Re: [PATCH] Fix mlockall for PROT_NONE mappings

[Andi Kleen <ak@muc.de>]

On Thu, Oct 09, 2003 at 08:17:48AM -0700, Linus Torvalds wrote:
> 
> On 9 Oct 2003, Andi Kleen wrote:
> > 
> > Ok. But what is with mappings that have MAY_READ not set ?
> > [not 100% this cannot happen]
> 
> Well, PROT_NONE will not have MAY_READ set, and no, they won't get locked 
> into memory.
> 
> But neither did they get locked in before either, so it's not like it's a 
> new issue. Now we'd just ignore them nicely rather than abort the 
> lockall(). Hmm?

Ok, here's a new version.

-------------------------------------------------------------------------

Fix mlockall when there are unreadable mappings in the address space
(happens frequently on x86-64 where ld.so creates an unreadable mapping
for libc). Ignores mappings that don't have MAY_READ set in the high level 
mlockall function.

As an improvement it continues mlockall now for all mappings even when an 
error occurs. The error is still reported, but only the first one.

This should do the right thing for applications that do mlockall(MCL_CURRENT)
but don't want to worry about some weird mappings e.g. from DRI in their
address space.


diff -u linux-test7-work/mm/mlock.c-o linux-test7-work/mm/mlock.c
--- linux-test7-work/mm/mlock.c-o	2003-09-28 10:53:25.000000000 +0200
+++ linux-test7-work/mm/mlock.c	2003-12-04 16:33:58.000000000 +0100
@@ -135,7 +135,7 @@
 
 static int do_mlockall(int flags)
 {
-	int error;
+	int error, nerr;
 	unsigned int def_flags;
 	struct vm_area_struct * vma;
 
@@ -154,9 +154,15 @@
 		newflags = vma->vm_flags | VM_LOCKED;
 		if (!(flags & MCL_CURRENT))
 			newflags &= ~VM_LOCKED;
-		error = mlock_fixup(vma, vma->vm_start, vma->vm_end, newflags);
-		if (error)
-			break;
+
+		if (!(newflags & VM_READ))
+			continue;
+
+	        nerr = mlock_fixup(vma, vma->vm_start, vma->vm_end, newflags);
+		/* When an error occurs continue with the next mapping, 
+		   but report only the first one */
+		if (nerr && !error)
+			error = nerr;
 	}
 	return error;
 }