unmap (see vfs.git#for-next)
* that goto out; in sys_mmap_pgooff() is an instant leak - you miss
fput() that way
* again, prot handling for !MMU case differs from MMU one.  Sure, we
can duplicate both, but it's getting seriously ugly ;-/
* while we are at it, checking address in validate_mmap_request() is
really odd, seeing that we have the only call to validate_mmap_request()
followed by
        /* we ignore the address hint */
        addr = 0;
        len = PAGE_ALIGN(len);