Re: [RFC PATCH 1/2] mm: introduce MAP_FIXED_SAFE

On Mon 20-11-17 10:10:32, Florian Weimer wrote:

On 11/20/2017 09:55 AM, Michal Hocko wrote:

quoted

On Fri 17-11-17 08:30:48, Florian Weimer wrote:

quoted

On 11/16/2017 11:18 AM, Michal Hocko wrote:

quoted

+	if (flags & MAP_FIXED_SAFE) {
+		struct vm_area_struct *vma = find_vma(mm, addr);
+
+		if (vma && vma->vm_start <= addr)
+			return -ENOMEM;
+	}

Could you pick a different error code which cannot also be caused by a an
unrelated, possibly temporary condition?  Maybe EBUSY or EEXIST?

Hmm, none of those are described in the man page. I am usually very
careful to not add new and potentially unexpected error codes but it is

I think this is a bad idea.  It leads to bizarre behavior, like open failing
with EOVERFLOW with certain namespace configurations (which have nothing to
do with file sizes).

Ohh, I agree but breaking userspace is, you know, no-no. And an
unexpected error codes can break things terribly.

Most of the manual pages are incomplete regarding error codes, and with
seccomp filters and security modules, what error codes you actually get is
anyone's guess.

quoted

true that a new flag should warrant a new error code. I am not sure
which one is more appropriate though. EBUSY suggests that retrying might
help which is true only if some other party unmaps the range. So EEXIST
would sound more natural.

Sure, EEXIST is completely fine.

OK, I will use it.
 

quoted

quoted

This would definitely help with application-based randomization of mappings,
and there, actual ENOMEM and this error would have to be handled
differently.

I see. Could you be more specific about the usecase you have in mind? I
would incorporate it into the patch description.

glibc ld.so currently maps DSOs without hints.  This means that the kernel
will map right next to each other, and the offsets between them a completely
predictable.  We would like to change that and supply a random address in a
window of the address space.  If there is a conflict, we do not want the
kernel to pick a non-random address. Instead, we would try again with a
random address.

This makes sense to me. Thanks, I will add it to the cover letter.
-- 
Michal Hocko
SUSE Labs