From: Andrew Jones <drjones@redhat.com>
Date: Fri, 12 Feb 2010 18:39:58 -0500
Subject: [mm] xen: make mmap() with PROT_WRITE
Message-id: <1265999998-13996-1-git-send-email-drjones@redhat.com>
Patchwork-id: 23264
O-Subject: [PATCH 5.5] [xen] [mm] Make mmap() with PROT_WRITE
Bugzilla: 562761
RH-Acked-by: Rik van Riel <riel@redhat.com>
RH-Acked-by: Miroslav Rezanina <mrezanin@redhat.com>

Bugzilla 562761

Using the PROT_WRITE flag with mmap() causes a SIGSEGV on first read.
This was patched for the bare-metal kernels for all arches since kernel
-95, but the *-xen files were missed. This patch makes the corresponding
changes to the xen files so mmap bug is fixed on xen hosts as well.

The commit for the bare-metal files is
c9ee5e25fb02218bf5786a1cea19a0d9f5eaaa6d

Tested by me on x86_64 with the following test program. Before the patch
the program segfaults. With the patch it completes fine.

main(void)
{
        char c, *p;
        p = mmap(0, 1, PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
        c = *p;   /* Reading a byte from the page */
}

Brew build is brewing.

diff --git a/arch/i386/mm/fault-xen.c b/arch/i386/mm/fault-xen.c
index 9561f24..1e3a18c 100644
--- a/arch/i386/mm/fault-xen.c
+++ b/arch/i386/mm/fault-xen.c
@@ -573,7 +573,7 @@ good_area:
 		case 1:		/* read, present */
 			goto bad_area;
 		case 0:		/* read, not present */
-			if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
+			if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
 				goto bad_area;
 	}
 
diff --git a/arch/x86_64/mm/fault-xen.c b/arch/x86_64/mm/fault-xen.c
index 5ff589a..8a9a09d 100644
--- a/arch/x86_64/mm/fault-xen.c
+++ b/arch/x86_64/mm/fault-xen.c
@@ -530,7 +530,7 @@ good_area:
 		case PF_PROT:		/* read, present */
 			goto bad_area;
 		case 0:			/* read, not present */
-			if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
+			if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
 				goto bad_area;
 	}