Re: [PATCHv5, REBASED 9/9] x86/mm: Allow to have userspace mappings above 47-bits

[Michal Hocko]

On Thu 18-05-17 17:50:03, Michal Hocko wrote:
> On Thu 18-05-17 18:41:35, Kirill A. Shutemov wrote:
> > On Thu, May 18, 2017 at 05:27:36PM +0200, Michal Hocko wrote:
> > > On Thu 18-05-17 18:19:52, Kirill A. Shutemov wrote:
> > > > On Thu, May 18, 2017 at 01:43:59PM +0200, Michal Hocko wrote:
> > > > > On Mon 15-05-17 15:12:18, Kirill A. Shutemov wrote:
> > > > > [...]
> > > > > > @@ -195,6 +207,16 @@ arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
> > > > > >  	info.length = len;
> > > > > >  	info.low_limit = PAGE_SIZE;
> > > > > >  	info.high_limit = get_mmap_base(0);
> > > > > > +
> > > > > > +	/*
> > > > > > +	 * If hint address is above DEFAULT_MAP_WINDOW, look for unmapped area
> > > > > > +	 * in the full address space.
> > > > > > +	 *
> > > > > > +	 * !in_compat_syscall() check to avoid high addresses for x32.
> > > > > > +	 */
> > > > > > +	if (addr > DEFAULT_MAP_WINDOW && !in_compat_syscall())
> > > > > > +		info.high_limit += TASK_SIZE_MAX - DEFAULT_MAP_WINDOW;
> > > > > > +
> > > > > >  	info.align_mask = 0;
> > > > > >  	info.align_offset = pgoff << PAGE_SHIFT;
> > > > > >  	if (filp) {
> > > > > 
> > > > > I have two questions/concerns here. The above assumes that any address above
> > > > > 1<<47 will use the _whole_ address space. Is this what we want?
> > > > 
> > > > Yes, I believe so.
> > > > 
> > > > > What if somebody does mmap(1<<52, ...) because he wants to (ab)use 53+
> > > > > bits for some other purpose? Shouldn't we cap the high_limit by the
> > > > > given address?
> > > > 
> > > > This would screw existing semantics of hint address -- "map here if
> > > > free, please".
> > > 
> > > Well, the given address is just _hint_. We are still allowed to map to a
> > > different place. And it is not specified whether the resulting mapping
> > > is above or below that address. So I do not think it would screw the
> > > existing semantic. Or do I miss something?
> > 
> > You are right, that this behaviour is not fixed by any standard or written
> > down in documentation, but it's de-facto policy of Linux mmap(2) the
> > beginning.
> > 
> > And we need to be very careful when messing with this.
> 
> I am sorry but I still do not understand. You already touch this
> semantic. mmap(-1UL,...) will already returns basically arbitrary
> address. All I am asking for is that mmap doesn't return higher address
> than the given one whent address > 1<<47. We do not have any such users
> currently so it won't be a change in behavior while it would allow
> different sized address spaces naturally.

I basically mean something like the following
---
diff --git a/arch/x86/kernel/sys_x86_64.c b/arch/x86/kernel/sys_x86_64.c
index 74d1587b181d..d6f66ff02d0a 100644
--- a/arch/x86/kernel/sys_x86_64.c
+++ b/arch/x86/kernel/sys_x86_64.c
@@ -195,7 +195,7 @@ arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
 		goto bottomup;
 
 	/* requesting a specific address */
-	if (addr) {
+	if (addr && addr <= DEFAULT_MAP_WINDOW) {
 		addr = PAGE_ALIGN(addr);
 		vma = find_vma(mm, addr);
 		if (TASK_SIZE - len >= addr &&
@@ -215,7 +215,7 @@ arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
 	 * !in_compat_syscall() check to avoid high addresses for x32.
 	 */
 	if (addr > DEFAULT_MAP_WINDOW && !in_compat_syscall())
-		info.high_limit += TASK_SIZE_MAX - DEFAULT_MAP_WINDOW;
+		info.high_limit += min(TASK_SIZE_MAX, address) - DEFAULT_MAP_WINDOW;
 
 	info.align_mask = 0;
 	info.align_offset = pgoff << PAGE_SHIFT;
-- 
Michal Hocko
SUSE Labs